Disk I/O Question 7.0-STABLE

[Karl Denninger]

On Fri, Apr 11, 2008 at 10:33:16AM +0100, Thomas Hurst wrote:
> * Karl Denninger (karl at denninger.net) wrote:
> 
> > I have disks on the internal ICH7 adapters (on the motherboard), SATA,
> > and also a TWE controller with two disks.
> > 
> > When hitting the TWE controller hard I can hose the I/O performance on
> > the primary (onboard) adapter quite severely to the point that it
> > seriously impairs the system's overall performance.  "systat -vm"
> > shows the primary channels going to 100% busy but the transaction and
> > I/O rate count is not all that high.  The twe disk I am hammering, of
> > course, is saturated (and blowing data at a high rate of speed -
> > 70MBps+)
> 
> I see exactly the same behavior on a dual dual core AMD box using amr(4)
> (8 port LSI MegaRAID SATA) and an 8 port Marvell 88SX6081 ata(4)
> controller.  I have 3 RAID-1 arrays configured on the LSI and a single
> drive on the Marvell; I can freeze up IO to all of them for several
> seconds just by running find / >/dev/null for up to 30s.
> 
> gstat shows the IO queue for one drive in the high hundreds (all
> writes), and it seems waiting for it to flush is what kills the rest of
> the system.
> 
> My best guess is this is a result of the syncer flushing atime updates.
> I should try mounting my /usr noatime and seeing if the problem goes
> away.  Not the greatest of solutions, but meh.

Mounting with noatime does help a bit but does NOT prevent the problem from
showing up.

Note that softupdates is OFF on the disk that is being hammered when
everything goes to hell so its not a buffer flushing problem there either.

--
-- 
Karl Denninger (karl at denninger.net) Internet Consultant & Kids Rights Activist
http://www.denninger.net	My home on the net - links to everything I do!
http://scubaforum.org		Your UNCENSORED place to talk about DIVING!
http://genesis3.blogspot.com	Musings Of A Sentient Mind