Why use RAID10 if all data is being replicated with VSAN?

Trey · Thu Jun 04, 2015 4:19 pm

This was the question that my manager asked me when we were speccing out our VSAN setup. We currently have two hosts with about 10TB of raw storage on each. With RAID10 we are halving our usable space by mirroring. Then we are duplicating this data on the other host. Why not use RAID0 on each server to maximize performance and usable space?

I couldn't articulate why using RAID0 exclusively was a bad idea. The only reason I could think of at the time was for local redundancy. If a hard drive fails in one of the servers, it doesn't take the entire server down. You can replace the drive and be back up and running with no downtime for either host. What other pitfalls am I overlooking? This seems like flirting with disaster but I need to be able to explain why.

: YEs.jpg (22.24 KiB) Viewed 13144 times

lohelle · Thu Jun 04, 2015 4:53 pm

Well..
If one of the hosts fail (or down for maintenance), I would feel a bit worried for the other node..?
What if the RAID-controller failed on node1, and you had to wait for days for a replacement? Would you feel safe in the meantime?

darklight · Fri Jun 05, 2015 8:53 am

Hi guys.

RAID0 grants top performance for sure. And since you always have an actual copy of data using StarWind it's safe enough I believe.
Of course, it's better to have a spare HDD for a quick replacement to minimize server downtime and possible risks, but it's up to you anyways.

Fri Jun 05, 2015 1:49 pm

There are two things here:

1) StarWind takes care of so-called "data locality" (will finally have a document on that, right, Max?). Quite similar to what Nutanix does and VMware does NOT

So we prefer VMs being run where VM data is hosted. This leaves our cluster nodes as a "islands" of compute and storage resources and each of them is stand-alone unit. This means we really like having SOME protection of data locally so RAID10 (non-LSFS) and RAID5/6 (LSFS) are recommended. So host would allow to have partner AND local disk failure *and* keep going after that. So-called ( N - 1.5 ) scenario

2) If you follow 3-2-1 rule (see links below) and you have primary data, on-site and off-site backups then you can go with RAID0-all-around

http://windowsitpro.com/blog/why-3-2-1- ... akes-sense

http://blog.trendmicro.com/trendlabs-se ... -2-1-rule/

http://www.veeam.com/blog/how-to-follow ... ation.html