Question regarding traffic behaviour in HPV 2012

[Eson]

Ive read up a bit on the changes in HPV 2012 with regards to redirected access during backups etc. I was under the impression that we had moved passed all the traffic flowing across the CSV-NIC.

For example, if I backup a server in Veeam that resides on a CSV that the host currently isnt the owner of all the traffic to Veeam passes through the CSV-NIC. Is it supposed to be like this and the only improvement for 2012 is that the disk isnt actually in Redirected Access during the backup so that regular traffic can flow as usual with direct access via Fibre Channel?

My setup: 3 x HP DL380G8 server in a HPV Cluster with FC to a HP P6350 SAN. Veeam is installed on a separate G8-server but Im running on-host backups.

[veremin]

is that the disk isnt actually in Redirected Access during the backup

It depends on which VSS provider is being used by VB&R instance. If it’s a hardware one, the CSV will stay in the redirected mode only during the snapshot creation process; once the snapshot is taken, the CSV will return to Direct access mode.

In case of software provider, the CSV will stay in redirected mode till the backup process is finished.

Thanks.

[Eson]

Im using software VSS.

What you are writing was the case for 2008 R2 but im pretty sure this has changed in 2012 by using among other things Oplocks. Check out this video:

http://channel9.msdn.com/Events/TechEd/ ... 012/WSV430

Its a bit long but check from about 27mins in.

Also check out this link for a summary:

http://www.aidanfinn.com/?p=12844

As I said in my post the CSV is not in redirected access during the backup, by this I mean it doesnt have that status when you check the disk overview. The traffic is going over ethernet though. The reason Im questioning this is Im pretty sure when I had just installed Veeam a couple of months ago for tests it did not behave in this way and no backup traffic was going over ethernet.

[chrisdearden]

I bumped into Aidan at the weekend and he confirmed this - the IO should be direct even during snapshot. Are you making use of the hardware VSSprovider? If you right click the hosts , you can see which providers each CSV is using ( if you have multiple ones installed )

[Eson]

Im using software, Microsoft CSV Shadow Copy Provider, we took the decision to not purchase the license for our HP SAN based on the fact that Server 2012 wouldnt keep the volumes in redirected state. And as I said the volumes themselves arent in redirected state but all the traffic when a host doesnt own the CSV passes through the network adapters assigned for CSV-traffic.

[chrisdearden]

I think this might be one to take up with support - I'd also be interested to know!

[Eson]

Ok, I might do that, I tried this at another customers cluster with a P2000 with SAS, same behavior. Im afraid though that this might be how things work.

[chrisdearden]

Its certainly not by design - It may be possible that you've unearthed a deeper issue. All of the research I've done indicated that hosts not owning the CSV should use direct access after the snap has been taken ( with 2012 ) If you could post the case ID up, or drop it to me in a PM it would be very useful as I'd like to track the case.

[davida]

Same problem here and it been like that since the first day of having Hyper-V 2012 and veeam for backup.

My case id is 00253912 if you want to follow up as well.

[foggy]

Our QC has confirmed this behavior to be the case (traffic going through the CSV NIC while backing up VM residing on non-coordinator host using software VSS provider). That said, in Veeam B&R v7 the agent will be always started on the coordinator node so the issue will be eliminated.

This seems to be a good scenario to test in the upcoming v7 beta, so it will be much appreciated if you guys could reach out to your Veeam sales rep to apply (we usually keep betas limited for a number of reasons).

[Eson]

It seems like when traffic is going through a host not owning the disk all the read data gets cached in RAM somehow and eventually leads to the host running out of available memory and the backup of the machine fails cause the host cant reach the CSVs any longer.

[veremin]

Emanuel, It will be much appreciated, if you open a ticket with our support team and post it here, since we are more than willing to get all relevant information regarding issue. As mentioned above, you might also want to contact your sales representative in order to apply to limited beta testing and see whether the aforesaid problem is fixed there.

Thanks.