Servers:
x4500-05:
X4500, Solaris 10 10/08 s10x_u6wos_07b: nfsd threads 388 / 1024,
load 1.70, 1.66, 1.58
x4500-10:
X4540, Open Solaris svn117: nfsd 16/1024,
load 0.05, 0.06, 0.07
na05:
netapp FAS940: 6.5.2R1. Under medium load.
Clients:
prov01: intel Solaris 10 1/06 s10x_u1wos_19a,
load: 0.08, 0.07, 0.11
ftp04: intel Solaris 10 10/08 s10x_u6wos_07b,
load: 0.04, 0.03, 0.02
All machines are currently live, but fairly low usage at this time. The
netapps are very old, and being phased out.
If I take a tarball of something, in my case MTOS-4.261-ja.tar.gz and
attempt to untar it, I see the following times:
prov01:/# mount -o vers=4,hard,intr,quota x4500-05.unix:/export/sd01/www
/mnt
prov01:/# mkdir /mnt/.test4 && cd /mnt/.test4
prov01:/mnt/.test4# time gtar -zxf /var/tmp/MTOS-4.261-ja.tar.gz
prov01 to na05: 0m11.114s (GOOD)
prov01 to x4500-10: 5m11.654s (vers=4)
prov01 to x4500-05: 8m55.911s (vers=3)
prov01 to x4500-05: 10m32.629s (vers=4)
ftp04 to x4500-10: 3m59.203s (vers=4)
I would feel confident that it's not an ZFS issue (hence I post here) as
I get:
on x4500-10: 0m0.478s (GREAT)
on x4500-05: 0m1.361s (GREAT)
(On the same volume, but under a new directory, no deletions; etc).
All servers and clients are using 1000MB, duplex=2 everywhere. In fact,
scping the tarball to the server is as fast as you would expect.
(1 second or so, with the ssh challenge)
snooping has not told me much of anything. A lot of setattr calls etc. I
have tested 2 different NFSD servers, 2 different NFS clients, that the
zpool itself is very fast, that the network appears fast. I feel I have
ruled out that is not just one server, or one client, that is at fault.
What would be my next step in trying to work out why it takes more than
5 minutes to untar a 4MB tarball?
Any hints would be appreciated!
Lund
--
Jorgen Lundman | <lundman at lundman.net>
Unix Administrator | +81 (0)3 -5456-2687 ext 1017 (work)
Shibuya-ku, Tokyo | +81 (0)90-5578-8500 (cell)
Japan | +81 (0)3 -3375-1767 (home)
