On Fri, 2002-04-26 at 18:10, Mihai RUSU wrote:
>
> Hi guys
>
> Excuse me for dropping in but I wanted to say some things about this
> subject.
>
> We are using an XFS partition for a high I/O load (many IOPs, not too
> much transfer).
> The partition is 100Gb and its using its default log size (as per
> xfs_info) of 3200 blocks (4096 bytes for a block).
> Also I am monitoring (MRTG graphs) the IOPs as from /proc/stat | grep
> disk_io . Would it be right to consider from /proc/stat the firs value
> as the number of I/O transactions?
> Ex:
>
> disk_io: (3,0):(10899763,847269,16900622,10052494,159221833)
> (48,0):(52710982,15933971,287172411,36777011,545459600)
>
> Whould 52710982 be the number of I/O transactions of the device 48,0 ?
>
> If yes then would help very much to get statistics from them and maybe
> answer the question of a log of X size would lower the performance of XFS
> partition of Y size and Z transactions/s.
>
> PS: having a single XFS partition of device 48,0
>
> Thanks
>
> ----------------------------
> Mihai RUSU
>
> Disclaimer: Any views or opinions presented within this e-mail are solely
> those of the author and do not necessarily represent those of any company,
> unless otherwise specifically stated.
If you look in cmd/xfsmisc in the cvs tree you will find a script
called xfs_stats.pl, make this an executable and it will dump all
sorts of stats out of xfs. There is text further down the file which
explains what the numbers mean.
I have come across a command called iostat which you can get here:
http://perso.wanadoo.fr/sebastien.godard/
which will report io requests nicely.
Steve
|