What throughput speeds should I expect?

[aman4God]

I am new to Veeam, but not to backup. I have a UCS environment with 25 ESX hosts running 5.5U2, on two separate UCS stacks. Both Stacks have (4) 10Gb connections to a 5596 nexus. We have a VMAX 20k with (8) 8Gb FC assigned to the storage group for UCS in the same NX switch. I also have a CS500 FC Nimble with (4) 8Gb FC in the same switch for my Repositories. I currently have 4 repositories configured and connected to 4 individual Win2012 servers. When I run backups, I can't get more than an aggregate of 250 - 260 MB/s for all the backups. The backup jobs all say the source is the bottleneck. Everything is internal to the same VM environment and interconnected within UCS at 10Gb. I feel like I should be able to get way more than 250MB/s for the entire environment considering the backend of the systems. Any help would be appreciated. I am running v8 with the latest patch.

[Vitaliy S.]

What transport mode do you use for your backup jobs, direct SAN access or network?

[aman4God]

It is set to automatic, but runs in appliance mode most of the time. There are a couple of jobs that report dropping to Network mode, but from what I can tell network mode is providing better throughput even though according to Veeam documentation Network mode is not preferred. We also don't have any physical proxies or repositories so we can't run in SAN mode from what I understand.

[Vitaliy S.]

If you have decent network connection, then network mode can work much faster than hotadd. Our documentation is referring to 1 Gb links. Do you see the same performance for network and hot add modes? The reason I ask this is that network management interface on the ESXi host is throttled, so if you have a network mode in use and the bottleneck is source, then direct SAN mode would give you better performance rates.

BTW, if you look through job session stats, what operation takes most of the time? Is it VM processing or something else?

[aman4God]

Currently I am running a test with a single system to determine optimal backups times. The first was network transport mode, tonight I will use the automatic config.

After looking through the session stats it seems that always the VM processing is taking most of the time. I recently moved the repository block size from the windows default to 64k and that seemed to increase speed some. For direct SAN mode though, don't I need a physical proxy that is zoned in the fabric with the storage? All my backups are coming off a VMAX through the UCS/Vmware environment onto the Nimble. Alls are connected through the same fabric, but we are 99% physical and cannot add a physical proxy.

[Vitaliy S.]

For direct SAN mode though, don't I need a physical proxy that is zoned in the fabric with the storage?

Yes, with FC connection, you should have physical proxy server and HBA adapter plugged into it.

[emachabert]

You should also verify your network configuration, such as DCB, QoS and PFC parameters.
Perhaps you have some bandwidth throttling that occurs during backups to maintain FCoE performance.

Are you talking of active full speed or incremental speed ?
In active full mode, you should saturate a 8gb/s link with no problem with your configuration.