lkml.org 
[lkml]   [2007]   [Mar]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [RFC][PATCH 0/6] per device dirty throttling
From
Date
On Tue, 2007-03-20 at 20:38 +1100, David Chinner wrote:
> On Tue, Mar 20, 2007 at 09:08:24AM +0100, Peter Zijlstra wrote:
> > On Tue, 2007-03-20 at 18:47 +1100, David Chinner wrote:
> > > So overall we've lost about 15-20% of the theoretical aggregate
> > > perfomrance, but we haven't starved any of the devices over a
> > > long period of time.
> > >
> > > However, looking at vmstat for total throughput, there are periods
> > > of time where it appears that the fastest disk goes idle. That is,
> > > we drop from an aggregate of about 550MB/s to below 300MB/s for
> > > several seconds at a time. You can sort of see this from the file
> > > size output above - long term the ratios remain the same, but in the
> > > short term we see quite a bit of variability.
> >
> > I suspect you did not apply 7/6? There is some trouble with signed vs
> > unsigned in the initial patch set that I tried to 'fix' by masking out
> > the MSB, but that doesn't work and results in 'time' getting stuck for
> > about half the time.
>
> I applied the fixes patch as well, so i had all that you posted...

Humm, not that then.

> > > but it's almost
> > > like it is throttling a device completely while it allows another
> > > to finish writing it's quota (underestimating bandwidth?).
> >
> > Yeah, there is some lumpy-ness in BIO submission or write completions it
> > seems, and when that granularity (multiplied by the number of active
> > devices) is larger than the 'time' period over with we average
> > (indicated by vm_cycle_shift) very weird stuff can happen.
>
> Sounds like the period is a bit too short atm if we can get into this
> sort of problem with only 2 active devices....

Yeah, trouble is, I significantly extended this period in 7/6.
Will have to ponder a bit on what is happening then.

Anyway, thanks for the feedback.

I'll try and reproduce the umount problem, maybe that will give some
hints.

-
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: 2007-03-20 10:49    [W:0.045 / U:1.008 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site