After a few days of meetings revolving around architecture, I figured this blog post would be fitting. It is an everyday conversation that normally ends up with the words “Architecture Matters” coming out of my mouth. It is all too common for people to want to leverage the newest IT shiny gadget. After all, most people in the IT field have a passion for technology. It doesn’t matter what it is phones, computers, tablets, watches, and even their IT infrastructure.
Let’s dig a little deeper into some of the causes or effects of this happening.
Chad Sakac wrote a great blog post on the different storage architectures. The key is they were all designed to excel at some requirement. This will enable them to be best of breed at what they were designed for. An EMC example here is the VNX and Isilon. The VNX is good at file services. However, it cannot compare to Isilon. It was purpose built (different architecture). The most important thing is to understand what strengths and architecture affords and at what cost.
When we look at this in the data center as a whole, not just storage. Understanding the requirements are king. Let’s look at an example. A retail chain wants to put an infrastructure in its remote locations to handle all local IT requirements. The company’s lead IT person loves technology and was reading about the new hyper converged infrastructure, EVO:RAIL. He is passionate about being on the cutting edge of technology, and is pushing to get it in his environment. Some of the business requirements are AD, DNS, DHCP, SQL, POS, file and print, and the Security system. Can we meet them requirements with hyper converged? Yes. The real question is should we? All the servers should fit easy. The component that makes it not the best option is the security system. It has 80 HD cameras recording 24hrs a day and they are looking to keep video for 90 days. This is about 25TB of required space. We could use (4) EVO:RAILs to get the correct space needed. The solution would be easy to manage, and fast to configure. It would give you 16 ESX hosts when you may only need 4. This is because CPU and Disk do not scale independently. EVO:RAIL could be used, it would not be the optimal architecture.
Looking at the above use case and just making a small change to drop the security system and adding 100 virtual desktops may make EVO:RAIL the best option. Yes. We could do both by pushing the video out to a better suited platform… if it was allowed.
The example above is just looking at a hyper converged use case. This same issue is true on smaller scales as well. Blades vs Rack servers is another common architecturally decision as well to show small scale. All designs need to look at the way its components will effect scale, protection, availability, agility, and management to name a few.
Architecture by design, not by accident.