Backup using TSM (Tivoli Storage Manager): 100% CPU load (W/O soft-compression)

[Victor M. Gomez Hornill]

Hi,

At this moment we are using OmniBack 4.10 for backing up an Oracle SAP R/3 database.
SAP R/3 and Oracle run under HP-UX 11.11.
We usually use brbackup (SAP tool) with OBII backint for backing up the database.

For bussiness reasons, we are evaluating TSM backint (Tivoli Data Protection for R/3) for doing our backups in a short future.
This test is done throught our SAN (LAN free backup, using TSM StorageAgent) to an IBM LT0 (ULT3580-TD1) device.

The results in a ONE CPU server (N4000-550MHz) were:

OmniBack (backint)--> 90GB/h --> 30-35% CPU ( backint + sapback+ bma CPU load)

TSM (backint)-------> 52GB/h --> 95-100% CPU ( backint + dsmsta CPU load)

I have done those two kind (OBII and TSM) of tests done using the same scenario. I mean, the source files an the destination device were the same. In neither of the cases sofware compression was used.

In TSM tests we have changed some values as: Multiplexing (from 1 to 6), Sessions (1 and 2), all of them with similar results (100% CPU load). TSM Server is runnning in another host (AIX), not in HP-UX server.

IBM says that this problem is not a TSM one in HP-UX platform, but state that this behaviour (100% CPU load) happens only in our server (may be for HP-UX kernel configuration, etc).


Does some of you use "TSM Storage Agent" (LAN free backup) under HP-UX?
Does some of you use together "TSM TDP for R/3" (backint) and "Storage Agent" (dsmsta) under HP-UX?
If yes, please let me know your background about the above.

Best Regards.

Victor

[Eckert_2]

Hi Victor,
it seems I am in the same situation.
Did you get any response?
How did you solve the problem?

Best Regards,
Torsten.

[Victor M. Gomez Hornill]

Hi Torsten,

Unfortunatelly we were not given any solution from IBM technical support. We have found a configuration option ("Validate Protocol") and we have checked that when it is disabled, our HP-UX TSM client CPU usage decreases a bit. This option is related to CRC check of backup data. But even when the CRC check is disabled, in our installation TSM client used the entire CPU in a single processor N4000 server. In a four-processor N4000 server, I saw that TSM consumed 350% of aggregated processing power (3.5 CPUs).

Finally, IBM recognized this behaviour (such a big CPU use!) as "normal" taking in account the TSM architecture. That is, the data is transfered to several buffers (secuencially) before writing finally to the tape (even in "LAN free" backup).

Because all the above problems, we desisted on the change of our backup tool, from OmniBack II 4.1 to TSM. We have been using OB II for years without relevant problems and we are planning to upgrade to Data Protector 5.5 within this year.

Be aware that our information and experience with TSM may be nowadays a bit outdated and perhaps your problem may be not the same as ours.
I don't know if IBM has made some improvement in the TSM for HP-UX in the last two years.

Best regards,
Victor