RE: [Xen-ia64-devel] Time for hybrid virtualization?

["Xu, Anthony" <anthony.xu@xxxxxxxxx>]

Hi Alex,

The system call would not incur such big degradation.

Guest process switch is the root cause of the degradation.
We need to paravirtualize the schedule path for better performance.


Anthony



Alex Williamson wrote:
> Hi all,
> 
>    I got a pointer to re-aim-7 and thought the high-systime workload
> might give us a good worst case scenario.  This workload does no I/O
> and a lot of the tests produce very high system time.  Here's the
> results: 
> 
> 
> 
>                  Max JPM
>              % of native
> Native
>                242347.19
>                  100.00%
> Para-virt
>                218694.95
>                   90.24%
> Full-virt
>                 58289.32
>                   24.05%
> 
> Not a pretty picture for HVM.  My test system is a 2-socket, dual core
> Montvale w/ threads disabled (4-way), 16GB RAM total.  This is the
> same system I used for the kernel build tests.  The guest domain was
> configured for 4-way, 8G, and I used mem=8G to get the native test to
> use the same.  PV-on-HVM drivers did not improve the full-virt
> numbers. I've attached a graph of the runs.  This is obviously a
> concern for a VT-mode dom0.  We probably need to profile the HVM run
> to see if these are things that could be easily paravirtualized for a
> hybrid mode guest. 
> 
>       Alex

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel