lkml.org 
[lkml]   [2000]   [Mar]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: kernel performance & profiles under heavy disk I/O
Date
From
> 2.2.5 had significantly greater IO performance and this attachment is
> roughly what changed between 2.2.5 and 2.2.10 or so........
>
> This change has been demonstrated to clobber direct Block-IO with a 50-60%
> penality............just a thought.........

That would make sense for devices with a low block size. Note that part of
the change is a bug fix (clamping to end of media) but the rest of it subtly
changes the I/O pattern so that it wont be reading contiguous bursts of
read-ahead, but instead bursts of read-ahead+random which will mean that
on 1K block I/O you will issue more I/O requests on a dumb controller as
you will tend not to merge 4 requests/page each time.

Andre - if you change it so that the blocks queued to read is clamped
at readahead as before - but keep the overrun check

(ie

blocks = number_user_wants;
if(blocks < readahead)
blocks = readahead;
if(blocks >= the end of the disk)
clamp to end

)

do you get the performance back


-
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/

\
 
 \ /
  Last update: 2005-03-22 13:57    [W:0.061 / U:0.252 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site