How many datastores for?

[jimbobt11]

Hello

Might be a hard one to answer with no real perf metrics however I would like to gain some info on scaling out app volumes 2.14 for around 30 appstacks and 1000 concurrent users, appstackes using around 120gb at present on one datastore running on SSD San.

At the moment there's very little users doing the PoC I need to understand when we ramp up the usage will I need a handful of datastores or just one?

What's the best approach for distributing the load?

Create a storage group and replicate the stacks? My colleague reckons one will be ok and it will perform fine as mostly reads on the Hitachi San when mounting the vmdks.

We will have 1000 users using horizon 7.7 with floating instant clone pools. A mix of laptop users and zero clients.

There will be ten + pools with appstack assigned to user at logon.

Our users will be on o365 eventually and using writeable volumes the uia template was increased to 30gb thin provisioned. Created at logon and assigned to a group containing all users. Uem is setting the location for outlook 2016 to the snapvolumetemp.

With the writeable volumes we are going to have 10 * 3tb datastores and create a storage group to spread the WV across the group and this seems to work quite well for a couple of small tests I've done granted again only a handful of users on the PoC. Windows 10 1803 with office 2016 click to run in the gold image.

We have horizon vrops in the environment but to be honest my understanding of it is not the best.

Any info on scaling this out would be helpful we are using vcsa 6.5 with esxi 6.5.

We went to our partner who helped set this up but no response as of yet.

[sjesse]

I'd like to see what larger environments see as well, but from my research if you have don't have flash storage you want to limit standard appstacks to 120 attachments per data store. All flash data stores are supposed to be able to 1000 per datastore, but thats depended on your use case.

VMware Knowledge Base

I have a hybrid array so I'm trying to keep with that 120 attachment per datastore limit, and I haven't seen any issues. I monitor latency on the datastore, and I can get iops based data between vcenter/ vrealize monitoring and from Liquidware Labs Stratusphere. Stratusphere  can do in guest statistics as well of grabbing statistcs from vmware, and has various reports you can use to watch how many iops each vm is using.

[jimbobt11]

Thank you very much our whole San is SSD.

Forgot to mention we have two appvol managers with 4gb ram. The agent in the gold imAge points to a DNS alias. I don't think it sits behind an actual load balancer.

Not sure how I can profile as we don't have liquidware or equivalent. I will research vrops and look at kB you provided.

Back to WV we have set in uem outlook 2016 admx to cache 3 months of emails into the ost we  are not downloading shared mailboxes first time It synced into a new WV / OST it took the best part of a day from o365 and at the end of it my ost was 5gb and windows search edb 3gb.

Should have really looked at iops then but did not....whoops

[Ray_handels]

If you are using full SSD, IOPS is the least of your concern.

We are using hybrid storage and have not seen that many issues with IOPS in regards to writables and appstacks. Mostly we saw issues when we did a back-up while also assigning and using appstacks, that was just a tad to much.

In regards to writables. Best pratice is to not have more than 500 on a datastore, mostly due to the fact that browsing on a datastore takes a lot of time. We had a lot more writables on a datastore and when creating a new writable or searching for one on a datastore you almost could get a cup of coffee. Luckily when attaching writables and appstacks, we didn't see this issue at all.

In regards to appstacks, I would always use  a storage group with at least 2 datastores in it, even if you only have a few users, at least you have some redundancy build into your environment and it is really easy to set up and works quite well.

[sjesse]

Ray how many do you appstacks in general have in a datastore, I don't have as big as environment as that I've gathered you have, but with our Nimble cs3000 does around 5000 iops total right now, but I noticed in one of our first data stores that was over 100 attachments we started getting issues.  Nimble states that 500 number as well, but we had a health assessment that uses that vha appliance a bit ago and that popped out the 120 for non SSD datastores, which is why I use that one. We do have some users doing non desktop like workloads as our in guest monitoring tools(Liquidware Labs Stratusphere) was seeing some uses using around 1000-2000 iops, but thats more on the instant clone datastore then the appstacks one.

[Ray_handels]

Ray how many do you appstacks in general have in a datastore, I don't have as big as environment as that I've gathered you have, but with our Nimble cs3000 does around 5000 iops total right now, but I noticed in one of our first data stores that was over 100 attachments we started getting issues. 

We are also using Nimbles but only deal with the VDI side of it. We don't look at what the datastore does in regards to errors. What did you notice after that 100 attachments? Did it get slower? Were appstacks not getting attached?

We have 2 datastores for Appvolumes and have about a max of 1000 concurrent users. We don't see issues in regards to appstacks or peformance.

[sjesse]

its was just performance, and maybe it wasn't related, but we put a lot of the best practices that tool spit out into place and noticed improvements afterwards.