By no way am I an NFS expert, but ...
Simply put, a "cat /dev/zero > [home on nfs]/tmpfile" on a SRS
almost stalls the server. (ls takes >1s!)
It happens on any computer that has its NFS ressource unavailable or
saturated.
The following shows you're searching in the right direction !
Are we the only ones experiencing this?
You're the only one dealing with *your* users, at least ;)
Our users tend to work with very large files, so reading or writing
a couple of GB at a time does happen frequently enough to be a real
problem.
"Reading or writing a couple of GIGAbits" at a time has to take some
time, you know.
If some people are doing it concurrently, then it's no surprise
your NFS ressource is saturated.
Sometimes 1 + 1 + 1 does not equals 3, disk sharing is one of these
examples.
It may take 20 seconds to access a single file, and it could take 60
seconds if you access file A,
THEN file B, THEN file C, but it may take *much* more than 60 seconds
to access files (A+B+C)
at the same time, because the OS may just spend its time moving the
hard drive heads,
which is a "slow operation". These are not real numbers, but it reads
1024 bytes of file A, then moves
the heads to file B, reads 1024 bytes, move the heads to file C,
reads 1024 bytes, and so on.
What happens to you is to some extent the same symptom as a
fragmented disk, only that it's
the i/o time that is fragmented.
Thus your NFS ressource is no longer available.
And your SRS freezes.
We have a fairly simple setup: A single Solaris/ZFS based NFS
server with loads of disks and two Linux based SRS servers
mounting /home. Saturating the servers links using other clients
does not cause any problems, but once one client's link is full,
all other processes on that client wait ages on i/o.
Considering the number of users some of you put on one box, this
issue must have come up before. What am I doing wrong? How do you
all solve this? Or, does anyone have an idea how to fix it?
Does using an automounter help? Or NFS4? Or a different NFS server?
I don't know about the performance of automounter or NFSv4 vs NFSv3 -
though I'd like to! -,
but it would definitely help if you had a SAS disks server (faster
than SATA!), using 15k disks
(one third faster than the typical 10k disks, and twice as much fast
as 7200 rpm "budget" disks
found on many high-volume SAN/NAS/servers).
By the way, the Sun x4500 thumper uses 7200 rpm SATA disks ;)
Also, it's no workaround but you can have a look at NFS mounting
options, like rsize, wsize, syncronous
i/o and so on...
--
-- Sam
_______________________________________________
SunRay-Users mailing list
[email protected]
http://www.filibeto.org/mailman/listinfo/sunray-users
