New Nimble Customer - VDI Sizing Questions

[vnelsontx132]

Hello Nimble community, my company is about to become a Nimble customer and we are very excited about what this means to the performance of our Virtual Servers and VDI environments!

I have been reading tons of information on proper sizing for VDI and Virtual Servers but have found some conflicting best practices and wanted to get some validation from the community.

I am very aware that Best Practices don't fit all use cases but here are my questions:

1. Block size for VDI Gold Images should be 4K?

2. Gold Images, Replicas and Linked-Clones should reside in their own respected datastores (multiple datastores for Linked-Clones)

3. Should I apply different performance profiles to each type of datastore for VDI? Gold Image, Replicas and LC's?

4. We use Veeam as our backup solution, is it better to use Veeam as deep backups and utilize Nimble Snapshots for <72 Hour Point in Time restore points? or the other way around?

5. When building Fault Tolerant Datastores will it be completely de-duped? e.g.: File Server is in Datastore 1, FT instance is in Datastore 2 (zero block change because of CASL and De-Dupe?)

6. Do people separate Exchange into it's own datastores and apply performance profiles or include into other Server Datastores?

7. Would like to get some feedback from users on how they designed their file server, from Nimble's document there are various configuration options. Does anyone have experience with the Win2012 Clustered scenario?

That should be a good start!

I am curious to hear from other VDI shops how you sized/built your datastore designs, pain points you experienced or lessons learned.

Again we are very excited to become a Nimble shop because of the great benefits and features of the Hybrid Storage design.

Here is the first of many posts!

Thank you in advance.

[nglick57]

Hi Brian,

Thanks for the great questions.  I'll answer what I can, but remember, your mileage will vary depending on the uniqueness of your environment.

1.  Sounds like you're running VMware Horizon View with Composer and Linked Clones.  I would choose 4K as the block size.  You can use ESXi or VMware performance profile and I pretty sure they're 4K.

2.  Yep, it's always a good idea to separate out the gold images and replicas just in case you need to make changes at the volume level.  This way you don't affect one or the other.

3.  4K should work for the performance policy, but you may want to apply different policies for metrics.  When you start sending data to Infosight those different performance policies will show up and you'll be able to tell where space is being used, how, estimated projections, etc.  The more granular you are when you set up your Nimble, the more granular the data will be when it comes back after being digested by Infosight.

4.  Sorry, not a Veeam guy, hopefully someone will chime in.  My opinion about backups though.  Snapshots are great for your user's data and then it can be backed up with more traditional methods, unless you need to get tapes offsite ASAP.  This way you're not rushing to backup data through traditional methods.

5.  Nimble doesn't de-dupe.  We do inline compression.  It is enabled by default, but you can turn it off if you choose to.  The nice thing about your Linked Clones is they're going to be pretty much de-duped already since all of the users will be working from that original snapshot and only making changes to their delta disk.

6.  Yep, good idea to separate out different application data.  I'd select the performance profile for Exchange too.

7.  I think you're referring to a CIFs server correct?  Unfortunately I don't have a ton of experience creating Microsoft File servers, so hopefully someone will chime in.

Take a look here too, it's a document Mike McLaughlin and I wrote with regards to SE Sparse:

https://connect.nimblestorage.com/docs/DOC-1291

Here's also more information on my blog site regarding it plus a ton of other VDI information:

Glick's Gray Matter: VMware Horizon View - To SESparse or Not to SESparse, That is The Question!

I hope this helps!  Please feel free to let me know if you'd like more information or clarification.

Thanks,

Neil

[vnelsontx132]

Neil, thank you very much for the quick response. If there are additional documents that I should be reading, please feel free to point me in that direction! The SE Sparse links you sent are a big help.

Responding to your answers:

1. Yes we are running VMware Horizon View 5.3 with Full VM's and Linked-Clones. Any difference between the ESX or VMware Performance Profile or are they the same.

2. Good to know, Best Practices Document states to keep replicas and LC's in the same Datastore

3. Do you have any suggestions for what specific policies I should apply to the various datastores? Gold Images, Replicas and LC's

4. We run Persona Management back to our File Server so we don't feel there is a need to snapshot the cached persona data for each LC, more concerned with Virtual Servers than the View environment for snapshots.

5. That was my mistake I meant to say compression, is there any specific compression ratios that you know of for FT datastores?

6. If it is wise to separate Exchange, I'm assuming it is best to separate SQL DB instances to its own Datastore? Is there a SQL Performance Profile?

7. Yes, CIFS is what we are planning to do, implementing the Nimble is giving us a great opportunity to redo/new servers. Nimble has a great document for various ways to server windows file services through clustering, iSCSI software for Windows, etc. Just not sure what is best design for Nimble, experiences, etc.

[nglick57 · Accepted solution]

Hi Brian,

Sorry for not getting back to you sooner!  How's work on the Nimble going?  To answer your questions:

1.  Nope, I think ESX and VMware performance profiles are both the same, 4K.  But it's a good idea to be as specific as possible because InfoSight will break everything down for you and you can show all kinds of cool metrics by different performance groups!  Have you had a chance to look at InfoSight yet?

2.  It's really up to you when you're placing the replicas and LC's.  Since we're compressing and not de-duping, you don't have to worry about like data being in the same volume.  Experiment a little and see what you like best.

3.  4K should really work best unless it's an application that runs 32k or 64k, etc.  The labeling is really the important thing for InfoSight.  So if you create your own custom ones, just select 4K.

4.  What are you using as your File Server, if you don't mind me asking.

5.  Compression is really going to depend on the data that the Nimble is trying to compress.  I know that's an "It Depends" answer.  :-)  But it really does depend on how compressible the data is.  InfoSight should be able to report on how well your compression is running and I'd be very interested in hearing back from you on that!

6.  Yep, I'd definitely separate out the different applications.  For both the performance policies, but again for the InfoSight data.  I believe there is both an Exchange and SQL DB performance profile.

7.  Take a look at this paper Brian.  I believe this is what you're looking for with regards to using Nimble for file sharing.  Let me know if it's not the right thing.

Best Practices Guide - Microsoft Windows File Sharing

Neil