I wrote:
A couple weeks ago, after we moved a user's home directory from a 5.7
box to 6.2, he came
to complain about slowness.
Unpack file from NFS-mounted directory to local disk, like /tmp, everything's normal. cd to the NFS-mounted directory, and unpack it
there, and it was
six to seven times slower. We repeated this a number of times, on a
number of machines. At
first, we thought it was an NFSv4 issue, and I filed a bug with RedHat
about it.
Then we did more testing; it seems to occur with 6.2; it's going to
happen if that's got ext4.
However, my manager saw something else, just today: he noticed one of
our nightly backup
servers was taking much, much longer to complete the rsync and delete.
Then he saw that I'd
updated that backup server to 6.2, and after the backups started running
again (probably
missing modules), they were taking a lot longer... about 6-7 times
longer. As in going from
between .5 and 1 hr, to 5.5 to 9 hours.
Well, I was googling on performance tuning NFS, and found http://www.softpanorama.org/Net/Application_layer/NFS/nfs_performance_tuning.shtml, where the author notes "One issue with migrating to NFSv4 is that all of the filesystems you export have to be located under a single top-level exported directory. This means you have to change your /etc/exports file and also use Linux bind mounts to mount the filesystems you wish to export under your single top-level NFSv4 exported directory."
I makes a *real* difference. I tested it, on one 6.2 server, mounting a foo directory I created in the usual 3 levels deep, and another, /tmp/foo1, and doing my same unpack ran at 2/3rds the time. Not good - 6.5 min vs. 1+min for 5.7 & NFS3, but better.
This alone suggests we need to go back to NFS 3 - it would be a Big Deal to relocate everyone's home directory.
Doing a yum list *nfs* shows something called sblim-cmpi-nfsv3.x86_64.
From the info, I can't tell if I can just drop it in and run that, or if
I'd have to yum remove nfs-utils & nfs-utils-lib. Anyone have a clue?
mark