POC HyperV Cluster, CSV, Software VSS

[drubs]

Hi, I'm just after any suggestions/thoughts on the Proof of Concept that I am currently planning for our Veeam deployment. Plan is as follows;

HyperV
-3 Node Hyper-V Cluster (Server 2012 R2)
-VMs stored over 2 CSVs
-Underlying FibreChannel Storage (no access to Hardware VSS, so using SoftwareVSS)

Veeam Backup and Replication v8
-Physical Server
-Direct Attached Storage (SAS to JBOD)
-Storage repository provided by Storage Spaces (6PhysicalDisk, 1PhysicalDiskHotspare, 2wayMirror, 3Column)

Network
-Everything is talking over 10GbE

Backup Jobs and Scheduling
-50VMs between multiple tenants

-2 Backup jobs per tenant (a job for TenantVMs on CSV1 and a job for TenantVMs on CSV2)
--CSV2 job only occurs after CSV1 job has finished
--And each tenant has non-overlapping backup windows
--Daily Incremental
--Saturdays are SyntheticFulls

-Each Job uses the following;
--OnHost Proxy
--App-Aware enable
---SQL/AD/SharePoint/Exchange VMs set to Require success but Copy Only
---All other VMs disabled

Currently have the POC working with a standalone HyperV server and its performance and stability has been perfect. The only thing that worries me at the moment is backing up VMs on a CSV using the Software VSS. So I guess if anyone has had experience setting up similar scenarios, and what they learnt, I would love to hear them?

Cheers
Drubs

[Shestakov]

Hi Drubs and welcome to the community!

Your infrastructure looks good in general. Just have 2 comments on it.
You are correct, Hardware VSS is a preferable option since software VSS has a number of limitations and can decrease the host performance.
And using On-host proxy is also is not a recommended approach. If you have an option to use offhost physical VM as a proxy, go for it to shift the processing out of the production hosts.

Thanks!

[nmdange]

There is no reason to have separate jobs for VMs on the two CSVs. You also don't necessarily need a job for each tenant, especially in v9 with per-VM backup chains. Ensuring different jobs don't run at the same time is a good idea.

You may want to consider enabling the "transform previous chains into rollbacks" if you have the "create synthetic full backups periodically" option enabled. You might also consider unchecking that and just use forward incremental forever. If you create a synthetic backup weekly, you will use more disk space on your backup repository.

If you use a scale-out file server cluster in front of your storage, you can do an off-host backup proxy without needing to have storage that supports hardware vss (and even storage that does support it often has bugs/issues with it). The environment you have is probably a bit small to justify that though.

[Shestakov]

Thanks for the qualified input, nmdange

Separating jobs by setting right job start time is a good idea, but I would not use backup windows. Jobs can work simultaneously with no issues due to parallel processing.

As regards to synthetic and active fulls / rollbacks etc. there is a great related topic you may want to get familiar with.

[drubs]

Thanks heaps for the input guys, will take it on board.

[Shestakov]

You are very welcome drubs.
More information about best pracices etc. can be found in this community forum.
If you can`t find an answer, don`t hesitate to ask us!