Converge Infrastructure Appliances - Hype or Transforming
Normally I write about home computing, mobile and home automation subjects. But in real life I am a pre-sales architect for the Enterprise Solutions Group @ Dell in the Healthcare vertical.
That means I spend my days developing solutions involving hundreds if not thousands of servers, petabytes of storage and all the networking that makes it all work.
With the advent of server virtualization, many of the traditional approaches to compute, network and storage solutions are being challenged. One of those challenges is the concept of converged infrastructure.
Converged Infrastructure is the approach of taking two or more pieces of compute, network or storage components and integrating them in some fashion to create a single manageable entity.
In recent years companies like VCE - the joint partnership between Cisco, EMC and VMware - have created rack sized product offerings that attempt to create that single managed entity. VCE's was called a Vblock. NetApp's was called a FlexPod and so on.
The goal of these products was to create a predictable procurement and deployment model. So size a "block" of compute, network and storage and that would be your minimum purchase that would run X number of VMs. When you reached X, you would buy another block. Simple right ? Wrong...
The challenge with these offerings is they are actually just a bundle of individual pieces/parts from the contributing companies and some attempt at hardware orchestration and virtual machine and resource management. It's not that they didn't work - but rather that they were typically very expensive and the orchestration/management tools were patchworks and not very good. Plus scaling at the rack level is not very efficient and mixing matching vBlocks just doesn't work.
Having worked for EMC when these offerings were being developed - the sales folks looked at any business application issue as a nail and offerings like Vblock as the hammer. Why ? Because they were big $$ sales and helped them retire quota against 3 solutions - compute, networking and storage all at once.
While this was going on in the Corporate IT market, another revolution was occurring and that was with the Google's, Facebook's, etc in this world. Folks like Google consume compute, network and storage resources at a rate that is unprecedented. To make their business models work, they need really cheap compute, network and storage resources and their apps are custom written to scale at huge numbers.
When you look at a Google Datacenter for example, they don't use traditional rack or blade servers. You won't find HP or IBM or Dell in a Google DC.
They have very densely packed server "trays" or "sleds" with a minimalist design, often with no chassis and none of the added extras that come with a modern server - so no remote management boards, no hot plug drives, no redundant power supplies, etc, etc. In fact Google works with what are called "Original Design Manufacturers" - ODMs to custom build motherboards, power supplies, etc to their specification.
Because their customer software provides all the scalability and availability, if a unit fails the just unplug it and slide in a new one and it rebuilds automatically. Google utilizes like 1 admin per 10,000 servers. Something Corp IT can only just look at with amazement.
And to give you an idea of the scale I am talking about - it is rumored that Google is Intel's single largest server processor customer. This approach to large scale datacenter computing has been so disruptive that Dell has created a whole group called Dell Custom Solutions (DCS) that is effectively an ODM. The DCS has created similar solutions for customers like Ebay and Yahoo..
Now to the storage side - you won't find huge storage arrays from EMC or HP at Google either. They would much rather buy 10,000 servers with some locally attached drives, with no RAID and then apply a software solution to provide them with high availability of data. The Google File System is that software and nothing comes close to supporting it's scale.
These approaches to extreme scalability and availability have been coined with the term Web Scale.
In the last few years some former Google folks are spinning up new start-ups that take the WebScale concept and turn it into a consumable appliance for Corporate IT. These appliances typically combine the compute and storage components of a virtualized datacenter along with specialized software to manage the storage.
When compared to the Vblock approach at the rack level, these solution are typically a fraction of the cost per unit, have more sophisticated orchestration and management tools and scale to levels most Corp IT will never see and no Vblock could ever come close to..
Solutions such as Nutanix, Sanplivity and VMware's EVO:Rail are a part of this wave of such appliances. Dell has recently signed partnerships with Nutanix and VMware to produce these appliances.
The question is - are these just hype or are they going to transform the CorpIT landscape ?
My personal belief is the later. For the better part of a decade now, CorpIT has been facing ever shrinking budgets and resources while still be asked to support larger and larger application portfolios and faster go to market deadlines.
They are also competing versus cloud providers such as AWS, Azure, etc. Companies would prefer not to capitalize their IT infrastructure investments or if they do, they want a predictable financial model.
The challenge in the past was as you scaled some applications, the infrastructure cost and complexity did not scale linearly at the same rate. These appliance attempt to address that.
Virtual Desktop Infrastructure (VDI) is one of those apps that exemplify the appliance approach. With VDI projects - customers often start with a small 100 - 250 user pilot on some existing servers and storage and then need to scale it to support thousands of users. Using traditional compute and SAN infrastructure the scale up can get extremely expensive - sometimes to the point of either A) killing the VDI project entirely or B) reducing the ROI to the point where the savings the project was designed to achieve are completely consumed and considered a failure.
Using the appliance approach - you could for buy a single appliance for your pilot. Then add more appliances as you scale up with each appliance supporting x users and scale linearly all the way up to your goal with increased performance and availability built in. No extra stuff needed. Oh and by the way - much less expensively than a traditional model. Additionally the support model is brain-dead simple. No multiple vendor calls - no finger pointing.
Converged appliance don't address every Corp IT need out there today - but they can address several and with the predictable procurement cost model may actually allow more projects to get approved. Of the solutions out there today I believe Nutanix has the best overall story, product line up and value.
Bottom line - Converged Infrastructure Appliances are very real, very good and very disruptive. They take the proven approaches that the largest providers in the world utilize and put them in your hands. If you haven't brought them in to your DC to at least test - do so. It may just save your job...
That means I spend my days developing solutions involving hundreds if not thousands of servers, petabytes of storage and all the networking that makes it all work.
With the advent of server virtualization, many of the traditional approaches to compute, network and storage solutions are being challenged. One of those challenges is the concept of converged infrastructure.
Converged Infrastructure is the approach of taking two or more pieces of compute, network or storage components and integrating them in some fashion to create a single manageable entity.
In recent years companies like VCE - the joint partnership between Cisco, EMC and VMware - have created rack sized product offerings that attempt to create that single managed entity. VCE's was called a Vblock. NetApp's was called a FlexPod and so on.
The goal of these products was to create a predictable procurement and deployment model. So size a "block" of compute, network and storage and that would be your minimum purchase that would run X number of VMs. When you reached X, you would buy another block. Simple right ? Wrong...
The challenge with these offerings is they are actually just a bundle of individual pieces/parts from the contributing companies and some attempt at hardware orchestration and virtual machine and resource management. It's not that they didn't work - but rather that they were typically very expensive and the orchestration/management tools were patchworks and not very good. Plus scaling at the rack level is not very efficient and mixing matching vBlocks just doesn't work.
Having worked for EMC when these offerings were being developed - the sales folks looked at any business application issue as a nail and offerings like Vblock as the hammer. Why ? Because they were big $$ sales and helped them retire quota against 3 solutions - compute, networking and storage all at once.
While this was going on in the Corporate IT market, another revolution was occurring and that was with the Google's, Facebook's, etc in this world. Folks like Google consume compute, network and storage resources at a rate that is unprecedented. To make their business models work, they need really cheap compute, network and storage resources and their apps are custom written to scale at huge numbers.
When you look at a Google Datacenter for example, they don't use traditional rack or blade servers. You won't find HP or IBM or Dell in a Google DC.
They have very densely packed server "trays" or "sleds" with a minimalist design, often with no chassis and none of the added extras that come with a modern server - so no remote management boards, no hot plug drives, no redundant power supplies, etc, etc. In fact Google works with what are called "Original Design Manufacturers" - ODMs to custom build motherboards, power supplies, etc to their specification.
Because their customer software provides all the scalability and availability, if a unit fails the just unplug it and slide in a new one and it rebuilds automatically. Google utilizes like 1 admin per 10,000 servers. Something Corp IT can only just look at with amazement.
And to give you an idea of the scale I am talking about - it is rumored that Google is Intel's single largest server processor customer. This approach to large scale datacenter computing has been so disruptive that Dell has created a whole group called Dell Custom Solutions (DCS) that is effectively an ODM. The DCS has created similar solutions for customers like Ebay and Yahoo..
Now to the storage side - you won't find huge storage arrays from EMC or HP at Google either. They would much rather buy 10,000 servers with some locally attached drives, with no RAID and then apply a software solution to provide them with high availability of data. The Google File System is that software and nothing comes close to supporting it's scale.
These approaches to extreme scalability and availability have been coined with the term Web Scale.
In the last few years some former Google folks are spinning up new start-ups that take the WebScale concept and turn it into a consumable appliance for Corporate IT. These appliances typically combine the compute and storage components of a virtualized datacenter along with specialized software to manage the storage.
When compared to the Vblock approach at the rack level, these solution are typically a fraction of the cost per unit, have more sophisticated orchestration and management tools and scale to levels most Corp IT will never see and no Vblock could ever come close to..
Solutions such as Nutanix, Sanplivity and VMware's EVO:Rail are a part of this wave of such appliances. Dell has recently signed partnerships with Nutanix and VMware to produce these appliances.
The question is - are these just hype or are they going to transform the CorpIT landscape ?
My personal belief is the later. For the better part of a decade now, CorpIT has been facing ever shrinking budgets and resources while still be asked to support larger and larger application portfolios and faster go to market deadlines.
They are also competing versus cloud providers such as AWS, Azure, etc. Companies would prefer not to capitalize their IT infrastructure investments or if they do, they want a predictable financial model.
The challenge in the past was as you scaled some applications, the infrastructure cost and complexity did not scale linearly at the same rate. These appliance attempt to address that.
Virtual Desktop Infrastructure (VDI) is one of those apps that exemplify the appliance approach. With VDI projects - customers often start with a small 100 - 250 user pilot on some existing servers and storage and then need to scale it to support thousands of users. Using traditional compute and SAN infrastructure the scale up can get extremely expensive - sometimes to the point of either A) killing the VDI project entirely or B) reducing the ROI to the point where the savings the project was designed to achieve are completely consumed and considered a failure.
Using the appliance approach - you could for buy a single appliance for your pilot. Then add more appliances as you scale up with each appliance supporting x users and scale linearly all the way up to your goal with increased performance and availability built in. No extra stuff needed. Oh and by the way - much less expensively than a traditional model. Additionally the support model is brain-dead simple. No multiple vendor calls - no finger pointing.
Converged appliance don't address every Corp IT need out there today - but they can address several and with the predictable procurement cost model may actually allow more projects to get approved. Of the solutions out there today I believe Nutanix has the best overall story, product line up and value.
Bottom line - Converged Infrastructure Appliances are very real, very good and very disruptive. They take the proven approaches that the largest providers in the world utilize and put them in your hands. If you haven't brought them in to your DC to at least test - do so. It may just save your job...
Comments
Post a Comment