Re: [Xen-users] Xen Performance with Charm++

[Anthony Liguori <aliguori@xxxxxxxxxx>]

Lamia M.Youseff wrote:
> Dear All,
>
> I am doing a performance study on Xen, and we are deploying Charm++ 
> parallel runtime library as part of our applications. However, we 
> observed a 20% slow down in Xen kernel that was not observed with 
> MPICH. It is very important for us to understand why this overhead is 
> happening.
> I am guessing that is caused due to the higher cost of system calls 
> like fork and exec, since Charm++ deploy a dynamic load-balancing 
> algorithm that relocates jobs across the cluster, and thus using 
> extensively the fork and exec syscalls. However, it is surprising for 
> me that the overhead is as high as 20%.
> My question is about the cost of fork and exec in the new Xen v3? My 
> guess is that they are higher than those published for v1 in Xen Art 
> of virtualization paper.
You'll have to expand a bit more on the particulars of your setup.  If 
you've got domUs running and your using bridging with any significant 
network load, you can expect a pretty good amount of CPU is going to be 
going to that.
Regards,

Anthony Liguori

> Thank you,
> Lamia Youseff
>
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxxxxxxxx
> http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users