Re: [Xen-devel] SMP dom0 and AMD64? + some benchmark figures

[Nicholas Lee <emptysands@xxxxxxxxx>]

On 10/7/05, Paul Larson <plars@xxxxxxxxxxxxxxxxx> wrote:
> Ian Pratt wrote:
> >SMP dom0 should work fine, but non-SMP is more stable.

Right.  Once Xen gets to 3.0.5+ or similar will there be a recommended config?

How does the performance profile work with SMP and the interaction
between domU and dom0 for CPU bound I/O?

Are the CPU slices allocated independantly for each CPU in SMP domains?


> Since we're supposed to be flushing out bugs, could we consider changing
> this to default to SMP dom0 to help get more exposure here?

I should be able to test out my NX based desktop over the weekend.

Of quick interest might be one of the benchmarks I use:

nic@wuwei:~/sys/ufc$ ./speedf  # 32bit Debian/Sarge 2.4.26-1-386
Running UFC-crypt for 100 seconds of virtual time ...
Did 153237.030000 fcrypt()s per second.

nic@wuwei:~/sys/speedf/ufc$ ./speedf # 64bit Debian/Sarge 2.6.8-11-amd64-generic
Running UFC-crypt for 100 seconds of virtual time ...
Did 300135.843750 fcrypt()s per second.

nic@wuwei:~/sys/speedf/ufc$ ./speedf # 2.6.12-xen0 #3 SMP
Running UFC-crypt for 100 seconds of virtual time ...
Did 293966.906250 fcrypt()s per second.

speedf is pretty much a single processor testing, but Xen 3.0 holds up
well  only dropping 2%.

I don't have 2.0 figures yet, and I'll get some domU figures later this weekend.

--
Nicholas Lee
http://stateless.geek.nz
gpg 8072 4F86 EDCD 4FC1 18EF  5BDD 07B0 9597 6D58 D70C

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