On 18 May 2000, Dimitris Michailidis wrote:
> Andre Hedrick <andre@linux-ide.org> writes:
>
> > On Thu, 18 May 2000, Alan Cox wrote:
> >
> > > Find out what has ruined disk I/O throughput.
> >
> > Possible combination elevator conflict with disk hardware.
> > If internal device write buffer cache sorting is enabled, you have double
> > labor for a single task. The rules for the elevator may conflict with the
> > drives look-ahead/read-ahead tables on the drive (OEM rules apply).
>
> One of the sources of poor performance under moderate/heavy disk I/O is the
> current management of the request pool. It is exhausted quite easily and it
I just cleaned up part of the request_queue setup in ATA, but until I
change it to device based queue there will be some goofys.
> causes an insane number of sleeps/wake-ups. I have seen situations where
> each attempt to acquire a request structure loops through sleep/wake-up an
> average of 10 times. Each time we wake up we have to scan the whole pool
> linearly, destroying the data cache in the process, and chances are we will
> fail and will need to do it all over. The elevator has only exacerbated the
> situation, with default settings at least.
>
> --
> Dimitris Michailidis dimitris@engr.sgi.com
>
Andre Hedrick
The Linux ATA/IDE guy
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue May 23 2000 - 21:00:16 EST