On Tue, May 11, 2010 at 10:16:13PM -0400, Martin Langhoff wrote:
> What you are saying is that this is a different issue from what dsd
> tracked down -- entirely possible that these are 2 orthogonal issues.

Yes, entirely possible.

> Does the frequency and/or std deviation change if you tweak the
> vm.dirty_* sysctl values?

Yes, significantly.  What I've observed is most likely entirely due to
vm.dirty_* behaviour.

The default on os123 is

vm.dirty_background_bytes=0
vm.dirty_background_ratio=10
vm.dirty_bytes=0
vm.dirty_expire_centisecs=3000
vm.dirty_ratio=20
vm.dirty_writeback_centisecs=500

Changing this to

vm.dirty_background_ratio=90
vm.dirty_expire_centisecs=60000
vm.dirty_ratio=90
vm.dirty_writeback_centisecs=0

causes the dd to finish in 7 seconds instead of 92 seconds, the sync to
take much longer, and the write latencies are much smaller.

So the effect I was seeing was documented ... "process which is
generating disk writes will itself start writing out dirty data.", per
vm.txt in Documentation/

-- 
James Cameron
http://quozl.linux.org.au/
_______________________________________________
Devel mailing list
[email protected]
http://lists.laptop.org/listinfo/devel

Reply via email to