In the vain attempt to troubleshoot why our backups are so crappy for a particular volume, we've found some interesting results:

OES2 SP1 fully patched (this would be the "client" server)
NSS volume with lots of tiny files.
Same volume data was copied to a Reiserfs partition on the same server
Same data was also copied to a Sun Solaris server using ZFS.

Both servers (the Solaris box and the OES box) are using Fibre Channel to a Xiotech Magnitude SAN using 2GB connections. NIC in both are gigabit.

The software being used to backup?

Symantec NetBackup which is running a master server and 4 media servers all on Sun Solaris. On the OES2 server we're using the native Linux agent (as we should). there is NO TSAFS involved anywhere because we're not using the NetWare client to backup the OES2 server. 100% pure linux as it were (from NetBackup point of view)

Here's where it gets interesting:

NSS volume full backup
116,803,954 (basically 116 GB)
3,968,844 files
Time to backup across the LAN to tape: 10 hours, 25 minutes

Reiserfs volume full backup
113,454,828 (basically 113 GB)
3,968,834 files (NSS has a bit more because there's other files it puts there)
Time to backup across the LAN to tape: 5 hours, 45 minutes

ZFS volume full backup
Same basic stats (I'm not a Unix admin, so I dont' have direct access to the Solaris server).
Time to backup "local" (the server the disk is on also contains the media server, so I'd expect this to be a bit quicker)
3 hours and 30 minutes to backup

Why is NSS so darn slow on a full backup, when the same data on the same server, just on a diff. file system is significantly faster?

Differential backups are even worse (instead of twice as slow, the NSS volume for diff. is about 3-4x slower than the data on the Reiserfs "volume").

I have no other backup software to test against, so I'm wondering if this is something specific to NetBackup or if it's NSS

Has anyone elase performed similar testing (my guess is that it's specific to large amounts of small files). We have no other file system (NetWare, Windows, Solaris, Linux) that has any similar data to compare against (ie, other than this particular volume).

I have other NetWare volumes with significantly more files (11 million), although the average size is a bit large and they backup just fine. The Solaris servers (80+) don't have any existing data sets with that many millions of small files and neither do our Windows servers, so I don't know if they all suffer the same fate.


I'd open an SR with Novell but they'll just close it right away and tell me I have to call Symantec and then I have to try to convince Symatnec to open a TSANET call with Novell.