fuser commnad slow on v class

[Joseph Hoh]

Has anyone seen where a commnad such as fuser is very slow on a V-class system. If I run "fuser *" on any directory it is very slow on my V-class servers. If I run the same command on a E-class w/ 11.0 it runs very fast or if I run it on a K-class w/ 10.20 it is very fast. Any ideas?

[Dan Hetzel]

Hi Jeff,

Does it make any difference if you run 'fuser .' instead of 'fuser *' ?

Could the argument list be a lot longer on the V class than on your other systems?

If you're on NFS filesystem, did you apply the latest ONC/NFS performance patch ?

Best regards,

Dan

[Joseph Hoh]

I tried "fuser /etc/passwd". On the V it pauses for a few seconds and then comes back with no processes. On the E it comes back instantlly. All the files systems are local. If I do the same command on my SUN servers they run very fast.

[Les Schuettpelz]

Yes, we're very familiar with how fuser behaves on V-class, it has been like that from the beginning. Any script that used 'fuser' to see if a filesystem could be unmounted started running much longer. Our V-class are all similar configurations, so we can't narrow it to V-class with specific adapters. It may have something to do with the crossbar and processor/cache architecture. We haven't noticed that running fuser impacts performance of other processes, but that could be a concern.

[Les Schuettpelz]

Also, the recommended fuser patch, PHCO_21901, has been installed on our V-class servers and doesn't provide much if any improvement. It is based on PHKL_18543 and requires PHKL_19186 which has been superseded, PHKL_21700 is 'best patch' and PHKL_22588 is most-current.

[Jitendra_1]

check your PATH . It might be searching for a while to find fuser . Still better try absolute path to fuser like /usr/sbin/fuser /etc/passwd.

[Royal]

This problem is not specific to V-class, NFS, directory length, or where the fuser command exists in your path.

At the same time we rolled out V-class early last year, we also upgraded all of our D,K,and T-class to 11.0

The problem with fuser running excruciatingly slow appeared at that time. We see it as well on N and L class
systems running 11.0 . NFS or local, same slow response. Absolute path to fuser, same response. Fuser to a file or filesystem, single or wildcard, same response. It's ugly.

It's an 11.x issue, not a system architecture issue. Has anyone found a specific patch or bundle which as alleviated or eliminated the problem entirely ?

Thanks