VMware vSAN 6.7 U3 R6 v R5 & R27 U3- All flash

We are looking to migrate to a vSAN environment, and will deploy at least a 12 node cluster. 90% of the environment will not be particularly IOPS hungry or latent sensitive. And my first approach, was to utilise a storage policy of R5 FTT1, thin provisioned. Any performant will be thick provisioned on R1. But, from reading R5&R6 suffer from poor performance and higher latency compared to using R1. Since R27U3, I read that there has been improvements of using erasure encoding in conjunction with dedupe/compression.

R6 has a double write parity, and with my number of nodes, I should run a FTT2, but I am looking at 100TB of storage. Is there a big difference write/latent wise R5 v R6? Are there any updated benchmarks comparing R1-R6 on R27? I am migrating via HCX, and I don't want to to retrospectively change my storage policy on that volume of data, although I could change problematic VM's if need be. Our vSAN is NVMe cache and all flash storage. Coming from a storage background, anything R6 was generally deemed archive workloads. Ideally, money no object R1 FTT=2, but this is too costly. 10% of my environment is running DB's, but the majority should fall to R6 FTT2, but I am worried of the downsides of R6. Are the improvements of R27U3 performance documented anyway and substantiated.
Has anyone completed any testing, any advice, pointers?





0 Kudos
1 Reply

You know that you can decide storage policy even on per vm-disk level ?

You can set up R5 FTT1 for your nonprod, use R6 FTT2 for non-critical prod and R1 FTT1 or even R1 FTT2 for critical things

Regarding the performance hit from R5 and R6:

Please check over excellent Vmworld sessions from Paudie o'Riordan

R5 and R6 utilize much more network streams due to the write amplification

Your network config will be critical for VSAN performance

Think about things like: NICs speed, MTU, NIC Teaming, switchport buffers