Help figuring out how to speed up a large Windows restore of millions of small files

[smccambly]

Hello Folks,

I've tried everything I know to speed up the performance of a restore we are attempting to a Windows file system than consists of 10 million folders and 20 million small files (5TB total).

The TSM server is 7.1.1.1 on Linux and not overly busy. The data is spread across 3 3592-E07 tapes, reasonably well collocated. Query session on the server most often shows SendW for the I/O thread sessions (there are two) and we are getting an aggregate speed of only around 10MB/sec.

I am restoring without NTFS permissions. I have set TxnByteLimit to 3098624K and TxnGroup to 65000.
Of course it keeps getting more complicated with the fact that the client is a VM writing to a NAS share (SAS disk based with an SSD index).

The CPU on the client never hits more than 50% and the 10GB of memory is only about 30% used.

Oh, and of course we have disabled the antivirus software.

Can anyone suggest somewhere else or something else I should be looking into to find what is limiting the performance?

Thanks in advance!
Scott

 

[marclant]

You can review this:
http://www-01.ibm.com/support/docview.wss?uid=swg21142185

In your case, you want a NQR restore (see technote).
Use a multi-session restore (resourceutil > 2)
SendW means the server has to wait for the network to deliver the data to the client, check with your network admin.