6

We're planning to move some production workloads to a UCS environment using vSAN and all flash. We are looking at vSAN with replication as we have two identical nodes in two data centres.

Is there a performance hit from using replication that would negate the benefits of the all flash storage?

We are currently using LeftHand and the write latency introduced by a commit having o be confirmed from all nodes before the write commit is passed up to the storage consumer makes it very slow. I've tried to find documentation on this but came up short.

  • I'm not really sure I know what you're talking about. You say "replication", which for me means vSphere Replication. But actually it sounds like you are talking about a stretched cluster. Could you tell me which of these two you are planning to implement? – Mario Lenz Oct 22 '18 at 18:26
  • Have one, looking to implement the other. – Matt Oct 22 '18 at 22:13
  • I'm still not sure what you mean. What's the "one" you have and what's "the other" you're looking to implement? Do you plan to build a vSAN stretched cluster? – Mario Lenz Oct 23 '18 at 20:14
  • If you were familiar with the technology I was referring to in my question, that would be clear. If you're not, then this probably isn't the question for you. Besides which, this question is answered. – Matt Oct 30 '18 at 19:36
9

If you are talking about storage replication, you should check two important things:

  1. Storage performance - in your case it shouldn't be a problem.
  2. Network performance between the servers.

Is there a performance hit from using replication that would negate the benefits of the all flash storage?

I am using StarWind VSAN and have zero issues with write performance on all flash HA config.

| improve this answer | |
  • 3
    Thanks, that looks very interesting, and is certainly priced way more competitively than VMWare vSAN,. – Matt Sep 25 '18 at 23:48
3

Is there a performance hit from using replication that would negate the benefits of the all flash storage?

It depends on how much 'spare' resources you have left on your cluster after all your VM load. If there's lots of spare CPU/memory/network-bandwidth/disk-bandwith-and-IOPS then no you should see little if any impact during replication, essentially it'll work as a background task and use those spare resources to do so.

If you've not left adequate resources then the impact will be greater as the replication does take an amount of those resources away while it replicates.

| improve this answer | |
  • Your comment about it being a background task has piqued my interest. Fairly sure with the :eftHand this isn't the case which seems to introduce the write latency. – Matt Sep 25 '18 at 23:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.