Very Slow Backups over CIFS to VNX 5550

[dzeleski]

Hi all,

I have a UCS cluster pointed towards a VNX 5500, our VMs run on the block side and our veeam backups run to the file side. Copying a file to the same mount our backups go to shows full 1gbps speeds (80-110MB/s). But Veeam backups crawl at 10-20 MB/s at most. I have another UCS cluster at another site that runs off of a Dell MD 3620i clustered array and we backup to a Dell r510 with 12x 2TB 7200k drives and I get 50-70MB/s, the backups at this cluster are larger and complete sooner than the ones with our VNX.

Im beginning to think that is is a problem with a CIFS share. On the slow performing cluster veeam states that the bottleneck is "Throttling" under summary. At the cluster that is performing well the bottleneck states "Target" which is what I would expect. I could test mounting the file pool as an NFS mount but I figured I would post here before jumping through too many hoops.

As far as the network is concerned, the VNX is connected to a Nexus5 while the dell MD is connected to a 3750 stack with 10gig FCoE.

My case number is: 00635182

Thanks in advance,
Dylan

[foggy]

Hello, Dylan.

Copying a file to the same mount our backups go to shows full 1gbps speeds (80-110MB/s). But Veeam backups crawl at 10-20 MB/s at most.

You should not take sequential I/O performance from copying files into account (unless you're comparing it with the initial job run). What backup mode do you use (forward or reverse incremental)?

On the slow performing cluster veeam states that the bottleneck is "Throttling" under summary.

That means that you've throttled the network connection between proxy and repository and it is currently the bottleneck.

[dzeleski]

You should not take sequential I/O performance from copying files into account (unless you're comparing it with the initial job run). What backup mode do you use (forward or reverse incremental)?

Agreed, but im not getting anywhere near the speeds I should be seeing. We run reverse incremental.

That means that you've throttled the network connection between proxy and repository and it is currently the bottleneck.

The thing is we dont have anything set under the repository for throttling.

Thanks for such a fast response.

[nefes]

The thing is we dont have anything set under the repository for throttling.

There are two ways to use throttling in job:
1. Repository throttling. It is set up in Repository wizard, Repository page ("Limit combined data ingestion rate to:") and applies to speed of writing to repository.
2. Network throttling. It is set up in Traffic Throttling wizard from main menu, and applies to network connection speed between certain hosts.

Could you please check both of that places?

[Gostev]

Agreed, but im not getting anywhere near the speeds I should be seeing. We run reverse incremental.

Also, how many spindles does this go to? Rule of thumb for reverse incremental performance is no more than 2-3 MB/s per spindle.