On Tue, Dec 30, 2003 at 04:05:49PM -0600, Eric Sandeen wrote:
> can you try the same kernel on an ide disk? What's
> the result of that?
On the same system (where I have boot IDE disk) with XFS on IDE disk with 35GB
partition it seems to be ok.
> How big is the underlying filesystem? Please add
> the xfs_info output.
(1.75TB) (8x250GB disk in raid 5, stripe 64k)
meta-data=/mnt/export isize=256 agcount=32, agsize=13354352 blks
data = bsize=4096 blocks=427339008, imaxpct=25
= sunit=16 swidth=112 blks, unwritten=1
naming =version 2 bsize=4096
log =external bsize=4096 blocks=20000, version=1
= sectsz=512 sunit=0 blks
realtime =none extsz=65536 blocks=0, rtextents=0
> Apart from that if you could run strace to see where things stop,
> or better yet get kdb and backtrace the process. KDB is part
> of the xfs cvs tree.
strace shows only read and write call, nothing more. I do not know how to run
kdb while whole system freezes. No process runs (I run top -d 1 at the beginning
on the console and it freezes as well.) Kernel responds to ping request.
I remember that bdflush and some other daemon (kswapd I think) does high cpu
utilization (99%) just before it freezes.