lkml.org 
[lkml]   [2005]   [Sep]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: kernel 2.6.13 buffer strangeness
Based on a better understanding of dirty_ratio and dirty_background ratio (thanks Nate) I just tried
the following test:

dirty_ratio set to 95
dirty_background_ratio set to 1

From Nate's description of these parameters, this should mean that the disk writes
start almost immediately, and the kernel will allow 95% of RAM to become dirty before
applying the throttle.

Ok, so with 25Mbytes/s coming in, and 17Mbytes/sec going out to disk, the dirty pages should be growing
at 7Mbytes/sec. With these parameters set as above I should see about 3 minutes of full speed
video before the throttle is applied since I have about 1.3Gb of RAM free for buffering..

*But* when I try this experiment I hit the throttle after only 65 seconds - an improvement to
be sure, but still a long way short of the 180 seconds that it ought to take.

Part of the test works as expected - the disk writes begin almost immediately due to the low
value for dirty_background_ratio, but the rest is a mystery.

It really looks as if the pages aren't being marked as clean fast enough after they're written.

How else can it take only 70 seconds to reach 95% dirty when I have 1.3Gb of available RAM and data coming in at 25MBytes/sec and out at 17MBytes/sec? It doesn't make any sense...

regards, Anthony

Nate Diller wrote:
> yes, on 2.6 there are two tunables which are important here.
> dirty_background_ratio is the threshold where the kernel will begin
> flushing dirty buffers, so it should change how soon the disk becomes
> active. dirty_ratio changes when the write-throttling code kicks in,
> which is what Anthony is seeing. The purpose of the write throttling
> code is to limit the dirtying process to disk bandwidth, so that is a
> Feature. Anthony, try *increasing* dirty_ratio, you can go up to 100,
> but you could trigger an OOM if you let it get too high, so maybe try
> setting it at 85 or so. This should effectively disable the write
> throttling and give you the bandwidth you want.
>
> NATE

--
Anthony Wesley
Director and IT/Network Consultant
Smart Networks Pty Ltd
Acquerra Pty Ltd

Anthony.Wesley@acquerra.com.au
Phone: (02) 62595404 or 0419409836
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-09-10 02:49    [W:1.506 / U:0.040 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site