Subject: Re: NFS mounts
To: None <,,>
From: C Kane <>
List: current-users
Date: 12/17/1998 13:56:51
Frank van der Linden wrote:
> On Wed, Dec 16, 1998 at 06:11:10PM -0800, C Kane wrote:
> > Doing an NFS mount under NetBSD-1.3I/i386 seems to be rather slow.
> > It's certainly somewhat slower than HP-UX but Linux performs
> > best of all.  
> That's strange, because it's the exact opposite of what I have seen
> on various systems. How are you measuring the performance?

I mean the time required to do a "mount -a".  It's several times slower
than the HP-UX system.  My HP-UX system takes 4:06, and NetBSD takes
30:08 to do the same mounts from the same servers on the same network.

The NetBSD system is a dual 200MHz Pentium Pro system with 256M memory.
The HP-UX system is a dual 120MHz J210.

> > On a separate but related note, NetBSD uses a reserved socket port
> > by default for NFS mounts.  Is this the default on commercial UNIXes
> > as well?  Does each mount point get its own port?
> Each mount point gets its own socket/port. This does indeed give
> a limitation, as there are a limited number of reserved ports
> available (< 1024). Then again, why would you want to use so many
> mounts at the same time?

Let's just say I wanted to test the limits of the OS.

HP-UX doesn't seem to have this limitation, so it's apparently not
using reserved sockets.  Do any commercial UNIXes use reserved
sockets for NFS?

This change means I am unable to copy the NFS portion of the fstab
from the HP-UX system to the NetBSD system, because NetBSD will
require a flag to not use a reserved port.  I'm using "-p" now in
the options field; is there a word I could use instead?

The more things stay the same, the easier they are to manage.

-- Chuck