Re: Time for a pre-patch-2.0.31-3?
Richard Waltham (dormouse@farsrobt.demon.co.uk)
Thu, 3 Jul 1997 18:59:29 +0100 (BST)
Dr. Werner Fink wrote:
>
> > The drop in read performance occurs only when Werner's patch is included on
> > top of pre-patch-2.0.31-2. Without Werner's patch pre-patch-2.0.31-2
> > performance is similar to 2.0.29, 2.0.30.
> >
> > And as you've CC'd this to Leonard this is including a scsi patch supplied recently
> > by Leonard.
> >
> > Richard
>
>
> Please try the following change:
>
> -------------------------------------------------------------------------------
> --- linux/mm/filemap.c.old Thu Jul 3 13:31:01 1997
> +++ linux/mm/filemap.c Thu Jul 3 13:31:39 1997
> @@ -191,7 +191,7 @@
> default:
> /* more than one users: we can't throw it away */
> /* if we visit a non buffer page we should age it */
> - if (!bh)
> + if (!bh && !test_bit(PG_referenced, &page->flags))
> age_page(page);
> /* fall through */
> case 0:
> -------------------------------------------------------------------------------
>
>
> If this will not help, then simply remove the statment
> `if (!bh) age_page(page);'
>
> and test the interactive performance under X Window System under
> high load.
>
>
>
> Werner
>
Sorry, Werner, but I do not have X running on my system, however, neither
suggestion makes any difference to the disk performance with bonnie.
The only comment I can make is that during reading there is a lot of disk
seek activity. This is no doubt what is reducing the performance.
I am trying to get some profiling info but profiling doesn't seem to want to
play at the moment.
Richard
--
Richard Waltham | Work: richard@digtalinterface.demon.co.uk
At home in | Home: dormouse@farsrobt.demon.co.uk
Southampton UK | Compuserve 100421.1276