Concurrent Snapshots question

[dynanet]

Hi @ all

I have a little problem with concurrent snapshots.

We have the following two scenarios:

1. Hyper-V Cluster (two Server 2012 R2 machines) with an underlying SOFS Cluster, also Windows Server 2012 R2
I've added the SOFS Cluster under SMB3 Clusters and the Hyper-V Cluster under Microsoft Hyper-V Clusters, in the "Backup Infrastructure" tab
2. Standalone Hyper-V Server (2012R2) which has "local" storage trough iSCSI. Also added under Standalone Hosts.

Veeam B&R v7.0.0.871 is installed on one of the two Cluster machines, the server where Veeam is installed acts also as an Backup Offhost Proxy. The Backup Repository is local storage on the same server, connected trough iSCSI.

While backing up VMs of the scenario 1, it's using the Fileshare Provider (mode: Crash consistent) and its also using the Hyper-V Offhost Proxy. Unfortunately i can only snapshot one VM at the time, the other VM's of the same backupjob are getting queued, Veeam says: "Resource not ready: Snapshot"

In scenario 2, Veeam uses the Microsoft Software Shadow Copy Provider 1.0 (mode: Crash consistent). This job immediately starts with 4 VM's, which is set in the Manage Volumes tab under "Max concurrent snapshots" on the server. The job runs onhost, because we dont have an VSS hardware provider from that storage.

Why is the first scenario only snapshotting one VM at the time? This takes ages to complete the whole backup job.

Did i miss something?

Thanks for your help.

Kind regards

dynanet

[foggy]

While snapshot is being created, the corresponding VM owner node is locked, as well as the CSV owner node, so that other snapshots involving the same resources cannot be created until the current snapshot is released. This is by Hyper-V design, only one snapshot creation operation is possible at any given moment in time. So if your VMs reside on the same cluster node or share the same CSV volume, they cannot be backed up in parallel.

Btw, are you intentionally performing crash-consistent backups (i.e. disable application-aware processing in the job settings) or the job fails over to crash-consistent mode for some reasons?

[dynanet]

hi foggy

sorry for my late reply.

I did not intentionally disabled anything. I just set up a new backup task with default settings.

Kind regards

dynanet

[foggy]

Then there's something that prevents VSS backup of this VM (for example, due to Hyper-V Integration Services failure, or if the VM guest OS has dynamic disks).

[dynanet]

Hi foggy

the backup works fine over SMB3, its just the fact, that I can only backup one VM at the time.

I also experienced, that Veeam takes much longer for snapshotting (around 2 mins) than it takes, when i take manual snapshot directly in the Hyper-V console.

You can imagine backing up 200 VMs over night could get quite tight.

My questions are:

* Why does the snapshot in Veeam takes so long?

* What does get locked, the CSV offered by the SOFS server or the Hyper-V cluster node?

* What would you recommend in our scenario (Hyper-V Cluster with SOFS SMB3 Server)?

Thanks in advance for your help.

Kind regards

dynanet

[foggy]

* Why does the snapshot in Veeam takes so long?

When you take manual snapshot in the Hyper-V console, Hyper-V checkpoint is created. While Veeam B&R performs volume snapshot that includes host VSS freeze with all that it entails.

* What does get locked, the CSV offered by the SOFS server or the Hyper-V cluster node?

Both.

* What would you recommend in our scenario (Hyper-V Cluster with SOFS SMB3 Server)?

If you get at least two volumes, one per each SOFS node, you will be able to get parallel snapshots for VMs residing on different cluster nodes.

[dynanet]

thanks for your info.

you could close that thread.

[TSIEnergy]

Since this logic is clearly defined wouldn't it be fairly straightforward to have Veeam interrogate the Hyper-V cluster to get the nodes that VMs are currently running on and their corresponding CSV LUN (you can with a WMI call). It seems to me that Veeam could add a feature to "optimize" the backup job as a whole so you can specify what VMs in the cluster you want backed up and in what priority (not necessarily order) and Veeam can dynamically select when individual backups occur based on current conditions. Right now I have separate jobs for each of my CSV volumes to minimize the VSS freeze-out BUT with the owner node and CSV both needing to be available to start things always delays jobs or forces me to retry.

Mark Ringo

[foggy]

Mark, just FYI, Veeam B&R v8 will add the option to backup multiple VMs from the same volume snapshot, so VMs from that single snapshot will be processed in parallel.