FlashArray//C VMware Managed Snapshot Performance
For more information from VMware on this topic, check out their blog post on backup performance improvement recommendations.
Pure's FlashArray//C was introduced to provide a capacity-optimized storage platform. FlashArray//X on the other hand, is a performance-optimized storage platform designed to provide the lowest I/O latency and highest IOPS performance. When comparing the relative performance of FlashArray//X and FlashArray//C it should come as no surprise that most, if not all, latency-sensitive applications will be less performant on FlashArray//C. The difference in average I/O latency becomes most pronounced in cases where single-threaded and low thread count read I/O are used on FlashArray//C. In practical terms, it means being able to perform one read I/O every ~300 microseconds (//X) vs. one read I/O every ~3,500 microseconds (//C). Such was the case with some VMware-managed snapshot operations on FlashArray//C prior to Purity version 6.1.9.
All of the below tests were run on a VMFS6 datastore with individual snapshots that were less than 350 GB in size. For snapshot creation, about 7.5k write IOPS were run on the arrays to generate data for snapshots; for deletion, the arrays had no load other than the deletion operation. The tests were run on a single VM that was powered on for the duration of the testing.
For deleting all snapshots, there were 3 snapshots present with about 350 GB in total snapshot size reported by vCenter.
Test | FlashArray//X50R2 Purity 6.1.10 |
FlashArray//C40R3 Purity 6.1.1 |
FlashArray//C40R3 Purity 6.1.10 |
---|---|---|---|
vCenter manual snapshot creation | 2 minutes or less | 2 minutes or less | 2 minutes or less |
vCenter manual single snapshot deletion | 8 Minutes | 30 Minutes | 9 Minutes |
vCenter manual delete all snapshots | 10 Minutes | 56 Minutes | 14 Minutes |
Why is there a test duration difference between Purity versions for FA//C?
Starting with Purity 6.1.9, a performance optimization was added to improve read latency for single-threaded I/O streams (like VMware-managed snapshot operations). As VMware's managed snapshot operations make use of low thread count and single-threaded read I/O, the duration of various snapshot activities (like consolidation, deletion) are heavily influenced by storage array read latency. As FlashArray//C read latency for single-threaded operations had been 3.5-4ms prior to Purity version 6.1.9, the duration for these activities was also considerably longer than FlashArray//X which provides less than 1 ms read I/O latency.
VMware has a related KB article.
More information on VMware's snapshot deletion operations can be found in this VMware KB and on this VMware area page.