question re stretch vsan cluster and vm protection at recovery site

[tekhie999]

hi

i am looking to deploy VSAN6.2 Hybrid in an active/active configuration across 2 sites (witness at 3rd site)

am i right in understanding that if i suffer from a Datacentre failure, and my VM's are recovered at their secondary site, there is no protection for the VM's whilst running at the secondary site whilst in this type of DR mode ?

So if i was to lose a DISK / CONTROLLER / HOST at the site that is available, i would lose the VM's running on that DISK/CONTROLLER/HOST, as there would be no +1 to recover with. 

Is that correct ? if so is there a smart way around this scenario ?

Thanks !

[zdickinson]

Good morning, it depends on how many hosts you have at each site.  If you're running 3 nodes at each site and have a VM with FTT = 1, then if a site fails you can then have 1 host failure at the active site and the VM will still run.

As an aside, with dedupe and compression enabled, all flash can be cheaper than hybrid depending on how well your data dedupes and compresses.  http://www.yellow-bricks.com/2016/03/04/hybrid-vs-all-flash-vsan-are-we-really-getting-close/

Thank you, Zach.

[tekhie999]

Hi Zach. Thanks for the response

i am planning on deploying 3 clusters. 1 with 5 hosts at each site, 1 with 6 hosts at each site and 1 with 8 hosts at each site

The witness for all 3 will be deployed at a third site.

would the nimber of host faiures to tolerate be the same regardless of how many hosts are left running at the active site? or would you be able to tolerate more host failures relative to the number of active hosts ?

thanks

chris

[zdickinson]

Good morning, it's a little bit of both.  If FTT = 1, then one of the three hosts that contain the VM data/witness can fail.  What the additional hosts allow for is rebuild to start so you don't have to wait for the failed host to be back online to be protected.  Thank you, Zach.

[tekhie999]

hi zach .... thanks ... but from what i understand you can only have an FTT=1 with a stretch VSAN cluster. So if i lose a datacentre, the vm's will start at the opposite site using HA, but will NOT create a new copy of them selves whilst running at that site... and therefore reenabling the FTT=1.  They will be running off their replica. And although the witness is still available, if a Host fails, then the 'replica' object has gone, leaving only the witness .... therefore the vm(s) will become unavailable

Is that not correct ?

Thanks

Chris

[zdickinson]

My apologies, I forgot the stretched part of this thread.  You are correct, FTT = 1 is the max as 3 fault domains (Data 1, Data 2, Witness) is the max supported.  I had not thought it all the way through, but it seems you are correct on the second part too.  If you lose Data 1, it won't rebuild a second copy of the data within Data 2 as that would be in the same fault domain.  Thank you, Zach.