[Devel] Re: Container Test Campaign

[Sam Vilain]

Marc E. Fiuczynski wrote:
>> My guess is that you just can't leverage that kind of benefit
>> from a hypervisor approach, but I don't really know enough about
>> how they work under the hood to be able to say.
>
> Right, our current benchmark shows that even w/o the benefit of FS
> unification that a container-based approach (in our case vserver)
> still significantly outperforms Xen on both SPECWEB'99 and OSDB
> benchmarks. Please take a peek at our evaluation sections in the
> paper that I forwarded before.

Yes, and very nice paper it is indeed, though I'm still curious as to
why all the dbench scores go down for SMP, even on the vanilla Linux
kernel.

I guess that demonstrating the effects of the unification would have
to come under the scalability benchmarks section, and probably with a
non-IO bound test.  If Specweb doesn't support testing a bunch of
machines at once, perhaps you could fudge it by using IP Virtual
Server on another machine to distribute the load across the VM instances.

Sam.