Slow SQL Application Restores

[mmcd]

Hi All,

I am having a great experience using Veeam but seem to have run into a problem when testing a restore of a SQL DB to server SQL01.

[*]The DB is large, 400GB in total.
[*]The mount point is set as the Veeam Backup and Replication Server (also the backup repository).
[*]It is originally on a server which is 2TB (backup size is around 400GB in repository).
[*]Random I/O reads on the backup file return around 70MB/sec with dskspd.
[*]I have configured the DB files to restore to 3 of the drives in the new server.
[*]When restoring, the VeeamFLR folder on the SQL01 server is populated with the mappings to the backed up server drives.
[*]Read speeds are quite slow during the restore, around 20MB/sec and response times are 400ms when reading on the drives mapped back to the backup repository.
[*]Respone times back on the Veeam backup repository are 2-4ms when I monitor the backup file being read.
[*]Also, it looks like Veeam is not restoring the files in parallel also, seems to start with one DB file, then the next. Is this normal?
[*]During the restore I can copy data to and from the SQL01 server from the Veeam backup repository at over 100MB/sec

Has anyone else ran into such issues? Would be great to hear some of your thoughts.

Martin

[PTide]

Hi,

[*]The mount point is set as the Veeam Backup and Replication Server (also the backup repository).

What's the connection between the mount server (repository) and ESXi host? What's the bandwidth of a management interface on your ESXI? Also what kind of repo do you use?

Thanks

[mmcd]

Hi,

Thanks for the reply.

The Veeam B&R Server (Mount Server) is a physical unit, connected at 10G to a switch stack.
I am just using the "Default Backup Repository". It is on decent storage, as I said above, DSKSPD reading random I/O on the backup file was around 70MB/sec
The ESXi hosts are on a separate VLAN on the same switch, also at 10G.
Port 902 is the only port open to the ESXi hosts from the Veeam B&R Server, could this be the issue?
When I monitor on the guest OS I am restoring to, during the restore I can see an iSCSI connection back to the VB&R Server on port 3260. Assume this is the mappings located within VeeamFLR folder. Once it starts reading these files, the speed is poor and the response times on read operations are around 400ms.

See below.

Code: Select all

netstat -ano |findstr XXX.201
  TCP    192.168.XXX.151:445    192.168.XXX.201:52384  ESTABLISHED     4
  TCP    192.168.XXX.151:1433   192.168.XXX.201:52304  ESTABLISHED     1368
  TCP    192.168.XXX.151:1433   192.168.XXX.201:52385  ESTABLISHED     1368
  TCP    192.168.XXX.151:6160   192.168.XXX.201:52334  ESTABLISHED     173168
  TCP    192.168.XXX.151:6160   192.168.XXX.201:52335  ESTABLISHED     173168
  TCP    192.168.XXX.151:62917  192.168.XXX.201:3260   TIME_WAIT       0
  TCP    192.168.XXX.151:62923  192.168.XXX.201:3260   TIME_WAIT       0
  TCP    192.168.XXX.151:62924  192.168.XXX.201:3260   ESTABLISHED     904

[PTide]

Please contact support team as they have a fix that might improve the iSCSI performance. Kinldy post your case ID once you get one. Looking forward to get a feedback from you.

Thanks

[mmcd]

Thank you, I opened a support case - #01839781

[mmcd]

Hi,

I was on a call with Veeam support and they suggested installing the Veeam Console on the target server, as in the server I am restoring to and executing the job from there. This greatly increased my restore, by around a factor of 5. Currently running at around 110MB/sec. Same speed as my dskspd test ran at on the backup file.

This is not ideal but seems the only option. The Veeam support tech didnt seem to know of any patch for this. Hopefully its something they remedy in upcoming versions.