Local failures to tolerate in Stretched Cluster

[_steez]

Hello,

I have a question about a Stretched cluster Failures to tolerate setting when creating a storage policy:

Do I need to tolerate any failures in Primary site? When testing failure scenarios - a host (containing VM data) was forsibly turned off, VMs still restarted on other host in the same site.

What should be the optimal configuration in this scenario (4+4 hosts)? As selecting No data redudancy saves storage space masivelly. Only downsides I can think of is impacted network performance (as VM data now has to travel between sites).

In what cases you would want to keep Failures to tolerate to RAID1/5 in stretched cluster deployments?

[depping · Accepted solution]

It depends on your expectations. there's a big benefit to having local redundancy:

• If a host fails and your VM is impacted read IO will still be from the local fault domain
• rebuilds can happen from the local fault domain
• you can withstand two failures across the fault domains, instead of 1

but if you feel that you don't need that, or some VMs are less latency sensitive and a prolonged resync time doesn't matter, just go with no redundancy within the fault domain.