2.4.4 works great -- most of the time.
I started with a 2.4.4 clean tree and applied the
linux-2.4.4-xfs-TEST.patch.
(I have a hard time using cvs to get the tree -- our firewall is
default deny)
My hardware is Tyan 1854, PIII 866, 3com 905b, 3ware 6200 ide
controller, ibm 75G ide disk.
My base OS is SUSE 7.1
Linux linux 2.4.4-xfs #1 SMP
NFS v2 turned on. NFS v3 turned off.
Stress Testing:
---------------
Linux box acting as NFS server for solaris 2.8 boxes.
3 Solaris 2.8 boxes running 'while(true);do bonnie -s $size;done' on a
nfs v2 mount of a 68G xfs filesystem. I run the $size value
differently on all boxes so the read/writes are fairly randomized.
100M, 500M, 50M...
The xfs box will run for almost 1 hr before it hangs. Will not respond
to ping. X-windows are hung. No virutal consoles available. Something
is not right. I have repeated this about 4 times in the last 12 hours.
Right now it appears to handle heavy nfs loads better than my
reiserfs/knfsd patched kernel while it is running, but the reiser
testing hasn't hung the box this bad.
Questions
---------
1. what debug can I do?
2. Is this a via chipset issue?
3. I'm using default args in mkfs.xfs and mount. Any suggestions for
changes there?
Thanks -- xfs looks pretty good right now.
eric
--
__________________________________________________________________
Eric T. Whiting AMI Semiconductors
|