Subject: nfsd loading
To: None <tech-kern@netbsd.org>
From: None <rick@snowhite.cis.uoguelph.ca>
List: tech-kern
Date: 06/04/1999 10:40:08
Hi,

I've noticed that on NetBSD machines that the CPU usage by "nfsd"s seems
to be "non-averaging", compared to say an Ultrix NFS server.  Compare:

[stuff clipped]

Does this actually make a difference in the grand scheme of things?  I
guess we're headed towards using kernel threads for nfs anyway...

Nope, no real effect. Each kernel process/thread is identical. The one picks
up requests, if it is busy the next one picks up requests and so on...
(If they all end up doing a fair amount of work, you need more of them, since
 the last one should be almost always idle. If the last ten are almost always
 idle, you won't do any harm reducing the number of them, although the
 savings will be minimal.)

An nfsd is basically a kernel thread, since it never comes back to user land
unless it is killed. (There was an exception to this, if you used the
kerberized code, but that has been stuck unfinished for eons, so...)

Have fun with it, rick