There are no shortage of ways to design infrastructure to solve a specific problem. One popular new trend is the modular server, as Rick Vanover explains.
One of the things I enjoy in my professional capacity is that I constantly am exposed to different environments. Most of those environments are focused on delivering an optimized VMware or Hyper-V infrastructure, which makes sense today.
For midsize or larger organizations, I’m indeed a fan of the pod-based approach for data center provisioning techniques. But for smaller organizations, there is a great opportunity to deliver all of the features for a virtualized infrastructure; however, there are clear guardrails around price and complexity.One way to solve this is the modular server approach. I spoke to a system administrator at a non-profit organization (very price-conscious) recently, and this person totally sold me on this approach for the right size organization. The basic principle of a modular server approach is that there are a limited number of compute resources, but they have direct access to a storage array. This is the key difference between a modular server approach and a blade environment. Figure A below outlines how the modular server approach would deliver four servers running VMware vSphere:
Shared storage - with no SAN
This logical expression of an infrastructure is much like others that may leverage shared storage and other typical components associated with a vSphere infrastructure. The key difference here is that the storage is provided via a local storage array, in many cases with modular storage, it is a SAS interface. SAS interfaces can be provisioned as multi-port. Think about the file system that may be used for vSphere environments -- in this arrangement, a VMFS volume is by itself a clustered file system. This means that there is no requirement for a storage coordinator node, and the storage itself can manage the access by multiple ESX(i) hosts.I’ve seen this configuration used in a number of configurations, including supporting retail operations and smaller data centers. The person I spoke with at length previously was using an Intel Modular Server Chassis like that shown in Figure B below:
Note that there is a good amount of (potential) storage available for the (up to 8) servers on the chassis. That is the difference from the blade approach (which also has its benefits). You might ask, "What if the chassis or the storage array fails?" That’s a fair point. I spend a lot of my time around data protection, so I’d design around failure in that regard.
The fact of the matter is that the flexibility here is great. You can provide VM migration capabilities, and the array locally on the server is supported to be VMFS -- which means you can use VMware vMotion. Don’t believe me? Check the vSphere Compatibility Guide online. The small environment does not have to settle for local storage only in this situation and gets (some) of the benefits of shared storage.
Have you used a modular server chassis? What is your experience with this compute platform? Share your comments below.