System slow, syncer responsible?

[dictum9]

One of the users complained that the system (11.23 Itanium, rx5670) was very slow, his commands were hanging. He says he noticed syncer was consuming 99% of the CPU cycles. Is it possible for syncer to cause this? The user says the slowness is on and off.

I noticed syncer in the process table, now it's back to normal. I couldn't figure out what started it, I thought it ran only once at start up time. It also doesn't appear to be in cron jobs.

The other idea what that the user was accessing a NFS filesystem... But I didn't see any NFS mounted filesystems, so that's not a problem.

[Yang Qin_1 · Accepted solution]

It looks like a LAN configuration problem. Run "lanadmin -x 0(or 1, 2)" and see how it is configured and then check with you network engineer to see it match what they set for your LAN connection or not.

[Steven E. Protter]

Shalom,

Seems there might be a network response issue here.

You need to look at the top command yourself as an admin or run performance stats rather than rely on the user to say what the information is.

http://www.hpux.ws/system.perf.sh

This may help.

SEP

[James R. Ferguson]

Hi:

The 'syncer' daemon runs every 30-seconds or so and is responsible for flushing modified file buffers from the Unix buffer cache. Its performance has been improved on 11.23 relative to prior releases. The number of modified buffers present in the buffer cache will affect the amount of work each 'syncer' pass has to perform. A buffer cache constituting about 5% of overall memory, particularly on a multi-CPU platform, should not mean that users will see any "pulsations" in performance.

If the filesystems in question are NFS ones, I'd look to your network settings. In particular, is your network switch configurated to match your LAN? If you are running a gibabit connnection, make sure that you have auto-negotiation enabled at both points. If you are running 100MB connections, you generally want to set full-duplex and 100MB speed at both points.

Regards!

...JRF...

[Bill Hassell]

The syncer is used to flush writes from the buffer cache to the disk, so if your system is continuously writing, the syncer will consume a lot of time. It is particularly efficient with more than 3 CPUs, but in order to consume a huge amount of time, the disks would have to be seriously overloaded. Your disk array cache setting may need major adjusting, or you have some disks going bad. And if the writes are going to an NFS disk, this would be quite normal as NFS disks are extremely slow.

As far as commands hanging, excessive syncer time would not affect command unless the commands are also on an NFS disk or the commands are struggling to access data on an NFS disk. Note that if you are running automounter, you won't see the NFS disks until they are used.