xfs
[Top] [All Lists]

Re: Performance problem with xfs

To: Danny Cox <danscox@xxxxxxxxxxxxxx>
Subject: Re: Performance problem with xfs
From: Feizhou <feizhou@xxxxxxxxxxxxx>
Date: Sat, 28 Feb 2004 15:05:42 +0800
Cc: XFS Mailing List <linux-xfs@xxxxxxxxxxx>
In-reply-to: <1077894047.25678.7.camel@pip>
References: <403F1934.3020701@xxxxxxxxxxxxx> <1077894047.25678.7.camel@pip>
Sender: linux-xfs-bounce@xxxxxxxxxxx
User-agent: Mozilla Thunderbird 0.5 (X11/20040208)
Danny Cox wrote:
Feizhou,

On Fri, 2004-02-27 at 05:17, Feizhou wrote:

I am having performance issues with a box running 3ware 0+1 mode where one of the filesystems is xfs based.

xfs_ili and xfs_inode are the values that change

 xfs_bmap /var/catchall/bulkdb/
/var/catchall/bulkdb/:
        0: [0..7]: 42235064..42235071


        Well, you didn't say what version of the kernel, so we can't figure out
if your XFS is very old or no.  That might be a clue.

XFS 1.3.0

The latest rh 2.4.20-30 kernels at
http://atrpms.physik.fu-berlin.de/


Also, kswapd chews substantial cpu.


        That's a clue right there.  If you're already swapping, NO filesystem
will provide much performance.  Can you try installing more memory?  You
might try running vmstat for a bit (vmstat 5 5 or vmstat 5 10), and
check out the "si" and "so" columns.  These are the "swap in" and "swap
out" stats for the past N seconds (5 in this case).  If they're anything
but "0", then you're swapping.

Sorry, wrong. Zero swapping. si and so are all zero. bi and bo are not very high either (under 1000 usually) so I/O bottleneck it is not. There are always a number of blocked processes which are mostly related to the xfs filesystem.

        Any others with insight?



<Prev in Thread] Current Thread [Next in Thread>