[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-discuss] VM tuning
From: |
Stephan von Krawczynski |
Subject: |
Re: [Qemu-discuss] VM tuning |
Date: |
Fri, 4 Oct 2013 18:23:57 +0200 |
Hi Peter,
I know the classical qemu tips, they're all done.
In short: deadline, AIO=native, cache=none, queue=<numvpus>
AFAIK dirty_ratio lets IO become sync above its ratio. This means that all
processes have to wait, and I guess that's the lag.
--
Regards,
Stephan
On Fri, 04 Oct 2013 09:09:53 -0700
Peter Kieser <address@hidden> wrote:
> 1. Make sure that the cache is disabled in the guest (cache=none)
> 2. Look at the Redhat profiles ('tuned') for virtual-host which
> specifies the following sysctl knobs:
>
> kernel.sched_min_granularity_ns=10000000
> kernel.sched_wakeup_granularity_ns=15000000
> vm.dirty_ratio=10
> vm.dirty_background_ratio=5
> vm.swappiness=10
>
> Having a higher dirty_ratio will cause larger amounts of I/O to be
> cached in memory - but it will also cause the I/O to hit the disk all at
> once and have un-even I/O patterns which could be causing the issues you
> are seeing.
>
> -Peter
>
>
> On 2013-10-04 9:07 AM, Stephan von Krawczynski wrote:
> > Hi Peter,
> >
> > I know the deadline stuff, this is already in. But the problem arises
> > nevertheless. My idea is that mv.dirty_ratio should be a lot higher to
> > prevent
> > the lagging, probably around 80.
>
>