Re: Fw: Some Xen performance measurements

[Keir Fraser <Keir.Fraser%cl.cam.ac.uk@localhost>]

Martti,

> > I tested NetBSD/xen domain0 and domainU (3.99.2) to get some idea
> > if this whole thing is suitable to run several low-CPU-intensive
> > hosts (Linux and xBSD) on a single 1U Dell PowerEdge 750 with 4 GB
> > of RAM.

It's great to see some promising performance results!

> > - One would assume that running 4 VMs would take 4x the compilation time
> >   compared to 1 VM. Any ideas why the time is only 1600 instead of 2000?

Unless you do a parallel compile (make -j #) you are unlikely to be
saturating the CPU. Running multiple VMs will increase the chance that
they fill in each others 'I/O gaps' and better utilise the CPU.

So in fact your benchmark may have been testing I/O performance as
much as Xen performance: doing a 'make -j 4' or 'make -j 8' in each VM
will not only make each one hit the CPU harder but you will
experience more process forks and destroys and generally stress the
VMM interfaces a lot more. That might give you a better worst case
bound for performance.

 Regards,
 Keir

> > - Use of vnd(4) instead of direct access to the harddisk does not give
> >   any noticeable slowness, at least with this small number of VMs
> >   running at the same time.
> >
> >
> > 
>