Difference between cloud and VPS?

Portal Home > Knowledgebase > Articles Database > Difference between cloud and VPS?

Posted by olgaki, 10-26-2010, 06:08 PM
Hello guys I wanna ask you to explain me the diferences between cloud and VPS I mean its seems the same you have a server and split the resources around the clients but whats the diference?
Posted by DiegoRBaquero, 10-26-2010, 06:10 PM
Cloud basically spreads the load of the virtual instance into many servers, you will never run out of your limited resources(unless you actually use them all). Cloud allows rapid fail protection, resources spread and balanced in servers, so as the internet links. You won't experience downtime if one server goes down, while if a dedicated server goes down, and your VPS is hosted there, expect downtime. Cloud is more expensive, but more reliable.
Posted by KarlZimmer, 10-26-2010, 07:19 PM
For 90%+ of cloud implementations there is no difference... That is much of the reason I hate the term "cloud" in the first place.
Posted by fancypocket, 10-28-2010, 07:50 AM
Are cloud providers able to oversell their resources?
Posted by WebGuyz, 10-28-2010, 09:17 AM
The only way anyone makes money in this business is to oversell.
Posted by FastServ, 10-28-2010, 09:30 AM
Exactly... By the 'modern' definition (redundant storage, redundant hypervisors, automatic failover), we've been offering 'cloud servers' for the past 4 years and just calling it VPS... I guess it's time to get on the cloud train. Toot Toot!
Posted by FHDave, 10-28-2010, 09:35 AM
Just curious,wWhat redundant storage, redundant hypervisors, and the kind of automatic failover did you implement since four years ago?
Posted by FastServ, 10-28-2010, 10:40 AM
Well I mistyped, it was actually 2.5 years ago when we deployed our first Citrix XenServer cluster which includes live migration (what everyone refers to as 'failover' these days). 4 years ago it was plain open source Xen which had iffy live migration so we never used it until the XenServer days.
Posted by FHDave, 10-28-2010, 10:48 AM
2.5 years ago may make more sense. What redundant storage did you use four years ago? I don't know much about Xen's live migration, but isn't it similar to Virtuozzo's live migration? You don't need iSCSI backend storage for this, do you? And the live migration is migration over time. Four years ago, what cloud virtualization do you use that enables your VPS to be migrated instantly when the hardware node fails?
Posted by JasonD10, 10-28-2010, 10:58 AM
If there is no difference, then that is a misrepresentation by the provider as there is a very big difference between just simply "VPS" and "Cloud". A VPS is a Virtual Private Server. It's one container, no redundancy, no high availability, limited vertical scaling, harder to manage, rarely entirely manageable from a web browser (both the client side, and Cloud administrator), and it's usually a more fixed cost (utility billing in many Cloud providers). More and more "Cloud" providers are being called out, and as true Cloud is becoming more adopted these providers claiming to be Cloud that are not, will continue to be called out and the definition will continue to become more clear. This is a shift and it's happened before in the industry so it will take time.
Posted by JasonD10, 10-28-2010, 11:01 AM
Some providers can oversell (technically), but quite a few technologies such as the Xen hypervisor which allocate resources 1:1 so it's not happening as often as it is with many other hosting platforms.
Posted by JasonD10, 10-28-2010, 11:04 AM
With XenMotion, you need shared storage (SAN) and it could be done via iSCSI. XenServer is a nice platform (we use it on older pre-cloud virtualization managed hosting customers) but I'd hardly classify it as "Cloud", even with their cream of the crop license, redundant hardware everywhere (including multiple SANs), and feature set. At the end of the day XenServer is still just VPS's with some Cloud features. It's lacking the complete list that I just listed above.
Posted by FastServ, 10-28-2010, 11:11 AM
You need NFS, iSCSI, or FC for live migration; it's not possible with local storage. Originally we used NFS, and today a mix of technologies. On the original plain Xen setups, they used local storage, thus no live migration. That's why I revised my statement after your comment. I have no clue if it's similar to Virtuozzo, as it was never considered due to lack of HVM support (we're primarily a Windows shop on the VM side of things).
Posted by tim2718281, 10-28-2010, 12:23 PM
If you use the term "cloud" in a discussion, you cannot be sure that half or more of the particpants understand it to mean the same thing as you do.
Posted by KarlZimmer, 10-28-2010, 12:59 PM
Why does a VPS need to be restricted by your definition, "A VPS is a Virtual Private Server. It's one container, no redundancy, no high availability, limited vertical scaling, harder to manage, rarely entirely manageable from a web browser (both the client side, and Cloud administrator), and it's usually a more fixed cost (utility billing in many Cloud providers)."?? Even if it has all those things, isn't it still a VPS? VPS services have for YEARS included redundant back end storage, various forms of failover, etc. I don't see anything you list as not being able to be done with a standard VPS platform, such as Virtuozzo, that has been around for many years and is nothing new... Then how is a VPS inherently harder to manage? Web based VPS management has been around for years as well. The vertical scaling is determined by the hardware being used, not whether it is a VPS or a cloud, and the billing structure can be done with either, the way something is billed does not make it a different product. Personally, I find the term "cloud" being used primarily to allow companies to no longer need to explain the redundancy they have in place. Ask them what redundancies they have specifically and most sales reps will just tell you, "It's all in the cloud, don't worry about it." making it so you really have no idea what you're actually getting.
Posted by JasonD10, 10-28-2010, 01:25 PM
The VPS is not redundant/ha. It's running on one server. If that server goes down, so will it. Sure it may not be long.. but it still stops. It can only scale to ONE physical server. A true cloud application can and does scale vertically far beyond that. Harder to manage, absolutely.. they are one encapsulated container.. just one stand-alone OS for the entire application (whether it be a LAMP stack, or whatever). When you break that down into much smaller pieces (many independant and smaller OS's serving only that one particular piece of the appication) it's level of management is in a whole new mindset. I'm upgrading many applications at one time, no longer just one. And I'm doing all of this entirely from the web.. not some proprietary piece of management software that the company makes you use or buy. This is going to sound really silly but cloud is kind of like the first time you watch the Matrix. Once you realize what in the world is happening in the movie you say ohhhhhhhhh.. wow. Well that's how it is with Cloud. It's far beyond a VPS.
Posted by KarlZimmer, 10-28-2010, 01:32 PM
So you're defining the definition of cloud by simply restricting what a VPS is, even though HA VPS infrastructures have existed? To me, it all seems to be an extension of the exact same thing that a VPS is, just adding more management and abstraction software on top of it. Personally, I find the reliance on that closed source abstraction and management software to be more unnecessary overhead and an increase in management headaches, as you can't handle the issues internally, but to each their own. We've been helping customers make "cloud" applications and "cloud" based web sites for 6+ years now. The key is in development, properly developing a site, not in adding more overhead and abstraction on the infrastructure side. I believe that a cloud application can exist and is different, but I have yet to see anything that excites me as far as cloud hosting.
Posted by JasonD10, 10-28-2010, 01:41 PM
Sorry I am not making it more clear, but again your thinking is still tied to the technologies and methods you're used to. A VPS is just a virtual server.. let's set aside all "fancy" software. A VPS as you see serves it's purpose as a "regular server" but virtualized using virtualization technology. Right? You throw on a Linux OS under Xen, Viruozzo, etc and it's an independently controlled "server" (although virtual). A true Cloud application goes beyond that. It takes independent pieces of that VPS and breaks it down from the inbound and outbound networking of it, data storage and handling, and Apache/PHP, MySQL, and so on. So while a normal VPS may be 1 "instance", a Cloud application may be 8, or more if you like. Because of the independence of each component the scalability is far greater. A VPS is always going to be limited to ONE physical server in today's technology. Management of a Cloud application I beg to differ is far easier once the platform is developed. It can be re-used on many customers, and you can upgrade or maintain each independent piece for all your customers at once. The developing, staging, and launching of any independent piece is far easier. I'm not bashing VPS technology as I have used virtualization since XenServer was XenSource. I love virtualization, and it is a core function of Cloud. But Cloud is the next level of virtualization, plain and simple.
Posted by packetm0nkey, 10-28-2010, 02:51 PM
Sounds like you are using 3tera's Applogic product.
Posted by JasonD10, 10-28-2010, 03:11 PM
We used to use our own proprietary platform which did all of this.. but AppLogic does it way better than we did so we dropped our platform in favor of AppLogic earlier this year.
Posted by eming, 10-28-2010, 03:26 PM
Karl, you should get an award for telling it like it is... Even though it doesn't exactly support my personal business plan right now, I totally agree with the 4 statements above...And I think - in time - the cloud will be commoditised and hopefully the term will sieze to exist. Right now it doesn't do us (cloud hosting/software providers) any favours. D
Posted by packetm0nkey, 10-28-2010, 03:31 PM
I'm very familiar with their product, and while it makes sense for larger sites, deploying an application per customer who just wants the redundancy aspect of the cloud who can scale vertically easy enough will find it to be overkill in terms of complexity and cost. Now, should you choose to build a multi-customer hosting solution application on Applogic, and offer something similar to Rackspace's Cloud Sites, or Mediatemple's product I can see that assisting in the creation and management, but then you are just building what you could do with a normal server cluster virtually. Either way virtualization is cool and the term cloud is indeed convoluted. No one solution will fit everyone's needs, which is why there are many providers offering different services.
Posted by JasonD10, 10-28-2010, 03:39 PM
Thank you for wording that better than I could.. my big "tech head" gets in the way too often where I have a hard time wording things... simply. However, just one small piece of kickback.. we are doing this and launching a "virtual cluster" if you want to call it that in a matter of minutes. You simply cannot do that with physical hardware, even with all of the PXE bootup scripts and automation in the world (I know, I've done it for many years). This can even be done for smaller sites Ie: a $10/mo customer, actually we are doing that. We're also launching large dedicated customers that are paying us 4-5 figures/mo for this kind of service but it is cost effective and possible to do with much smaller sites too, and be multi-tenant. It all depends on how you want to do it but the technology is there.
Posted by KarlZimmer, 10-28-2010, 03:56 PM
A VPS itself might be limited, but simply use the VPS or dedicated server as a building block. It is quite simple to use load balancing technology that has existed for 10+ years, MySQL clustering, etc. to expand out your solution using standard, open source software, without needing to be entirely dependent on a closed solution. In addition, those solutions allow you to limit your additional overhead and abstraction to only those things that need it, not your platform as a whole. We've had customers splitting out VPSes and dedicated servers for different tasks, often 8 or more of them, for 6+ years, that is nothing new... Why can't you do that with physical hardware or a VPS environment? With various SAN solutions, etc. you can mirror the data across many different physical nodes in basically no time at all. I had NetApp in here showing me how they created 1000 copies of a VPS in about a minute.
Posted by packetm0nkey, 10-28-2010, 03:57 PM
I figured I could learn more from your website, so I took a look and I see exactly what you are doing. Very cool to see someone using applogic for that. It has been a few years since we last used them so I imagine it has only gotten better and more reliable. Those 10-45m restarts for large applications, and volume re-size times were a killer.
Posted by JasonD10, 10-28-2010, 04:05 PM
Thanks. Yes I believe they've come a long way, although I'm a newcomer so I can't really speak on behalf of their past. The large application restarts still do take a while as it is starting and stopping a lot of appliances, but we don't really ever restart a whole application anyways. There's just no need. Volume resizes are going to take time no matter what the technology is. It's mostly limited by the actual FS you're using. Some are faster, some are slower. We found a better approach is to simply create another volume, mount it by doing a quick reboot of that appliance then do what you want to do with it (add it as another volume permanently, rsync the data over and remove the old, etc). This way it's less intrusive.
Posted by JasonD10, 10-28-2010, 04:12 PM
Who said it's all closed? We're running various Linux and Windows distributions unmodified with a lot of open source software. Congrats on 1000 copies of a flat VPS in a minute, but what I'm talking about is an entire Virtual Private Data Center, very complex multi-layered applications in a matter of minutes. That's what impresses me, copying a simple flat VPS template I too have seen and done for many years. That's old news...
Posted by JasonD10, 10-28-2010, 04:25 PM
Here, to better illustrate what I'm talking about I just quickly created a VPDC consisting of various load balancers, nas's, gateways, port switches, linux and windows web servers, mysql and mssql servers, and a monitoring appliance. Hope it helps... Attached Thumbnails
Posted by KarlZimmer, 10-28-2010, 05:16 PM
You're saying AppLogic isn't closed source? When something goes wrong with it or behaves oddly, you need to contact them for support, you can't look into it and see what the code is doing itself and resolve the problem yourself, correct? I know that was a large problem, and why we've never considered using it, seeing the issues LayeredTech and others had with them early on, and then being entirely dependent on a third party. You have the images for the different applications as well, how is that any different? You have the images made, turn up an image for each of the different types of services, if you have a common set of applications just script it. We've been turning up VPS accounts with control panels on them, or VPS accounts for strictly nameservers, and other similar setups within minutes for years now. You need a Windows server, just load the Windows image, for a Load Balancer you load your load balancer image, not difficult.
Posted by JasonD10, 10-28-2010, 05:34 PM
That's a semantics discussion as whether I use Citrix XenServer or AppLogic (and I do use both) I expect, and want support from the makers of that software. However, I can and do troubleshoot them myself quite often. We're not dependant on a third party any more than you would be dependent on the manufacturer of your car that you drive. I have a high level access to AppLogic (root to all levels) and can get into it and do just about anything I want. Again, anyone can turn up individual VPS's, even with a bunch of custom templates but they are still independant VPS's. These are complex, completely integrated applications that are far beyond the comprehension of a bunch of individual VPS's. It's decoupled management of the OS and Application, VPS's are completely integrated into the OS so they are not decoupled. Here is CA's own words to describe it: AppLogic operates on the logical structure of the application, enabling you to package an entire N-tier application into a logical entity and manage it as a single system. This approach also makes it very easy to assemble, deploy, monitor, control and troubleshoot applications visually in a browser.
Posted by KarlZimmer, 10-28-2010, 05:40 PM
Or, you can just properly design your software in the first place instead of needing to be dependent on an abstraction layer with the additional issues and overhead that that brings with it. And AppLogic gives you the source? With the car, you own the whole thing, you have the "source" and you can modify it as you wish and sell it as you wish, it is your property. To the best of my knowledge, AppLogic is closed source and thus puts major restrictions on those sorts of things. If there is a general stability issue, you're dependent on that one commercial entity to resolve the problem, you cannot resolve it yourself, that is not the case with OpenVZ, Xen, or KVM.
Posted by JasonD10, 10-28-2010, 05:46 PM
Sure, if you want to completely develop, maintain, and do that you can and no one is stopping you but at that point you might as well become a software company and sell it like 3tera did for $100M as well. haha. I choose to focus on my business of hosting instead of developing and managing a software platform myself. But hey, to each his own.
Posted by KarlZimmer, 10-28-2010, 06:32 PM
Who is saying to develop your own software? I'm saying being able to support, modify, fix, and adjust the software that has already been developed.
Posted by JasonD10, 10-28-2010, 06:40 PM
Ahh I see you modified your post above and added OpenVZ, Xen, and KVM. Those are virtualization technologies not Cloud. It is a component of Cloud, just as Xen is a component of AppLogic. Anyways, at this point we're beating a dead horse so before we convolute the thread any more let's move along...
Posted by DiegoRBaquero, 10-28-2010, 06:43 PM
I wonder, how is bw distributed along the systems? 1gbps from 10 servers could achieve 10gbps then?
Posted by JasonD10, 10-28-2010, 07:10 PM
Yes, it does.
Posted by brentpresley, 10-28-2010, 08:39 PM
Even more. We are about to start working on Infiniband to see what would be required to get it to work with AppLogic 2.9.9. Our starting point will be 40Gbs, to utilize on the backplane for node mirroring.
Posted by hornet1999, 10-30-2010, 05:10 AM
Agree! It is just like the same thing viwe from the left vs. view from the right.
Posted by sylye, 11-15-2010, 07:06 AM
I have been searching in the net trying to find out why is the industry is so sick in selling marketing gimmick by calling their product as 'Cloud' but in fact they are NOT! I just want to know is there any provider out there who is truly Cloud Server and then I see your thread. I'm really glad as there is someone really know what he is doing by not mixing Cloud and a VPS. I'm really really sick of that. I am using Rackspace Cloud Server and only just know recently they are just another form of VPS and the customer support I contacted just now have admit their product is not a true Cloud, and he even claim there is no such thing on earth but I doubt it. I guess your statement has clearly prove they are wrong. There is people really doing it and by not knowing how to do it didn't mean it's not exist. Please give some technical detail if anyone out there think Rackspace Cloud Server is a true cloud. Cloud is just not VPS!! I think this statement say it all what is a Cloud is
Posted by JasonD10, 11-15-2010, 10:16 AM
Well, it's an unfortunate event that happens in many industries and definitely is not limited to just hosting but more companies want to ride the wave of Cloud because it's good marketing, and will bring in new business even when they don't technically do it. It's even easier to do because 99% of customers won't have a clue. How could they? Of course the Rackspace tech is lying, but to be fair I wouldn't hold the entire reputation of the company against one rogue tech. He is limited to his own knowledge and company and I'm sure he believes they are the best, and because they are limited to their own technology they think everyone else in the world is as well. You see this again in many other industries so you walk across the street and find someone else doing exactly what you want. I agree with Andrew's comment on Cloud for the most part, however it is not limited to LAMP, and the virtual environments are usually broken down much more than the whole LAMP stack. We are doing it by breaking it down into much smaller components. We have achieved this with WISA now as well (even Windows Server 2008 R2 (64-bit). Take what I said and that and sum it together and that is the true essence of Cloud.
Posted by jayglate, 11-15-2010, 06:12 PM
Well lets clarify a bit what the cloud is and what the cloud isn't in respect to WHT. When we here talk about the cloud, we really mean IaaS, Infrastructure as a Service. The term cloud really entails IaaS, PaaS and SaaS, but more on that later (maybe) So what do I think a redundant and scalable IaaS deployment or what everyone here is talking about a true Cloud Deployment should be? Network Fabric: 1gbps, 10gbps, inifiband it doens't matter what combination you run it it in, as long as it is redundant. Some stacks can be merged on the SAME pipe using VLans, I.e Public/Private IP traffic on a 10gbps pipe with SAN on an infiband pipe. Or any combination. But the key is redundant switchs and redundant cards. Storage Subsystem: Redundant when I mean redundant I mean redundant head nodes, with redundant HBA's, redundant JBOD's dual dual sas expanders and dual ported SAS drives with a fully meshed SAS architecture. Also a storage sub-system that can handle the IOPS and be a true replacement for a dedicated server. These are connected to the SAN Fabric via redundant 10gbps or inifiband ports. Hypervisor Layer: Redundant Power supplies in each hypervisor, redundant NIC Cards in each hypervisor if its inifiband or 10gpbs or 1gpbs doesn't matter.. once again redundant plugged into the redundant network fabric. That if one goes down the VM is immediately started elsewhere, and if you need more resources than that hypervisor has available your VM can be moved to a hypervisor with resources available to accommodate your needs. This my friend is as true "cloud" IaaS service. Now I am going to be a tad mean and say a few organizations that are just slapping what IMO is a true cloud. And are flat out using the world cloud wrongly and misleading you as to what you are getting and what they are selling you. It really grinds my gears how shameless these organizations are and how bold faced they are in misleading their customers and the public on their "cloud" which is just a standalone VPS server with a pretty gui that if it goes down you are hosed and hosed hard.
Posted by JasonD10, 11-15-2010, 06:48 PM
For the most part, that's right but I'm trying to base my definitions and reasoning on Wikipedia's definition of Cloud Computing. Being a neutral source it seems to be the most fair way to discuss it. As IaaS, you're still missing the accessibility of management of that infrastructure over nothing more than a web browser, economics (utility or subscription based, in either case not having the upfront cost normally associated with a hosting environment). We also need to see the ability for multi-tenancy, automated scaling, automated migrating, device and location independence. Whether public, private, or hybrid cloud these are still key factors. Where I'm getting at is, it doesn't stop at the hypervisor layer as any virtualization layer is capable of achieving your definition of Cloud. There is a much more advanced set of software on top of the hypervisor to truly bring this to the next level. Now this is debatable whether or not it's necessary to be in today's definition of Cloud, but I think decoupling of management from the application and os layer is another step that greatly improves how technology is implemented and will work it's way into the full next generation of Cloud technologies. Personally I feel this is the most important thing as this is what makes vertically scaling beyond a single physical server possible in today's technological limitations.
Posted by jayglate, 11-15-2010, 06:53 PM
The biggest problem you will run into with 3tera unless you use external storage ( and set it up correctly), is going to be your IOPS per TB of data.
Posted by JasonD10, 11-15-2010, 06:58 PM
Actually I would argue it's faster than any other technology. Why? Each server is an independent but redundant and high available SAN. So let's say you have a cloud of 15 servers, and storage is N+1. Traditionally you're setting up two SAN's for the same level of ha/redundancy (one standby), but far less capable speeds due to hardware limitations if we're talking apples to apples as far as total cost invested in a Cloud of 15 servers. So, 15 independent SANs even of lower speed drives is going to operate faster than 1 active SAN as we're talking about a whole lot more independent hardware not limiting each other. Networking the slowdown you're referring to? Connect them via Infiniband instead of Gbps...
Posted by jayglate, 11-15-2010, 07:04 PM
You are 100% incorrect a group of 15 interdependent SAN of 4 to 8 drives in each SAN will not outperform a well designed storage network like I described. And total cost might be a tad more due to the head nodes but storage expansion will be cheaper and provide more redundancy and more scalability, also having a setup like this allows for easier creation and management of different storage volumes, ala SSD, SAS and such. A properly designed redundant independent SAN will not be limited in anyway shape or form and if designed correctly can be expanded and grown far more cost effectively than the 3tera approach. And yes, having a 1gbps backplane for all the 3tera SANS to connect to would be IMO a huge limiting factor on how large, and how fast your SAN can ultimately grow to.
Posted by brentpresley, 11-15-2010, 07:04 PM
This is questionable at best right now, and we are in testing with CA/3Tera on integrating full 40Gbs Infinniband support into the kernel, which is not (yet) supported.
Posted by JasonD10, 11-15-2010, 07:07 PM
I would be interested to see the results, side by side of your proposed solution of this "properly designed SAN". In our initial testing we're taking larger implementations with shared SAN's (Promise arrays) and seeing drastic improvements in speeds (and cost savings) going to the 3tera local storage approach. This is on a 12 server cloud. Care to share some details to prove your theory?
Posted by jayglate, 11-15-2010, 07:08 PM
So you want to interconnect the 3tera nodes via Infiniband with IPoIB for communications? That would remove the network bottleneck even with the IPoIB overhead however I think 40Gbs is overkill make sure you take into account for the PCI-E bus speed and any redundancy concerns before committing to 40Gbs vs 20Gbs. Inifiband is a wonderful technology and fairly cost effective to build a 3Tera grid like you are talking about. But IOPS isn't just the network, IOPS is a factor or spindle and speed of the drives.
Posted by brentpresley, 11-15-2010, 07:10 PM
Please see my above reply, but this is just simply not the case. A good external SAN is bloody expensive compared to the 3tera approach. And as CloudWeb and I have pointed out, if you have a proper backbone network (10GbE or IB), then you can really crank out the performance for a lot cheaper than an external SAN.
Posted by brentpresley, 11-15-2010, 07:13 PM
With 3Tera, we have developed a method for implementing full RAID 0 or RAID 10 on their platform, so spindle speed is far far less a factor than when utilizing single drive setups per node. Plus, we have already taken into account the PCI-E theoretical limitation on Infinniband by using 16X onboard controllers that are not limited by the interface. Also, redundancy is built into the 3Tera grid system by default, that is the beauty of it by design. It's the "node" equivalent of RAID 5 or 6.
Posted by jayglate, 11-15-2010, 07:22 PM
MODS please correct this if I am not allowed to "self promote" but here we go. Our SAN for our upcoming cloud, is designed to my SPEC as outlined. Redundant SAS 6Gbps wide internal SAS paths. Giving me 24Gbps of speed per each SAS Wide Channel 2 Channels per HBA so 2 HBA's per system gives me 48Gbps of SAS connectivity. Volume 1 currently has a total of 20 Drives in it in diverse JBODS with running RAID10 across both chassis (so insane reliability). With SSD Caching on READ and WRITE. Each head node is maxed to RAM (so we can do a lot of caching on the head nodes before it even bothers the SSDs). We are able to achieve on Volume1 1600MB/s READ and 1300MB/s WRITE. (Volume1 is our low performance Volume, we have Volume 2 and 3 which blow this out of the water, in terms of speed and IOPS) So to convert it into English most of the WHT community can understand. 16 Gbps (almost 2x 10gbps circuits)and 13 Gbps (a little more than a 10Gpbs circuit) With 40Gpbs into our SAN fabric for the hypervisors to talk to. Yes the initial outlays we expensive, however growth of the storage sub-system is very very cost effective. Our SAN will out perform any SAN/Cloud in the market mass cloud market today and 90% of them in the specialized cloud market based on our testing across over 30 different "clouds" in the market place, in the given drive segment. Ala SAS/SATA vs SAS/SATA or SSD VS SSD, Fusion IO VS Fusion IO Including those that aren't real clouds just VPS's with a pretty gui calling themselves clouds. So there is the secret sauce behind DedicatedNOW upcoming cloud. it will only get FASTER....
Posted by brentpresley, 11-15-2010, 07:24 PM
And the cost?
Posted by jayglate, 11-15-2010, 07:27 PM
Expensive I wont lie, but it is what a TRUE IaaS cloud service should be IMHO and that is what my customers and the market deserves. But, adding extra capacity is a fraction of the cost of the initial capacity. So just like building any infrastructure its expensive to build but cheaper to grow and maintain.
Posted by brentpresley, 11-15-2010, 07:29 PM
For comparison, our only additional cost will be for the Infinniband switch and cables, as our current motherboards already have IB onboard. From that, our max theoretical speed is limited only to the speed of each node's local storage, at zero additional cost. And of course, there is always the option to add on an external SAN if something custom was needed for a customer. I guess what I'm getting at is this: if you want to use local or external storage, 3Tera gives you that option. I have not seen this from other cloud software vendors, at least not in an easy to implement perspective. And even utilizing the local SAN, it is a TRUE IaaS cloud service.
Posted by JasonD10, 11-15-2010, 07:39 PM
Expensive.. but yet your argument was a "tad bit more". That sounds like a ton more. Not exactly an apples to apples comparison.
Posted by KarlZimmer, 11-15-2010, 07:48 PM
Well, how much money is the 3tera licensing costing you? When we were talking to them before that wasn't cheap either...
Posted by brentpresley, 11-15-2010, 07:50 PM
Privileged info, but let's just say it was the best buy for us, by a considerable margin.
Posted by KarlZimmer, 11-15-2010, 08:00 PM
But then how does that recurring cost compare to the ONE TIME additional cost of the storage solution? My guess is that in a large deployment you're spending more money on those licenses, over time, than you would on the SAN anyway...
Posted by jayglate, 11-15-2010, 08:25 PM
Based on what you just said you have a massive single point of failure with no redundancy at all for your ENTIRE cluster. IB on board. I could be wrong, but it sounds like you only have a single IB controller and single IB switch to power your entire backend SAN network. Thus IMO you are close but no cigar in offering a truly redundant IaaS Cloud based solution to your customers and to the market.
Posted by jayglate, 11-15-2010, 08:27 PM
No, not really, when scaling since our JBODS have a lot of empty capacity it is just adding in additional volumes and more JBOD's without the need for more networking, head nodes, hba's etc. etc.
Posted by jayglate, 11-15-2010, 08:28 PM
Expensive can mean completely different things to everyone. But I am not saying it was the cost of a 3Par or an IBM XIV, but it wasn't the cost of throwing a few drives in a 2U server either.
Posted by KarlZimmer, 11-15-2010, 08:34 PM
I'm agreeing with you, saying that adding the overhead costs of a SAN are prohibitive, when it is a one time cost that then saves money and effort on expansion, whereas the licensing fees for AppLogic or a consistant recurring fee, being a much greater cost long term, or so I'd assume, based on the pricing I had seen from them.
Posted by brentpresley, 11-15-2010, 10:03 PM
You are going under the false assumption that I would be buying additional nodes just to expand my SAN. That is completely incorrect. I have never once purchased nodes for the sole need to expand the SAN.
Posted by brentpresley, 11-15-2010, 10:06 PM
You misread, but also mis-assumed. One IB controller PER NODE, and we can deploy as many switches as needed to power the backend SAN network. This varies by datacenter for us, but there is always at least one layer of redundancy w/ GbE as a backup.
Posted by JasonD10, 11-15-2010, 10:23 PM
While I don't mean to go off on a tangent, I do want to set some things straight as to why we (and I would think most anyone) would use a commercial Cloud platform such as AppLogic. First, we did not purchase AppLogic solely for a SAN solution as to do so would be taking away from everything else that it does. But on the data note, we had used everything from free Openfiler redundant SANs with HA+DRBD, Datacore SANmelody, as well as commercial hardware SANs when we developed our own Cloud based on XenServer. We've been using Virtualization technology for 4 years and were very pro Xen ever since XenSource v2. While SANs were of course a big consideration in Cloud, AppLogic was chosen for it's sheer ability to produce a true Cloud platform, and do it well. As I've said time and time again I have never seen any other product come close to offering what they do and the #1 thing they do for us is allow us to properly decouple management at the application to the OS. We're building near limitless scalable infrastructure's not only LAMP, but WISA on 2008 R2 for high profile sites, fortune500, and SaaS providers. Yes, AppLogic started as grid 7 years ago but beyond the marketing and buzzwords they have the best solution in the industry bar none, and I have nearly tried them all including our own. CA didn't go buy up this small company for $100M for nothing. To simply compare it for it's SAN benefits and nothing else is not a fair argument. We wouldn't compare a Ferrari to a Chevrolet and only talk about the gas tank so let's not dumb this down to that either.
Posted by jayglate, 11-15-2010, 10:30 PM
We were I think the 2nd partner for 3tera many many moons ago. Unfortunately it had more problems and redundancy issues than I can to go into. I think now they just finally fixed their redundancy issues where if the controller dies it all dies. And yes their "white board" is pretty pretty. I also really hated and still hate their limiting of in and out network paths, gets very annoying. I also feel CA bought them because well, they wanted to buy someone and with Vlad god rest his soul dying, 3Tera really had to sell possibly due to loss of investor confidence. But either way Bert and the others there are great people to work with, either way . Just not a fan of their technology.
Posted by JasonD10, 11-15-2010, 10:37 PM
Sorry to hear of your trouble with the platform early on. Redundancy issues have all been addressed at this point and it is entirely redundant including support for multiple WAN/LAN switches/ports, and the controller at least for our Cloud's are replicated on a handful of servers at least. Not sure about the limiting of network path's, I've yet to hit a wall *shrug*. But anyways I'm sure as you know being in this business a long time too when looking back many technologies throughout the years many things we probably thought were going to be flops ended up being the new standard. At this point there's no denying the shift and adoption of Cloud no matter the platform, and time will tell what platforms and methods emerge to set the standards for what will be remembered in another 12 years. In any case, thanks for the good points and talk. This kind of conversation is what makes WHT better.
Posted by KarlZimmer, 11-16-2010, 02:02 AM
No, I am not, I am talking about the overall cost structure. I am saying that significant money is being spent on licensing from a particular provider and that amount is probably minor compared to what people are complaining is a significant investment in a SAN. I am not saying one is better than another, I am saying that complaining the SAN is expensive is a ridiculous argument for saying it is a worse solution... if the cheaper solution were better, why pay those rates for cloud software?
Posted by brentpresley, 11-16-2010, 02:04 PM
I'm pretty sure you are missing my overall argument: 1) local SAN on 3Tera can be designed to be as good, and as fast as an external solution. 2) License costs, for us at least, were better than any other competing product. From that perspective, the decision was a no-brainer.
Posted by KarlZimmer, 11-16-2010, 02:17 PM
I had started that part of the argument with CloudWeb, who was the one who said, "Expensive.. but yet your argument was a "tad bit more". That sounds like a ton more. Not exactly an apples to apples comparison." I am not missing any argument on your part, you're the one who entered into the argument when it was already going... 1) Yes, to get something as fast or reliable as a true Enterprise SAN you're going to need to be spending millions of dollars on research and development. Not saying it isn't possible, but.... 2) Again, I'm not saying it is a bad decision or comparing licensing costs, I'm saying that spending the extra $50k one time on a SAN is nothing compared to say $50k a YEAR in licensing fees. I am saying it is a minimal cost increase to go with a SAN, all things considered.
Posted by brentpresley, 11-16-2010, 02:19 PM
1) Or just partner with the developing company and cut your costs to a few $k. Who do you think is paying for the Infinniband testing we are doing? 2) You once again are ASSUMING that I purchase even a single NODE for just the purpose of a SAN. This is 100% incorrect. I NEVER purchase an additional license unless I need the CPU cores and RAM for customers. The local SAN comes along as a "free perk". Therefore, from my perspective, it is spend 50k on an external SAN, or work from the local SAN and get good performance for ZERO extra cost. That's a no-brainer, as I have in the past pointed out.
Posted by JasonD10, 11-16-2010, 02:26 PM
$50K? The number is relative to the servers.. but assuming we're talking about the 12 server Cloud as we were comparing apples to apples, not even close.. as a matter of fact by the time you cross that number in AL licensing your hardware will be obsolete. But again, it's important to note AppLogic is not just some "SAN" technology, it's Cloud. It's not Virtualization, it's Cloud. It's why any company buys any commercial product. We can all go around using opensource all we want, and for many cases it does fit well but imho there simply is nothing even close to this level yet. I loved XenServer and was a big fan even from the old XenSource v2 days, and I happilly walked away from my Citrix partnership for this as it's worlds better. Even the largest Cloud platform (Amazon) doesn't have the features, stability, and capabilities of it.
Posted by KarlZimmer, 11-16-2010, 03:20 PM
1) And you're getting the same performance as an Enterprise SAN? if you're still testing, it doesn't even sound like a final product... 2) No, I am not. You're entirely missing my point, I've explained it twice already... I guess if I spell it out with numbers it will help? OK, so you build a cluster of 100 servers, that is say $400k in hardware, you spend $60k a year on software licensing, you spend $100k a year on space and power, that $50k for head end nodes of a SAN are simply a minor cost. As I already said, "saying that adding the overhead costs of a SAN are prohibitive, when it is a one time cost that then saves money and effort on expansion" was ridiculous, that is my sole point. I am not comparing, I am simply stating that to exclude something because in the grand scheme of things it might increase your costs by 10% is ridiculous. Why are we comparing "the 12 server cloud" where did that even come from? A cloud environment really makes sense with only 12 servers? I would think to get the best out of it you'd want to scale it as big as possible, like Amazon Web Services, etc. I wasn't even thinking of starting any sort of cloud service with less than 50-100 nodes... When you're looking at an enterprise cloud solution you're looking at a solution to support hundreds, if not thousands, of front end nodes, so you're correct, we aren't comparing apples to apples.
Posted by JasonD10, 11-16-2010, 03:28 PM
Well that says it all then. Your solution only works for Public Clouds and MIGHT make sense starting at 50 servers. What I was referring to was a 12 server PRIVATE cloud, aka virtual private data center. There are many cases where this makes a whole lot of sense. For this particular customer we're going from 28 servers (non cloud, traditional environment) to 12 in Cloud. It all depends on your customer base and what you're trying to do but that is highly problematic at those levels. Many customers simply do not want to be on a public cloud. It works for some, but not all.
Posted by brentpresley, 11-16-2010, 04:02 PM
This is my point, with Applogic, there are NO HEAD END NODES for SAN required if you use the local storage SAN. So I don't have that 50k cost. ZERO additional cost. ZERO additional hardware. GOOD performance, which can be scaled to various degrees depending on your back-end network I/O. Is that clear now?
Posted by KarlZimmer, 11-16-2010, 04:35 PM
Sure, I understand your point and have never had an issue understanding your point, but I still have no idea how it has any correlation of any kind to the point I am making....
Posted by KarlZimmer, 11-16-2010, 04:41 PM
This thread is about VPS vs. Cloud, right? You really think the OP was asking about comparing a single VPS to a 12 server private cloud? I would assume based on the discussion in general that it is VPS vs. Public/Shared Cloud. If we're now comparing a single VPS vs. a full private Cloud configuration that opens up a whole new can of worms... And yes, that makes sense when you have customers who are terrible at developing their own application... We've moved multiple customers from "cloud" environments to 10+ dedicated servers (one with 100+), because they properly crafted their applications and were seeing a waste of resources and IO issues on the cloud, thus dedicated servers were actually easier and saved them a significant amount of money. My guess is, that that customer could have seen the same savings, if not more, by properly engineering their application in the first place. Now, I do understand that some companies simply don't have the resources and thus there is a market for it, but it doesn't mean that is the best option.
Posted by JasonD10, 11-16-2010, 04:56 PM
The OP simply asked the difference, we later went into comparing type's of Clouds and implementations. This is the Cloud forum, and the thread has sparked quite the conversation. The conversation going on between us and Jason have brought other things to light, both wide and smaller scale. However now you're going into a whole other arena challenging a poorly designed application where Cloud had no role in that. Based on your argument then I could take your now optimized environment and scale it down in the Cloud yet again and achieve 100% high availability, redundancy, vertical scalability, and proper use of the resources across all the servers instead of leaving so many servers wasting unused resources. The 28 server client I was referring to was already highly optimized, and we are making very few application level changes (we're actually adding more layers which would appear more complex, but instead separating roles to provider more independence). By going Cloud we're bringing this down to 12. It would not be possible in a non-cloud environment no matter how optimized the environment was due to the complexity of it. We simply need that many OS's and in a Dedicated environment, with the latest and greatest hardware, it could not be achieved in any less than 16 servers and even then it has limited high availability as we're talking a mix of MySQL and MSSQL databases. HA in MSSQL is both expensive in achitecture, and licensing in a traditional environment. In Cloud, it's not.
Posted by KarlZimmer, 11-16-2010, 07:43 PM
So where was the private cloud brought up? The first I see it mentioned was your response to me, so how is that what we were talking about? If you're talking needing to mix MySQL and MSSQL databases it certainly sounds like the application could have been better engineered... Yes, it would be possible outside the cloud. The cloud is not magic, you still have the same resources both ways and the cloud adds MORE overhead. Sure, it might not be trivial and it may be cost prohibitive, but adding more layers will always increase overhead. As I said, it may certainly be the simplest or most economical solution, but that doesn't mean it is the most efficient solution. To note, I agree completely that there is a place for the cloud and a cloud market, I just feel there are a LOT of problems with it. First, that there is no real definition, everyone just calls their product a cloud without differentiating them, saying what redundancy IS in place, etc. Second, the idea that it is the correct solution for everything, which it is not. And finally, that it is some magical thing that somehow makes things simpler by abstracting more, but that somehow that abstraction also increases efficiency, which simply isn't true. Certainly, it can improve utilization of resources out of a poorly engineered environment, but if your application is properly engineered in the first place there should be no need for the cloud. The "cloud" should allow everyone to build high availability and scalable applications, something previously reserved for larger firms. Those larger firms have been building massive applications for 10+ years, without the "cloud" by simply following good development processes and building good code. I feel many see the cloud as a way to escape from many of those best practices and we already see terrible code where customers just say to throw more hardware at it instead of fixing their application problem. With the cloud, I simply see that as being a greater problem, as they don't need to worry about hardware limitations. To me, that seems to bring about less efficiency, as people will have the idea that resources are able to be infinite and program without concern for resource utilization. My issue there is more of a general and idealistic issue though. To me, efficiency is always the goal, and the cloud represents no longer needing to be efficient in programming or application design. People saying that it is then more efficient upsets me... We are working on a couple cloud service projects ourselves, I'm not anti-cloud in general, I just feel it has a lot of marketing hurdles to get over, as most of what I see is outright lies. Our goal will be to clarify what we're doing, exactly. I know it may alienate some users who don't want the technical details or just get confused by them, but to me being honest and not just a used car salesman is important to me. people need to understand what they're getting, and so many cloud providers, cloud applications, etc. don't really tell you what you're getting...
Posted by packetm0nkey, 11-16-2010, 07:47 PM
I'm a bit confused. How does the cloud offer vertical scalability past the resources that one physical node can offer, and how does that help your MSSQL HA problem of licensing and required resources?
Posted by badboyx, 11-16-2010, 08:54 PM
a new way to steal money out of your bucket
Posted by JasonD10, 11-17-2010, 12:32 PM
I feel you with the Cloud marketing challenges. It's one of the most convoluted technologies I've adopted in the 12 years I've been in the business and I do put a lot of time into "clarifying" that with customers, or potential customers. It's a shame, but it's a short term problem and in time it will clear up like all things. Now, when talking about your scenario, a cluster of dedicated servers, that is comparable to a private cloud scenario. You're not sharing that cluster of dedicated servers with hundreds of other clients, right? If so, then it would not be a dedicated server cluster for one customer but a shared hosting environment. So, in the cloud market the fair comparison is a private cloud platform as that too is a cluster of dedicated servers however it is adding the cloud layer to it. This particular customer is not just one application so I apologize for the confusion as I should have just simply said it is one customer, it is an Internet business with many independent pieces and several applications. Some of the components are MySQL, some MSSQL, some PHP, some .NET. Some of the software is their own, some is open source, some is commercial paid for software platforms, and so on. It's an online business, plain and simple. So in a perfect world let's say it's designed to the best of it's abilities as this has no direct correlation to Dedicated vs Cloud anyways as we'll see moving forward. You're right, the Cloud is not magic and neither is a properly designed Dedicated solution. There is no magic in anything once you understand it. Both infrastructure's require a serious level of skill to properly architect to this degree as you very well know. Providers than can truly do this right with or without virtualization, or Cloud, are unfortunately few and far between. As you said, many just want to throw hardware at the problem. Imho they do not have the customers best interest in mind when they do that as it is wasting resources, which is wasting money. What we want to show is the comparison of the technologies used and the outcome of that launched application, or set of applications in this case. To optimize that in a Dedicated environment without wasting money on unused resources it can even be more time consuming, and there is going to be more waste than the Cloud overhead you're talking about. I will get into that later on the Dedicated side. The overhead in a cloud platform we have seen is ranging from 8-12% per physical node. These are not just our numbers, but if you search for whitepapers on Xen you will see these are reliable figures and some sources cite lower numbers but this is the range we've seen so I want to be very fair here, with no fluff trying to point things in the Cloud's favor. Let's face it, I've done Dedicated much longer than Cloud so I'm obviously not anti-Dedicated either. So, let's say we're wasting as much as 12% on every server in this particular 12 server cloud. That's 1.44 servers going to waste just in overhead. But is it really? Let me lay it out to show you what this consists of and compare it to a non-cloud (dedicated) environment. This client is on a cluster of dedicated servers now with us, we have 28 of them. Here's the scoop. You can assume anything that is listed as 2x and 4x is for redundancy and high availability and the 4x for scalability as this business typically gets /.'d or other forms of media so traffic spikes very high. Right now, that is what they have (for the most part, just a few components are short of fully ha/redundancy but that is not my choice.. it was budget or otherwise administration limitations): 2x firewalls 2x load balancers 4x windows iis web servers 4x linux apache servers 2x linux apache servers (another application, must be a separate environment due to very different application requirements and admin level access to it) 2x proprietary environment 2x windows mssql servers 2x linux mysql servers 2x mail servers 1x admin 2x proprietary nas servers 2x SAN for DBs 3x servers for spares as the 25 listed above are not all identical servers, and even only having that much spare hardware is a little stressful on us. This is the dedicated server environment, and my comments earlier about moving them down to 16 would be using bigger and better hardware to take the 4x clusters to 2x (saving 4 servers), losing the spares (dumb but possible due to the 2x of each environment I guess), and combining applications which may not even be possible but for discussion sake let's assume it is. So there we have it, 16-21 servers in a dedicated environment if you want to achieve true fault tolerance, and when those spikes come from mass media attention, /., or other massive generating traffic sources we better have enough free resources on those servers to support it, right? Now comes the Cloud. In this case, we are building two Cloud's. One with enterprise SSD drives, the other... not. hehe. Otherwise they're the same CPUs and chassis's on all the servers. 2x Westmere's in each server. That's 24 hyperthreaded cores per server, or 288 in all 12. Oh, no more external SANs either. We're also adding redundancy and ha to all components of the infrastructure, some of which at a level above the core OS as with cloud we have found ways to decouple management at the application and OS level. This simply cannot be done in a dedicated server environment as you can only have one OS per physical server. There is no decoupling as the OS is tied to the server, and if you want to upgrade it, (here's a big one) replace the whole OS without effecting the application, replace the drives on it, upgrade the CPUs, etc it's going to go down and an engineer is going to have to manually work on it (in some cases they can do a few of those things remotely). Cloud, we do all of this from a web browser anywhere in the world. Physical hardware is just a commodity to the applications that are running on it. Nothing is tied to or dependent on a single, or particular piece of it. Now, here's what I hinted at before about resources. In a HT-aware environment which any good Cloud is we are just simply telling pieces of an application how much resources it can have. The cloud is scheduling, and evaluating the operating conditions of these environments as they are not tied to a server(!). Huh? Where are you going with this? It is quite simply using all of the available resources effectively on each server, and even going as far as to schedule that to use free resources on an idle core before a thread on a busy core. So we are using up servers efficiently and effectively, as well as sharing the BEST of those resources (idle cores) for each application running on it. At different times of the day and with a complex set of applications your demand for resources will vary and in a Cloud we are actually capable of using all of those resources and in doing so it is highly efficient. The same applications run faster as they are not limited to that one physical server they're on hitting busy threads once all the cores are no longer idle. Back to Dedicated, to even speak apples to apples let's use that 12% figure again. If you're building a dedicated cluster equally as efficient then every server in that cluster could only have a max of 12% bursting ability as otherwise the whole argument of efficiency in Dedicated is gone because that's what number we have seen. However as I've shown you're still limited to that one physical server and the amount of idle cores it COULD have. But, that's just talking flat efficiency on total resources, not considering any other benefits to Cloud. N+1 in Cloud is just... N+1. An 8 server cloud means 1 server, a 50 server cloud means... 1 server. And they're all the same hardware. It's just easy. Then there's all the other benefits that cloud provides that is inherent in the infrastructure. I don't know about you, but I like the idea of running these complex applications and managing entire businesses from a web browser. I like spinning up virtual environments, deploying development environments and testing upgrades before trying them in a matter of minutes. I like resizing the drives on the applications in a click of a button to make better use of my storage or find that I made a mistake and simply need more. I like not having expensive external SANs to manage and even think about any more. I like not having a ton of spare servers and parts for purpose built servers sitting along the floor and in storage bins in Equinix which is notoriously expensive. I like being able to replicate the environment I built for this customer and relaunch it for someone else in *minutes*, not weeks. I like being able to replicate, test, then upgrade ONE Windows IIS server (decoupled management of the OS and application) and it upgrades all of them on the Cloud automatically. I like being able to effectively use nearly all of the available resources in a Cloud without extensive, time consuming, and limiting architecture of a Dedicated traditional environment. I like the fact it just works. I like the fact it saves my customers money. I could go on and on, but hey that's just why I'm doing it and believe it makes sense for my customers. I'm not anti-dedicated, my business has been dedicated for more than 10x as long as it's been Cloud but I dug deep... very deep, and believe it is better to do it this way because I've yet to find a reason why it makes more sense to not do it this way. To each his own, but again.. that's just me.
Posted by tim2718281, 11-17-2010, 12:59 PM
And how many systems share that throughout?
Posted by KarlZimmer, 11-17-2010, 01:46 PM
CloudWeb, I don't know the applications, but I can't imagine that couldn't be made significantly more efficient using dedicated servers plus some virtualization. In addition, I can't imagine all those separate platforms are truly necessary and I'm certain development work could be done to combine things and redo things to make them simpler, easier to fit on a smaller number of boxes, etc. It seems that is proving my point, that a messy setup can certainly be "simplified" with the cloud, but if it is properly done in the first place, they wouldn't have all those varied requirements in the first place. Again, I don't disagree there is a market for it, it makes sense in cases because sometimes, economically, it simply doesn't make sense to do your own development work,etc. To me, it just then leads us to more inefficiency.
Posted by JasonD10, 11-17-2010, 01:59 PM
And again, efficiency of the application is irrelevant as shown in my argument. We could go there, but it is side tracking the point of Dedicated vs Cloud and in even a simple application that needs the scalability of multiple servers and full high availability and redundancy, you cannot do it better in Dedicated. I just showed you how it is more efficient going to Cloud. The efficiency of the application has no bearing on that and not even remotely correlated. If we cut the needs of this customer in half for then both the dedicated, and cloud requirements are cut in half. So there is no point or correlation. If you go and add virtualization to your Dedicated server environment then your whole argument of inefficiency is nullified as now you're adding that virtualization overhead as well. But, that's still just virtualization, you're missing all the other benefits that Cloud provides. If you're going to go ahead and add virtualization and the overhead of it you should just go all out and go Cloud as otherwise you're still paying the price of lost resources (which I just proved there is none, it's only perceived), and still missing all the other benefits.
Posted by KarlZimmer, 11-17-2010, 02:09 PM
Virtualization overhead will be lower than cloud overhead and would only be used selectively where needed, not across the entire platform. Even then, without virtualization, I'm not sure why the firewalls and load balancers can't be the same boxes, or why the mail servers need to be separate from everything else, why there needs to be separate pools of Linux web servers, etc. My point is you're comparing a non-efficient solution, so of course the cloud will be more efficient... To note, I do agree that the cloud may actually be the best way to do MySQL and/or MSSQL redundancy, but I find that more as a short-falling of MySQL and MSSQL than it is a true benefit of the Cloud.
Posted by JasonD10, 11-17-2010, 02:23 PM
Maybe the mods should split the thread a few pages back from it's original purpose. hehe. That's a bold claim and logically sounds nice and all, but let's see some facts please before making it. I must not be making this clear so let's try it a different way, and just for fun I'm going to scale it way down to something very simple. Let's take a glass and a mug. They're clearly different like cloud and dedicated are. We're talking about the mug, and the glass. But for some reason this conversation keeps talking about whether we're putting water or coffee or soup into them (the applications), and how they could be heated up, moved to the other one, and all this other stuff. All of that is irrelevant and has no correlation to the mug or the cup itself. It's sidetracking and taking away from the point of the conversation which is simply discussing the mug vs the cup (dedicated vs cloud). So back on track, assuming the application is perfect what further arguments can be made from mine above about how Dedicated could possibly work better?
Posted by FHDave, 11-17-2010, 02:34 PM
Karl, what kind of overhead, and how much, are we talking here?
Posted by KarlZimmer, 11-17-2010, 03:58 PM
We're coming at this from two completely different directions, I understand that, and we're not going to get to the same point because of that, but... But the contents DO matter. I'll take it to the extreme. It is possible to build code that itself is fully scalable, utilizes all the resources, etc. and does NOT require all the additional abstraction of a cloud environment. In that case, not using the cloud is certainly more efficient than using the cloud. Also, if code is so simple that it fits on a single account on a single shared hosting box, again, the additional abstraction and overhead of the cloud is less efficient. Where the cloud shines is on poorly engineered and designed code, code that was never designed to scale, or scale well, in the first place, which does include things such as MSSQL. This again then comes down to it being more of an idealistic dislike of the cloud, in that it allows people to be even lazier and to create worse code, while it still works out because of the cloud. Certainly dealing with this sort of inefficient code you can almost certainly achieve greater efficiency overall, as those applications don't understand efficiently using resources and efficiency of redundancy like the cloud does, but as I said, I see that as more of a short-falling of the code than a benefit of the cloud. The vast majority of software used falls under this category though, thus as a whole, the cloud does yes, make things more efficient, which is why we're working on cloud projects of our own. This does not mean though that adding these abstraction layers is the absolute most efficient method, as doing that sort of abstraction directly in the application, so you're only using it where and when it's needed, instead of across an entire platform, is certainly more efficient. If these sorts of cloud infrastructure platforms were really the most efficient thing, then why isn't Google using them? It is because it is more efficient to build the redundancy and scalability into the application. Yes, for 99% of people the cloud can probably make them more efficient, and I applaud it for that, but to say it inherently increases efficiency, it can do something that other things can't as far as efficiency, or to say it is some sort of magic bullet are simply not correct. I feel those sorts of statements and ideas then remove the responsibility from the developer to make efficient and scalable code themselves, which will eventually lead us to much less efficient environments, just as the increase in general hardware specs has led us to extremely bloated and poorly designed control panels, applications, operating systems, etc. Adding layers, adding abstraction, will decrease efficiency by adding more code. Simply logically, how does adding more layers, more abstraction, and more code equate to lower overhead? That is my proof, it logically makes no sense. Care to prove that my dog can't circumnavigate the world faster than you? The only way to prove it is to do it, compare them head to head, (which has not been done in any sort of Cloud vs. Virtualization comparison I've seen) or to simply think it through logically. Do we really need to compare those things head to head to know the answer or will logic get us there? :-)
Posted by JasonD10, 11-17-2010, 04:21 PM
Huh? Now you're talking like a SaaS company but we're all IaaS providers here. Customers are running their applications independent from us getting in their way, even in fully managed hosting environment we do not build or manage their application. If a customer wants to run vBulletin who are we to say that software is flawed and inefficient and cannot scale properly? What if they are using SugarCRM for hundreds of their employees? Or a .NET based frontend for their website? Basically what you're saying is the answer is to use your own proprietary software built to scale, to take away the customers freedom to use the software that has been developed, tried and true, by thousands of independent companies all around the world. Most of which they have happily paid for. And when you're talking that you're no longer an IaaS company, you're providing SaaS. If for some reason you think you are talking IaaS, then frankly you're not being realistic. Businesses don't want to do that, they don't want to reinvent the wheel and they surely don't want their host insulting their intelligence to suggest a ton of improvements when they are not soliciting that advice. In time they probably will and many of our customers do, but it's still a gradual change and at the end of the day there are plenty of excellent commercial platforms available for whatever the application is. Bottom line is they want to run what they have, have independence, and run it well. The Cloud allows this without reinventing their entire business to some proprietary or highly modified "thing" that must now be supported by yourself or whoever is doing this. Who in the world wants to do that? For the vast majority of websites, and Internet businesses they are using someone else's developed software. They are not software developers or architects nor do they employ them. They are building a business on readily available software looking for a platform to host it. Of the hosts on this forum I would bet that well over 90% of the hosts do exactly this. Who here really builds their own control panels and entire business from scratch? I know some do and that is absolutely fantastic.. but we can't deny the facts and majority rules. We do have some SaaS providers as customers and they do more what you're talking about, but as an IaaS provider of 12 years by far the majority of our customers are meeting their needs by using commercial software and what they want is a way to achieve redundancy, high availability, scalability, ease of management, save money, lessen dependence on purpose built hardware, costly physical engineers at the Data Center, decouple management from the OS and application WITHOUT modifying these ready-built applications that they already have owned and purchased. Boy I sound like a salesman and I truly don't mean to it's just the facts. But it can all be done on Cloud and a lot more as we've been talking about. And finally, again, I've already shown you how cloud is more efficient as logic is a perceived idea of efficiency. Just the numbers alone let alone the Cloud "software" benefits come out ahead unless you're talking about a custom designed SaaS like Google which we're not.
Posted by JasonD10, 11-17-2010, 04:22 PM
Since the question hasn't been answered, I will answer it for you. The answer is 1 thread of 1 cpu core per cloud, 2.5GB of space per ha instance, and 3/4GB of RAM. That's all it needs to add for us, I can't speak for other technologies.
Posted by uberhostNET, 11-17-2010, 04:40 PM
The lie of "unlimited storage & bandwidth!" is a bit less evil when referring to cloud vs vps. Cheers, Rick
Posted by gocloud, 11-21-2010, 08:07 PM
Hi Brent, Do you still use SATA (3.5" and how many?) drives in your RAID0 or RAID10 node (storage) setup with Applogic in combination with 40Gbps IB?
Posted by brentpresley, 11-21-2010, 09:13 PM
Infinniband is currently in testing w/ 3Tera/CA. We have not placed it into a production environment. With the servers we have laid out, we can RAID 0 up to 6 drives. We do not utilize RAID 10, as it would be wasted spaced since the 3Tera grid has similar built-in redundancy on the node level.
Posted by KarlZimmer, 11-21-2010, 09:13 PM
CloudWeb, you're not arguing against what I'm saying, at least that I can tell. I admit, my point doesn't really apply to the overall thread, but I had already clarified that. No matter what kind of provider you are, shouldn't you be providing your customer with the best solution and best advice possible? Whether you're an IaaS or SaaS provider, it doesn't matter, you should still be looking for what is best for your customer. Now, I also have stated on multiple occasions, that it is certainly possible that developing a truly scalable platform/application is not economically efficient, and I acknowledge most customers can't afford to re-invent the wheel. The issue there is though, that it seems many customers don't even consider efficiency when selecting existing software platforms, and most existing commercial applications scale terribly. Yes, the cloud helps make those applications scalable, but why do people continue to buy applications that aren't scalable, or try to force that sort of scalability? As I said, I agree that the cloud can certainly make current applications more efficient as far as scaling them out, adding redundancy, etc. but I also don't see why companies would continue to just buy old non-scalable apps and live with the status-quo, if what they want is scalable solutions. If you want the cloud, take a look at yourself, the applications you've picked to this point, and see what you can do to make yourself more scalable first, then look at cloud solutions.
Posted by gocloud, 11-21-2010, 09:23 PM
Brent, Is your RAID0 with 6 drives limited (to 6) because of the hardware you use or something else and do you use SATA drives (7K rpm) or SAS (15K rpm)with or without SAS HBA?
Posted by brentpresley, 11-22-2010, 07:43 AM
Right now, any RAID setup we use is limited just by the hardware choices we have made (chassis, RAID cards, etc.). If we really found that HDD I/O was limiting our customers overall experience, we could go with chassis that accommodate more drives and use a RAID card with more ports to alleviate that problem. We are currently utilizing 7200RPM drives b/c of their cost/GB. We get very good performance from this setup, and it allows us to offer larger storage to our customers than if we were on faster drives. We are looking into SSDs, as we know another 3Tera AppLogic provider is running some very serious tests with those, but until their data is back and we have a better idea of reliability, we have held off on deployment. Last time we looked at them, about a year ago, 3Tera told us that their in-house testing demonstrated poor reliability overall (on SLC-based drives). For that reason we have held off on deployment.
Posted by gocloud, 11-22-2010, 08:27 AM
Brent, Your RAID cards are HBA's or proper cached RAID controllers? But by reading Applogic release notes they clearly say 'AppLogic has not been certified with any hardware RAID solutions ...' and as such not certified by CA/3Tera', does it mean that you don't have their full support on this? I guess it will be a problem to run heavily loaded mysql database from a single sata drive in the server/node with Applogic, and because of that youv gone RAID0 route...
Posted by brentpresley, 11-22-2010, 08:37 AM
HBAs. You get minimal performance improvement going from HBA to a cached RAID controller if all you want is RAID 0. It's when you need to write a parity stripe (RAID 5/6) that a cached controller with a good onboard CPU shines. And I know what the 3Tera docs say. You will notice that they don't mention Infinniband either. BH has a very very close relationship with 3Tera. The two companies leverage this to test and deploy technologies prior to their incorporation into the official releases that 3Tera distributes to their other partners. For example, we have been running RAID 0 on all our nodes for about 9 months now. The documentation just needs to be updated to reflect the current state of development. I cannot comment on 3Tera's willingness to support RAID 0 with other providers, but since we were the first to implement it and provide them with the results of our testing, they have had no issue answering any questions about it we put to them.
Posted by JasonD10, 11-22-2010, 10:23 AM
I apologize if I'm being a little dense.. but maybe you're not understanding the difference of IaaS to SaaS properly. Just to make sure this is clear by looking at your website, or our website, we're IaaS providers. SaaS is something like Google Apps, or Salesforce. It is a specific application that is sold as license based software. SaaS providers purchase services from IaaS providers, or they build their own Infrastructure. I completely agree in providing the best possible service and advice, and believe this is how to do it as we've already talked about. Not all commercial platforms are bad. We all use some yourself included. Some things are more important than efficiency such as support, security, price, and reliability. All things must be considered and while efficiency is of course important, so is everything else when making the decision on what platform to use. And again, besides scalability you're still missing all of the other benefits of Cloud which there's no point in rehashing. It provides a whole lot more than just that.
Posted by gocloud, 11-22-2010, 12:10 PM
Hi Brent, Running RAID0 with SATA drives hmm, have you had any failures in 9 months? I mean this is not the most reliable approach to storage I guess I believe I read somewhere on forums that they 3Tera have some kind of 'patch' (not officially supported to run RAID0 on Applogic nodes as obviously software RAID solution (possible using LVM), you are obviously not using it? Yes, Infiniband is not mentioned but again in some of the forum posts Applogic said that few of their customers utilize it successfully (with their support!?) gaining quite high overall performance
Posted by brentpresley, 11-22-2010, 12:18 PM
Any hosting provider will have drive failures. We have our share. We have never had lost data from it.
Posted by KarlZimmer, 11-22-2010, 01:40 PM
No, I understand the difference. I don't see how that matters though, the objective is always the same, find your customer the best solution, whether it is the best solution for your company or not. if they can get what they want simpler and cheaper from a single dedicated server if they just changed their application, but you only offer cloud services, you still tell the customer what is best. I know that properly scalable applications don't exist much, but that is what upsets me. IaaS shouldn't really be necessary if people knew how to design and engineer software/applications. That it shouldn't be necessary doesn't mean it isn't a good business plan or business option, because what would make it unnecessary doesn't exist... Maybe that helps clarify? I'm upset at software development, which is driven be people's demands. I don't understand how IaaS, a much more complicated solution, became the better option than simply having better engineer applications, but it has...
Posted by JasonD10, 11-22-2010, 01:58 PM
Of course it matters. If you are working for yellow cab and driving people around all day or are working for Ford and making the cars that the taxi companies use it is a very different thing, just as IaaS is to SaaS. SaaS companies use IaaS as they do not wish to build these services in house. Just like it was with Shared Hosting, Co-Location, Dedicated Servers, and Fully Managed Hosting. All of those services are IaaS, Cloud is just the next advancement of technology and a new method to provide IaaS services. Not all companies want to do all of that in house so they outsource it. If it provides value to their business then it is a good thing. I have had many non-managed customers turn into managed customers through the years because they realized the value of outsourcing. More and more I find businesses are becoming smarter and focusing on what they do best, which is focus on their business and not so much on every detail about it and what makes it tick. Just like you and I instead of handling every single thing about our business we have other people in the organization who handles their jobs. With Cloud services starting cheaper than a Dedicated Server I don't see the point of offering Dedicated anymore worrying about dealing with failing hardware and expensive Data Center technicians, having so many types of spares, and all the other stuff that goes along with it. After doing it 12 years.. we are in the process of dropping it entirely but that's just me. As I said earlier I'm not against Dedicated and it has it's place I guess, but it's a ever growing smaller and we're willing to lose that market share in the process. If a customer isn't a good fit, we tell them. But that is not all that often as Cloud services evolve into more and more levels of services as we'll soon publish...
Posted by KarlZimmer, 11-22-2010, 04:23 PM
So you're saying the car company should be worried about the efficiency of the cars they make, but the taxi company shouldn't care about the features and efficiency of the cars they buy?
Posted by gocloud, 11-22-2010, 05:16 PM
brent, but are you not increasing your chances of losing the data when running raid0? even if you have applogic IPSAN (and 1:1 mirroring across the nodes unless you use more then 1 mirror copy)
Posted by brentpresley, 11-22-2010, 06:32 PM
Nope. No chance of data loss.
Posted by streaky81, 12-07-2010, 10:20 PM
Clouds don't have virtualization and limited resources, no matter how you define them - you shouldn't be able to define them - VPSs do. People think VPS deployment platforms are Clouds - they're not, they're systems for deploying VPSs. Which is fine, I'd hate to argue against the usefulness of what they are, and I wouldn't doubt the scalability or stability of a well-built one. Just think the world would be a better place if we called things what they were - then people would know what they're buying. If you can define something's size or scope - or it is resource limited in any way it isn't a cloud, basically. Feel free to argue otherwise, but I'm right and you're not
Posted by cAPTAIN^k, 12-08-2010, 12:07 AM
Amusingly the term "VPS" is almost non-existent in many countries outside USA. The rest of us call them VM's Last edited by cAPTAIN^k; 12-08-2010 at 12:11 AM.
Posted by Anykey, 12-08-2010, 11:23 PM
For the end-user, what would be the difference between VPS and Cloud, using Virpus.com as an example: http://virpus.com/unmanagedvps/ 200GB hdd / 1024MB ram http://virpus.com/cloud/ 40GB hdd / 768MB ram Would cloud users have same access to root console and starting/stopping/installing of linux apps as VPS user does? Also, what other differences would the end user notice in normal day to day opperations? Maybe more consistency of performance in a cloud than in a VPS which fluctuates depending if some users are hogging disk I/O etc?
Posted by axces, 12-13-2010, 06:15 AM
Even if "cloud computing" really did work _perfectly_ it would never be more efficient than to just use one computer and not that many people need the resources of more then one computer. It would be more efficient for someone to buy/rent a dual hexa core processor (or other bleeding edge fast server) then to use "the cloud". This is even if "the cloud" worked perfectly and the technology to enable the most perfectly ideal cloud were freely available. So in what situations are people in when even the current bleeding edge fast server (single server) is not fast enough? If someone is really in that situation they should at least have a programmer/admin that doesn't suck so much that he can't even make use of two computers. It's just not that hard to use resources of two or more computers for your application and the people that really need to do it wouldn't have a problem doing it.
Posted by streaky81, 12-13-2010, 07:45 AM
No because that's the problem: If you have the need for a server more powerful and your 'cloud' is really just some kind of virtualized environment you're limited by the power of one HV unless you start doing things you might as well do a cluster for in the first place - then you have options for uptime and scalability anyways. "vps clouds" are only really useful in situations where you don't need much in the way of resources - and that's in an ideal world where hardware doesn't fail, but bearing in mind it does - you may as well get a redundant dedi - it'll be cheaper. This idea that you can't consolidate without virtualization is getting a bit silly
Posted by boskone, 12-14-2010, 12:14 PM
In a short amount of time, all services will be delivered using cloud technology. Dedicated servers and collocation will become a very small, higher end niche, for companies and large corporates that require privacy above all else, or who have their own very extensive requirements - but they will also typically just run private clouds on that kit.
Posted by uberhostNET, 12-15-2010, 07:40 AM
Resistance is futile? Perfect! Cheers, Rick
Posted by xtrac568, 12-15-2010, 08:08 AM
that's cool, but first we need to define "cloud technology", like: -it's alien technology 4D8W1 -it's stealth hosting H117 -it's complex virtualization on demand ... so on. until then, the term is pretty useless.
Posted by boskone, 12-15-2010, 08:12 AM
By cloud tech, in a hosters perspective, I mean - large scale compute nodes, with resilient, high capacity interconnectivity - all primary storage off-node with resilient, performant SANs - running virtualisation on the compute nodes to deliver service - in time, running virtualisation on the SANs All delivered in a flexible, scalable, elastic, web and api facing service delivery model to the end user.
Posted by streaky81, 12-15-2010, 09:06 AM
And by all services you mean anybody who needs decent IO without epic latency and contention, and massive virtualization overheads in all its different types. Scalable virtualisation is a myth. Last edited by streaky81; 12-15-2010 at 09:11 AM.
Posted by boskone, 12-15-2010, 09:13 AM
What's mythical about the ability to scale virtual resources? Decent IO with low latency and five nines resilience is already being provider by cloud providers.
Posted by streaky81, 12-15-2010, 09:19 AM
You're limited by the grunt of one HV, unless you get into clustering at which point you may as well do it on bare metal and avoid all the overheads, oh and - instant redundancy. lol we're confusing solid performant IO with remote storage rather than what people are used to with uncontended direct-connect storage (sata/sas/et al), with serious latency/bandwidth issues. Once you ask the right question it all starts to fall appart...
Posted by boskone, 12-15-2010, 09:24 AM
SAN IO typically outperforms local storage (at least in my experience)
Posted by HostColor, 12-15-2010, 09:47 AM
Well, excellent discussion I enjoyed. Thanky folks for having it. May I just say that the opening thread was about difference between VPS and Cloud. As we all know Cloud refers to a Cloud computing technology which might or might not have anything to do with VPS and virtual machines.
Posted by streaky81, 12-15-2010, 10:48 AM
Well hey a SAN in it's own right is faster than a sata disk in a dedi no doubt. The question comes when you start loading a lot of other users on it and it starts grinding to a halt, which you know - doesn't take all that much - is it worth the investment or is there some better option available? That's also ignoring the fact that it's a single point of failure for everybody sharing that SAN, which is big danger because they do fail often, especially if you happen to be thrashing them.
Posted by boskone, 12-15-2010, 10:55 AM
It all depends on what people call a SAN ? Some cloud providers are using SATA disk in server chassis with large raid cards and calling it san. My experience has been with proper netApp clusters. They rock local storage every single time and scale almost without limits relative to user requirements. They are almost nearly impossible to 'thrash' given their capabilities. If you use a cloud that is 'on the cheap' you will suffer the consequences. A 'real' SAN is designed to 'never' fail (within normal tolerances)
Posted by streaky81, 12-15-2010, 10:59 AM
Well hey of course that's true but by which point you have to start asking questions about if it's really cost effective. You could also buy the cray XT5 and run your blog on it, and it'll definately stay up if you get dugg.
Posted by boskone, 12-15-2010, 11:10 AM
Cost effective works for both the customer and the provider. Spending money on the things that matter typically saves the hoster money. Look at dediserve.com as an example. 19.95 a month for a virtual machine, and running on netApp SAN and HP high end server nodes, with onapp cloud engine.
Posted by JasonD10, 12-15-2010, 11:20 AM
Well.. that's euro's and it's always important to consider all factors of a host. Now I don't mean any of this towards Dediserve just speaking about consumers in general. Dediserve may have a spectacular business. More often than not people talk about the things you mention (hardware, what virtualization backend or cloud engine, storage space, and sometimes SLAs which can be worthless..). Unfortunately customers rarely ask other questions such as core infrastructure, networking, Data Center, staffing, resiliency in the environments, upgrade policy, etc etc. So much makes up a company and unfortunately so much is hidden from customers as they're not walking into storefronts like a traditional brick and mortar. A provider could have a fantastic website with excellent information and do a great job at marketing, but produce a horrible service in the long run (GoDaddy hosting anyone?). More and more consumers are learning.. but it's still a long way away.
Posted by streaky81, 12-15-2010, 11:30 AM
No but this is the problem right here. Not much in the way of resources for that.. They don't do a vps as powerful/as much resources as my dedis, even though they're orders of magnitude more expensive.. Just leads me to ask the cost effective question again, especially given they're probably not even paying for onapp.
Posted by JasonD10, 12-15-2010, 11:36 AM
That's another issue comparing dedicated servers to cloud. Cloud obviously costs more, and it has a list of benefits above dedicated. Also something to consider is the other costs of business which mentioned above. For example, we host out of Equinix which is notoriously expensive. We routinely get phone calls from Cogent, HE, and other "cheap" Data Center's willing to take our business for exactly 1/4 of the cost we pay for all of our facilities, floor space, and peering in Equinix. That is our #1 cost in business but why we choose to pay 4x as much because it's worth it. Comparing apples to apples with providers is difficult to do, and for some customers they will just flat out find a better fit with a different "grade" provider. If cost effectiveness is the #1 thing for a customer, Equinix and DuPont Fabros are not going to be their choice when you can easily spend over $2,000/mo just for a rack and power, no ping. Fortune500's, high-profile websites, and those companies who are very serious about their hosting that pose to lose large revenues for sub-par quality have no problem paying higher prices because it provides value to their business and justifies that expense.
Posted by streaky81, 12-15-2010, 11:46 AM
I'm comparing apples to apples. Not for nothing but it's a bit too easy for the 'cloud' industry to make claims that just aren't true, and that's one of them - I'm fully aware how much a rack costs in a quality DC. Eq is 100% cloud providers? No of course it isn't..
Posted by KarlZimmer, 12-15-2010, 11:58 AM
What high profile web sites are hosted out of Equinix? Most high profile sites have the space and power needs to find wholsesale space from providers like DRT, not retail from Equinix. Most of the sites Equinix lists as customers don't have significant amounts of space there, simply POPs for connectivity purposes or a part of a CDN, etc. The reason to go with Equinix is the connectivity, that is it, there are plenty of other facilities that build out data centers to the same levels of reliability at a lower price.
Posted by JasonD10, 12-15-2010, 12:17 PM
I'm not at liberty to say who they are as they may not want that known. What I know is here in the DC area where we're at there's now.. hrm.. 6 I think Equinix Data Center's alone within about a block's radius, there are an awful lot of high profile sites, SaaS companies, CDN providers, newsgroup providers, core DNS services, banks, government customers, Fortune500's, and other very critical online businesses. I host some of those myself on our main business (Cloud Web is just a subsidiary). Some choose it for the excellent connectivity, but some go "pay grades" higher and want the other things it provides. From what I've seen through the years some companies which have outgrown Equinix still have quite a presence in Equinix but also have built their own Data Center's. There's kind of a rule of thumb among the carrier-neutral crowd and it's either you host in carrier neutral such as EQIX/DFT or you host yourself.
Posted by KarlZimmer, 12-15-2010, 03:12 PM
That was exactly my point, those companies USE Equinix facilities, but they're not based out of Equinix facilities. Most sites or companies that get large enough will be getting wholesale space, not retail space. Yes, a business unit of a Fortune 500 company will use an Equinix facility for a specific use, but to basically indicate they're largely dependent on such facilities for their data center needs isn't really true. They're used for inter-connectivity, not hosting core infrastructure. I'm not saying no one uses it for that, but they'll claim Limelight, Akamai, Google, Yahoo!, Rackspace, PayPal, EA, various banks, etc. as customers when those companies use Equinix for an extremely small percent of their overall data center needs. What are the "other things" Equinix provides?? Note: We are an Equinix customer in multiple sites. We would be listed as an Equinix customer, but Equinix makes up ~0.3% of our data center space.
Posted by JasonD10, 12-15-2010, 03:18 PM
That also is not true. I know for a fact many of these companies are running their core infrastructure out of it. I get what you're saying, and I'm sure many build out their own infrastructure's but Equinix is about as close of an infrastructure between retail and wholesale as you say (you buy cages, buy power in bulk, build them up as you see fit, buy bandwidth and cross connects as you see fit). I don't know, maybe you're talking about small customers who come in and buy a single rack in increments like a traditional Data Center but I know the larger customers don't. Everyone has private cages and decide exactly how to build that out. *shrug*
Posted by uberhostNET, 12-16-2010, 11:46 AM
In that case it is probably best not to use them as an example in the first place. Cheers, Rick
Posted by JasonD10, 12-16-2010, 11:47 AM
It's irrelevant to the fact that it's done.
Posted by boskone, 12-17-2010, 05:05 PM
CLoud is a lot more than 'vps' It's a new way of delivering solutions and yes, they can be a lot more powerful than equivalent 'enterprise' dedicated servers, The key with enterprise cloud offerings if that you have to really compare 'like with like'. If you wanted a dedicated solution to deliver what a cloud machine can, ok: 1 - set up both dedicated and cloud machines with the same spec 2 - for dedicated, add a second identical machine for failover 3 - for dedicated now buy switches and failover gear/load balancer 4 - now sort out a NAS or backup storage for the dedi 5 - now have an SLA with your DC to insert ram or storage in your dedi in under 15 minutes, 24/7. Now, what does the dedicated server solution cost? Apples with apples here, cloud wins every time. Providers get that, which is why everyone is scrambling to go cloud. Dedicated servers of the mid 2000's and that era is coming to an end.
Posted by streaky81, 12-17-2010, 09:57 PM
Basically everything you said is unnecessary with 15 seconds of capacity planning - if you really need ram in 15 minutes you're doing something wrong.. Switches on dedis what? Load balancing isn't a standard cloud in a vm-sense feature, you have to buy more clouds and set it up just like with a dedi. Providers are scrambling because it's the fashion, let’s get a grip on reality a bit here..
Posted by nonprofit-worker, 12-17-2010, 11:15 PM
I am tired of being a nonprofit worker so if there's some way I can sell some kind of mumbo jumbo for more money (the cloud) then I'm all for it. As I see it the best thing about "the cloud" is that it gets clients to quit asking questions because all the answers seem to be the same. Client questions are: What kinda raaaam? What kinda processor? What kinda harddrive? And if we can get the client to just stop being annoying as hell by saying "Unlimited!, Secure!, NEVER go down again!, It will help you build anything you want even tho you can't program!" Then I can answer every single clients questions with the same answer. It's a lot less work and gets the client to just stop asking questions. This seems to be the real advantage of the cloud and that's okay because it's all about the money sometimes. Just don't try to answer technical questions about the cloud with marketing propaganda.
Posted by JasonD10, 12-18-2010, 10:05 AM
Not really. Why overbuy the Dedicated server keeping spare resources on hand? It's more cost effective and easier to upgrade in Cloud when you want. Actually switches, load balancers, and various other plug and play appliances can and do happen with private ptp tunnel's between those vm appliances in Cloud. It's way faster, easier, and cheaper than traditional hardware as.. well.. it's virtual. It's not like a dedicated.
Posted by JasonD10, 12-18-2010, 10:09 AM
The technical side is that it is far easier, cheaper, and quicker to develop, deploy, and manage an infrastructure. Time to market is cut in half, and setting up new development environments and taking them to production that time is greatly reduced. It's nice when the technical and marketing sides align and it actually does IF you have a proper Cloud infrastructure.
Posted by axces, 12-19-2010, 12:18 AM
You're going to have to do more 'CloudWeb' then getting the last post in this thread to say some meaningless blanket statement of "Cloud is awesome!" that will only convince someone that is already clueless. You're going to have to get involved in some technical detail and provide some real numbers to convince anyone that has real experience. Also why didn't you reply to my post?
Posted by nonprofit-worker, 12-19-2010, 04:37 AM
You destroy the whole point of RAM in a computer if youre sending data across the network and through the bus of another computer. If you do this you might as well take the data youre supposed to send to RAM and just store it on SSD/hard drive or other slower media. Cloud services have created a specialized high speed zone in the cloud just to run mysql. Other clouds have just created a separate section of the cloud just for mysql servers that are not actually apart of the cloud. This should tell you something about the inefficiencies of trying to fork a process across a network to another computer. (Dont think most clouds can do this any way) The people that try to run mysql in the cloud obviously could just use mysql replication like everyone already does when they need to scale. Tell a company that is operating a rendering farm or a standard computing cluster (or super computer) to start using onApp or whatever and it will not go over well because they will be able to see a instant and obvious decline in efficiency. I know many companies that use render farms and they would never try to use the cloud. Isnt that odd? Some of the people that use clustered computing the most dont do the cloud. Please take time to address everything I said without "the cloud" marketing nonsense and unspecific generalities
Posted by eming, 12-19-2010, 05:58 AM
If you have 100% 24/7 utilization of your compute resources, and if you are content without redundancy/HA or a sexy UI then yes - it would be silly to deploy on OnApp. However, chances are that your load is not 100% 24/7, but perhaps an avg. of 50-60% or lower on avg. In that case it might make sense for you to have a dedicated server farm allocated to dealing with the constant load, and offload the spikes to a public cloud. It's one of the core business areas of Amazon (rendering, transcoding, numberchrunching etc), and OnApp has a handful of clients with setups specialized for this sort of thing. No, it's not odd at all. There would be no need for it in that scenario. If you have a constant load, no need for redundancy or a publicly shared infrastructure - then why bother doing the whole cloud thing. Wouldn't make sense. However, if you have specific rendering jobs to do it might make sense for you to have them completed in some sort of public cloud scenario, only paying for the resources consumed. Outsourcing compute power ... Is that unique for the cloud though? No - not at all. But it's just been conceptualised and commoditised - making it a whole lot easier to buy and sell. D
Posted by Mo-chi, 12-19-2010, 07:18 AM
Cloud has a big failover, your cloud will never down. When the first node down, the cloud will move automaticly to another nodes.
Posted by Cristi4n, 12-19-2010, 08:57 AM
Cloud or everything that should allow you to call your solution a cloud solution has it's benefits. I have clients that do not want to pay for multiple dedicated servers except a few hours a month. I also have to agree with some anti-cloud people here too, although they tend to over exaggerate sometimes. Usually companies that sell cloud solutions tell you that everything will fly around and you will never have a problem again, like Mo-chi, above, sees it and which of course is a big lie. Last edited by Cristi4n; 12-19-2010 at 09:02 AM.
Posted by nonprofit-worker, 12-19-2010, 09:43 AM
I only used a render farm as an example. Let's say 100% for 1 hour a day and 40 percent for the rest of each day. It's not silly to deploy OnApp anymore? How so? You're right it's not unique to cloud computing. You could sell a VPS to someone for a few hours if you want. You could strike that as a reason for cloud being easier to buy/sell. Could replace it with a number of other things tho such as.. Cloud computing is easy to buy because: 1. I'm totally confused about how a computer works 2. I like the idea of slowing down my application for redundancy even tho I can have redundancy and normal performance if I knew what I were doing Cloud computing is easy to sell because: 1. There's people that are too confused to prove that this a bad idea and it's not more efficient (exclude all things not unique to the cloud) 2. I have a few answers that all similar to each other. I barely have to read my sales emails, I can use auto-reply and it saves me a bit of time
Posted by Cristi4n, 12-19-2010, 10:20 AM
@nonprofit-worker too few people know how to manually configure/deploy a.s.o. virtual or dedicated servers. And then there are others that just don't have the time to configure and deploy manually everything, they don't care how much it costs they just need a solution asap. Take netapp for example, there are too many companies using netapp as a SAN. Some do not have the necessary time to care about a performance loss if there is any and do not have the time to start learning how a computer works since it's not their job. You can't expect for everyone to know everything. As stated above, there are cases when having a "cloud like" solution is ok and you should know that people will always go with the trend regardless if that trend is ok or not for them. OnApp from what I know was meant for service providers or at least for companies that need to automatically deploy multiple VMs pretty fast for third parties/clients, and was not meant for a single person. Anyway, I think there is too much spam with OnApp already here.
Posted by eming, 12-19-2010, 11:10 AM
Agree wasn't my point, but - depending on your specific setup - it may make sense to have a set allocation of resources that would deal with the 40% load, and then spike into a public cloud offering outside of your facility/infrastructure, like Amazon. True, you could replace the typical cloud offering with a whole bunch of other things, my point before (when mentioning that it has now been conceptualised) is that no-one (in the mass market) really sell's VPS's that way. Go to the VPS category in WHT and look for hourly offerings...I havent checked lately, but I doubt you'll find many. So doable? Sure, in theory - done? No, not really in practise. The whole "cloud" buzzword is not helping any of us, and I wish it would go away...and actually, I think in time it will. In the not too distant future it's not a buzzword, its just how stuff is done. D