VMware vSphere Design classroom training wrapped up today by covering modules 6, 7, and 8 which was Storage design, VM design, and Management/Monitoring design, respectively.
In the storage module, we talked a lot about protocols, IOPS, capacity, tuning, zoning, multipathing, and RAID types. Furthermore, we applied rating indexes to the protocols and technologies such as cost, complexity, and flexibility. Understanding this information aids in formulating design choices for the project. A few interesting tidbits I pulled out of today’s discussions:
- VMFS block size selection has a negligible effect on the design. Don’t stress over 1, 2, 4, or 8MB. If you want the most flexibility, choose 8MB and be done. There is no measurable performance impact and the amount of disk space wasted due to placing 10-20 VMs worth of small files on a 500GB LUN is negligible, with or without sub block allocation.
- This one I was a bit surprised at: “Don’t adjust LUN queue depth“. Rationale: if you are having performance issues in the form of queuing, it is likely the result of a larger problem elsewhere such as front end port saturation on the array. I will agree that leaving queue depth at its factory defaults on the HBA and VMkernel side (Disk.SchedNumReqOutstanding) makes implementation that much easier, but it doesn’t mean it’s always the right solution. In my VCDX design submission, I increased queue depth from the default of 32 to 64 (128 is max). By the way, the best whitepaper on this discussion I’ve ever found is Scalable Storage Performance. It’s quite short and in my opinion explains the key details in better detail than the VMware KB articles.
The VM design module was fairly predictable. It basically focused around right-sizing and choosing the appropriate virtual hardware. It touched a bit on security and naming conventions. Nothing earth shattering here.
The final module, Management and Monitoring Design, briefly covered scripted installation, infrastructure security, vCenter placement in the logical design, templates, time synchronization, CDP, SNMP, external logging (ie. syslog, splunk, vilogger), alarms, and performance monitoring.
Final thoughts… good class, I’d recommend it. I felt many ties to the VCDX design process and hopefully this class will be beneficial to those in pursuit of that certification (or the VCAP-DCD). My feedback on the class was for added days. We got through the entire 400 page classroom book in 3 days but we flew through some sections. The biggest compromise of time was probably the labs. I felt that we did not spend a lot of time hashing design decisions, justifications, and upstream/downstream impacts. This is where you need to spend considerable time thinking through each design decision, not only for the benefit of your customer, but you’re going to get hammered by the VCDX panel on decisions as well. Ample time spent here is a good practice.
The instructor is sitting the VCAP-DCD BETA exam the same day as I. He seemed a bit curious as to whether or not this course will cover all of the exam questions we both will face. I’ll solicit his feedback after the exam to see what he thought, and then compare it with my thoughts. I’ll post a recap on my exam experience, as I usually do, so stay tuned if your interested in the VCAP-DCD.