Making disk IO place nicely in a guest (VirtualBox)

[Adam Vande More]

On Thu, Aug 7, 2014 at 4:05 PM, Mike Tancsa <mike at sentex.net> wrote:

> On 8/7/2014 4:53 PM, Adam Vande More wrote:
>
>>
>>
>> Is your swap device ZFS backed like a ZVOL or something?
>>
>
> Hi Adam,
>         The hypervisor does indeed have swap on zfs.   I will try
> disabling swap in the guest.  I am pretty sure the panic was due to the
> guest hitting and stalling on swap inside the VM.  The guest dmesg is
> attached.
>
> There does not seem to be much memory pressure on the hypervisor.  The
> hypervisor disk is a little slow, as I am experimenting with hast to sync
> across to another zfs box. However, thats only for a zfs volume that holds
> the VM disk images. The OS is not "hastified"
>
> In the hypervisor
>
> # pstat -T
> 284/514622 files
> 51M/16384M swap space
>
> CPU:  2.6% user,  0.0% nice,  6.6% system,  1.7% interrupt, 89.1% idle
> Mem: 191M Active, 277M Inact, 14G Wired, 25M Cache, 1618M Buf, 315M Free
> ARC: 7831M Total, 491M MFU, 6986M MRU, 83M Anon, 39M Header, 232M Other
> Swap: 16G Total, 51M Used, 16G Free
>
> The machine has 16G.  Should I think of limiting ARC ?


The error messages shown are classic swap on ZFS errors which is inherently
unstable especially under high IO loads.



-- 
Adam