Have you ever had an influential teacher or mentor? I have. Mrs. Hanley, my Freshman high school English teacher. It was more than 35 years ago that I enjoyed that experience.
Mrs. Hanley was amazing: She used to get up on her desk to teach, especially when she knew we weren’t paying attention. She taught us how to write “paragraphs of proof” in which we had to have a at least three examples to prove any point we were making. Less than three was not acceptable: if we had less than three, the point was probably not worth making.
When we combined the Micron 5200 MAX and ECO SSDs with VMware’s vSAN 6.7 (update 1) and Dell R740xd platforms in our most recent vSAN Reference Architecture, we saw good things from these three:
- Optimized, balanced design: This all-flash vSAN 6.7 U1 solution using workload-specific SATA SSDs showed both performance and value. With cache tier SSDs (write-intensive, high IOPS, low latency and workload tuned endurance) and capacity tier SSDs (read-intensive, high capacity to bring immense virtualized platforms and data closer to the most demanding applications).
- A platform that balances capability and economy: Unlock pent up value so your infrastructure stays one step ahead of application demands and data growth. This platform is tuned and tested, with proven results and flexibility that enables you to build today to manage tomorrow’s demands.
- An engineer-designed and lab-validated blueprint: This solution enables fast time to deployment with predictable results, helping free valuable deployment teams from the drudgery of experimentation, testing and reconfiguration. This enables them to focus on higher-value tasks like rapid deployment, faster time to value, and building bottom lines.
We studied two very different working sets: one that fits entirely into the cache tier (as one might find with a pre-planned, fully documented application and workload rollout plan) and a working set that overflowed the cache tier (>50 percent larger than the cache tier capacity).
Why test both? Simple. While everyone in IT and project management strive to plan with complete accuracy, we also know that absolute precision is rare. The number of VMs grows quickly, the number of uses migrated is higher, and the value of consolidation is realized more quickly.
We also compared three different configuration profiles to a baseline profile — one configured to emphasize performance, one to emphasize density, and a third that was a balance of both.
That’s why we look at the unexpected as well as the expected. The “out of cache” working sets and the in-cache working sets show different results for different profiles. We want to help you know what to expect when you build your vSAN for your environment and your applications.
Here are a couple of examples that show why Micron validates both out-of-cache and in-cache working sets across four configuration profiles:
These two figures compare IOPS and latency for a 0 percent read (100 percent write) test where the working set fits into cache (left) and where it exceeds cache capacity by about 50 percent (right) across four profiles (note that the vertical axis scaling is different).
Here’s a similar comparison for a 70 percent read (30 percent write) test for the same two working sets:
Which one is right for you today? Which is right for your plans for tomorrow?
Those aren’t all.
This reference architecture shows how the read / write ratio changes both the IOPS and latency as the workload moves from 100 percent write through 100 percent read, enabling you to find the configuration that’s the best fit to meet your needs.
The Micron Reference Architecture provides detailed planning and deployment directions to optimized, pre-engineered solutions that have been fully validated at Micron’s Storage Solutions Center in Austin, Texas. We provide proven processes and benchmark data so end users, system builders, and independent software vendors (ISVs) can build next-generation solutions. This all requires less invested time and potential risk, enabling IT teams to focus on higher-value tasks. They can rely on:
- Precise sizing and component selection
- A set of targeted, standard of industry benchmark results
- Configuration and deployment guidance
There’s that “3” again; good things do come in “threes.” Mrs. Hanley would be happy with this I think.
How much of your infrastructure is already virtualized? How much of it will be?