How much nodes of vsan will be the best？

[lijianhui1]

I know the maximum of vsan is 64 nodes.

But how much is the vsan recommended a cluster of nodes achieve the best performance and stability？

If I set up a big cluster，such as 64 nodes，Should I divide many small area inside the big cluster？

[daphnissov]

But how much is the vsan recommended a cluster of nodes achieve the best performance and stability？

There is no such thing as "best" without knowing specifics including your goals. For some workloads (and the amount), 4 nodes are perfect. For others, more nodes are needed.

[MohamadAlhousse]

Hi lijianhui1​,

This is according to your business needs in terms of capacity, availability, RAID level, and how much hosts you accept to tolerate (FTT).

Please see the below table when sizing your cluster:

To achieve the best performance you can consider the following:

1. Choose All-flash VSAN rather than Hybrid-mode.

2. Although 3-nodes VSAN cluster is supported, the minimum recommended is to go with 4-nodes to be able to evict one node of the cluster with full data evacuation in case of a maintenance windows.

3. Consider using 10GB or 40GB uplinks to have a good throughput on the VSAN network.

4. Consider using multiple storage controllers and diskgroups per server to have redundancy in case one disk group goes down.

Again, all depends on your needs.

Hope I answered as much of your thoughts.

Regards,

[TheBobkin · Accepted solution]

Hello lijianhui1​,

64-node clusters can be unwieldy other than for heavily-automated, homogenous workloads (e.g. VDI) - I have seen a few of these in the wild and while they spit IOPS like fire they need to be immaculately well implemented and maintained (same goes for any 64-node non-vSAN cluster).

Putting all your eggs in one basket isn't necessarily a good idea either - this is a clustered solution and thus issues on one node can potentially affect other cluster members, the same goes for performance.

What would likely be a better solution is to spread these 64-nodes over appropriately-sized clusters for the workloads they are running, e.g. 4-node Management-cluster, 12-node Test/Dev, 24-node Desktops 12-node Prod, 12-node Prod-DR (or a 12+12+1 Stretched for Prod).

Doing it this way can help ensure workload isolation and would also allow for far better customisation to better suit the workloads e.g. All-Flash high-spec servers for Prod, lower-spec and size All-Flash for Desktop, Hybrid for Test/Dev etc.

Bob