Okay David,
I'm taking another stab at answering your question about mount directories
onto directories. Yes, the NFS idea seems to work. That is, when mounting
an NFS export doing "ls .." shows the proper parent directory listing.
I thought that there might be a chance that the performance would be at least
decent, after all, just because it NFS doesn't mean that it's really having
to access the filesystem over the network, just through the loopback device.
So, I used bonnie, a little benchmarking tool to check things out. First,
I ran bonnie on the my fileystem. Then I mounted part of the filesystem and
ran bonnie on the mounted portion, so that it was reading, writing, and
seeking through NFS. I was not please with the results.
Here they are...
**Test on my root filesystem...
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
100 2785 75.9 3807 46.9 1702 49.2 2541 92.0 4367 90.0 45.0 1.8
**Using an NFS mount
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
100 709 28.8 837 15.7 392 14.9 751 26.1 860 15.6 30.9 7.0
I'm looking at the K/sec readings when I say that I'm dissatisfied. Of
course, all
of the CPU %ages are lower. But I'm wondering if that may be skewed somehow.
Like, maybe the program can only measure to CPU usage "on the client side" of
the NFS
"wall," and that on the host side, there is still even more CPU usage, the
overhead of the NFS server. That seems a little strange at first, probably
because both client
and server are running with the same CPU. But imagine if I'd run bonnie from
an remote NFS client machine. Then do you think that the CPU results would be
the clients
or the hosts? or both? I'd be inclined to say that the client isn't going to
have the privilege of such information, being an entirely different box. So
it seems feasible to think that the same would happen with client and server
sharing a CPU.
But....I'm so completely stabbing in the dark on this one. Making a purely
logical conjecture so as to account for the differences in the two reports.
If any one
knows for sure, I'd love to know too!
--
======================== Mike Wilkerson ==========================
"You cannot go on 'seeing through' things forever. The whole point
of seeing through something is to see something through it."
C.S. Lewis, "The Abolition of Man"
PGP Fingerprint->
1E 5F 0F 39 54 28 73 4F 27 7C 48 69 E6 0D E9 3A
==================== [EMAIL PROTECTED] =====================
--
PLEASE read the Red Hat FAQ, Tips, Errata and the MAILING LIST ARCHIVES!
http://www.redhat.com/RedHat-FAQ /RedHat-Errata /RedHat-Tips /mailing-lists
To unsubscribe: mail [EMAIL PROTECTED] with
"unsubscribe" as the Subject.
