Re: [Xen-devel] [RFC][PATCH 12/13] Kemari: use signal to save qemu state for Kemari

[Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>]

Yoshiaki Tamura wrote:

> Stefano Stabellini wrote:
>> Ian Jackson wrote:
>>
>>> Stefano Stabellini writes ("Re: [Xen-devel] [RFC][PATCH 12/13] Kemari: use 
>>> signal to save qemu state for Kemari"):
>>>> I did a couple of quick tests and it seems that signals are faster then
>>>> xenstore but still xenstore offers good performances.
>>>>
>>>> I measured the time using rdtsc at the sender size and at the receiver
>>>> size, the difference between the two values is the following:
>>>>
>>>> signals:   368836
>>>> xenstore: 1984632
>>> This is with an unloaded xenstored, I take it.
>>
>> Yes
> 
> Thanks for measuring the numbers.  It made things clear.
> I think we need a signal-based interface as Ian said previously.
> 
> On our test environment, AMD Barcelona 2.3GHz, the total cpu cycles of Kemari 
> in 
> userland when running I/O intensive applications is around 3000000.
> Although Xen transferring code and QEMU saving code is processed 
> concurrently, 
> using xenstored for staring QEMU portion would lower the performance. 
> Especially if the xenstored gets slower when items are loaded.
> 

Using an event channel to notify qemu in a stubdom that a particular
event occurred improves things a lot: it is even faster than signals!

event channel: 20880

performances using an event channel with qemu as a process in dom0 are
about the same.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel