VxRail 7.0.100 unlocks the potential of NVMe functionality in cache drives, especially when supporting write-intensive applications such as in-memory databases. With the most recent improvements to vSAN, VxRail can now take full advantage of the speed and predictability delivered by high-specification drives, outperforming standard SSDs by up to 60 percent.
With previous vSAN versions, faster cache drives did not provide meaningful performance improvements, although they did provide more predictable latency. The following figure shows that at lower performance demands, the differences between cache drives is insignificant. As higher and higher performance is required, however, higher specification cache drives are required.
Figure 5. Maximum performance
We ran this test based on RAID 1, with no compression or deduplication set up with an RDBMS type workload. During the test, we saw the most significant performance improvement when testing large block operations on the most performant drives. This improvement can greatly benefit write-intensive operations such as those that take place in in-memory database workloads. To protect data from the risk of memory failure, in-memory databases persist in-memory data to storage media and periodically flush all changed data from memory to the underlying data volumes.
With mixed workloads and small block operations, performance was comparable regardless of the drive specifications. However, when we pushed the system, we began to see significant performance deviations depending on the drive’s specifications. In short, below a certain IOPS threshold, the cache drive is not a determining factor. Above that threshold, however, higher performance requires a high performing cache drive.
When we need to stress our VxRail system to achieve maximum storage performance, we have to include the highest-performing drives that are available. The investment will yield huge latency improvements and increased maximum obtainable throughput.