High CPU usage from syncer

[Keri MacNeil]

I just built an RP7420 from an ignite image of a N4000 (HPUX 11i) as part of a migration of the old server to the new server. The server has 8 CPUs, 16GB of ram of which 11GB are still available. The server runs an oracle database, but the server is essentially idle right now. Benchmark tests show good performance from the database prespective compared to the N4000.

As a last step before releasing it to the users, I'm running a make_net_recovery to our ignite server.

What normally takes less than 30 minutes has now taken more than two hours, and is only about 20% complete based on the size of the ignite file it's building. Looking at the server processes, /usr/sbin/syncer is pegging 7 of the 8 processors. What could be going on to cause this type of performance problem?

[James R. Ferguson]

Hi:

You don't differentiate between 11.11 and 11.23, but the first place I'd look is at your buffer cache settings. Buffer cache management is generally better in 11.23 compared to 11.11.

Do you have a large value for 'dbc_max_pct'? Since you are running (or will be running) a database, I'd make sure that the maximum size doesn't consume more than (perhaps) 5% or about 800MB of memory.

Regards!

...JRF...

[Keri MacNeil]

I can thank the boys in the networking group. The network port was hardcoded to 100MB Full duplex, and the NIC was configured to autonegotiate. It looks like it autonegotiated to 100 Half, and was creating all sorts of collisions. This apparently caused the syncer to work extremely hard to write to the NFS mounts. I changed the port settings to match the network setting, and no more problems.

[Keri MacNeil]

See comments above.