With all the latest mergers, acquisitions and cloud providers coming out of the woodwork, there has been lots of marketing fluff (as a marketer, I can say this) around the benefits and differences between Reference Architecture and Converged Infrastructure. When I first started looking into these, I admit it included a lot of raised eyebrows and general pondering about whether any of this mattered to end users, or if it made a difference in anything cloud. The truth is, I am still not sure. Here’s why.
You might have seen some headlines about EMC and their latet VSPEX release, or IBM and their PureSystems, or even the NetApp & Cisco partnership with FlexPod. These solutions would be classified as Reference Architectures as they are usually integrated and validated platforms that include server, network and storage components with a hypervisor on top. These vendors tout the flexibility as a major benefit as they enable users to mix and match as long as they keep the same basic format. This model uses open APIs and management tools which make them quick and easy to deploy, and really a risk-free option for private clouds. It’s a great way for budget SMBs to put together a cloud environment, although remember that you can theoretically break any type of model if you don’t know what you are doing.
One of the key drawbacks with Reference Architecture is that although they act as a pre-configured system, you still have enough flexibility in choosing components, which really just ends up making it the same as a piecemeal solution. Until the components are integrated together on a customer site and used for that customers’ purposes, it is impossible to predict how it will act. The sad thing is that it’s the end user that is on the hook if it doesn’t work, and if it doesn’t perform to optimal levels, it really undoes all the benefits this model is supposed to provide. At this point, almost all the support calls that result from this model are going to be related to configuration issues, and the finger of blame will be pointed at every component vendor, making it a pain to troubleshoot.
The other key issue is that sometimes these configurations are based on static sizing and architecture deployments that doesn’t bode well to resizing for future requirements. This means that the end user will have to ultimately have to reconfigure and reintegrate their solution and constantly find a way to use what they’ve done to date and get it to work with the new environment. But since you’ve gone ahead and resized, you’ve already gone out of scope with the original configuration which will bring up more support issues.
Single Stack or Converged Infrastructure solutions are another option for cloud environments. These all-in-one offerings include solutions like Oracle’s Exadata and Exalogic, Dell’s vStart, HP’s CloudSystem Matrix, and IBM PureSystems. These solutions have tightly defined software stacks above the virtualization layer, and might include bundled infrastructure and service offerings. These are a great option, but do come with a downside: vendor lock-in.
For example, if you decide to go with HP’s option, you are looking at certain network switches from supported platforms such as TippingPoint or 3Com. This means down the road if you decide to do hardware upgrades, you are tied to these vendors, instead of having the choice to use for example, a Vmware platform.
Speaking of Vmware, a true converged infrastructure is not only pre-tested and pre-configured, but it should also be pre-integrated. Single SKU is what we are looking for. Right now there is only one solution that fits this bill, VCE’s Vblock. These units leverage technologies from Vmware, Cisco and EMC to deliver a pre-configured, single product that is the same no matter where in the world you order it from. I personally love this model because although the environment is fixed, you know what you are going to get. The environment will perform pretty much the same way across the board, making support a lot easier as there is only one vendor (VCE) to work with. And the bugs in your system have probably showed up at some point with another customer as you’ve eliminated a lot of the variables that come with custom deployments.
The other nice thing is that because it is pre-configured and assembled, it has the quickest deployment time. A typical deployment can take between 30-45 days from procurement to production as opposed to a Reference Architecture which could take at least twice as long as you still have to rack, configure, test and get it up and running. This makes Converged Infrastructure a perfect choice for private clouds because the speed of deployment will give you an immediate reduction in your TCO.