From: Goetzman, Dan (Dan_Goetzman@bmc.com)
Date: Thu Aug 01 2002 - 11:29:19 EDT
From a systems admin point of view...
I am in the process of setting up long term perfomance monitoring and
modeling on my 5.1A TruCluster system. The cluster is used as a large scale
highly available file server and I wanted to view the amount of system
resources used by both NFS and CIFS/SMB file serving (via ASU). The ASU part
was easy, I just had to identify all the processes devoted to that task.
NFS seems a bit more difficult to track. The nfsd daemons do the file
serving, but it appears that they use "kernel threads" and the system kernel
threads are run under the system process "kernel idle", as are all the other
system kernel threads running on the system, nfs related or not. The nfsd
daemons appear to do little work. So when I have a large nfs file serving
load the performance monitoring (you could even use something like "top" to
see this) will show the "kernel idle" process as using a fair amount of CPU,
not the nfsd daemons.
My question, how to account for the kernel thread activity and track it
back to the "application" (NFS in my case)? Using normal tools and not
having to manually look at the kernel threads via "ps axml" manually.
I know you can see the kernel threads by using a few Tru64 options to ps
(like "ps axml"), but how about typical system performance tools like top,
etc?
TIA...
Dan Goetzman
This archive was generated by hypermail 2.1.7 : Sat Apr 12 2008 - 10:48:48 EDT