DB2 backup performance issues

Rigido

ADSM.ORG Senior Member
Joined
Apr 21, 2006
Messages
151
Reaction score
7
Points
0
Location
Rome, Italy
Hi,
probably I miss some DB2 info and the dba is not useful.
With a configuration like this: SAP on DB2, LPAR with AIX, LAN 10Gbit, SAN 8Gbit, V7000 and 7620 VTL, TSM 7.1.7, native DB2 backup commands, we can't go more than 200MB/sec. I tried with LAN, LAN-Free SharedMemory and LAN-Free TCPIP commethod. I still didn't try a "plain" dsmc backup but I'll do it tomorrow.
I read that I could specify buffers size, buffers number and parallelism parameters on the backup db command line, but I can't find any info or best practices, just the DB2 manual that states "DB2 will automatically choose an optimal value" :confused:

Does someone have any hint on this?

PS: With SAP on Oracle db I never found a dba that used just RMAN to backup database, they always told me that there were "other files" SAP needed to save for a reliable backup, so always used brbackup. This DB2 dba looks confident that this is how a SAP/DB2 backup must be done but, if so, why should IBM offer a TDP for ERP on DB2?

Thanks.
 
Is the DB2 LPAR environment sharing resources with any other LPAR - using the hypervisor?

If so, this maybe your issue. If you can move the LPAR to have direct access to the back plane, do so. You may be surprised to see a performance boost.

I never had issues using native DB2-to-TSM backup on standalone servers. Backup was fast.
 
Hi

I think narrowing down the problem would be better at this point. Here are some things I would check/test.

Storage capabilites of the DB2 system
Tablespace distribution across LUNs
Balancing of tables across tablespaces
Test backup to local Null device
Jumbo frames in use?
Test network speed from DB2 server to null device on TSM server (DD or similar)
Speed of receiving and writing data by TSM

We found that there were improvements to be had through adjusting the backup settings but by far the best gains for us were at the storage/filesystem/tablespace level and balancing out the load across multiple LUNs, filesystems and database containers.

Also worth tuning the networking. AIX default settings do not make the best use of 10G networking.

Haven't found much difference between going through VIO compared to physical however each setup is different.

That said, I think the best place to start is methodical testing from beginning to end to find the choke point.

Cheers
 
Thank you all for the support but we think the bottleneck is the 7620 VTL.
 
Back
Top