I am running one of the applications from the Stanford splash2 application suite ( http://www-flash.stanford.edu/apps/ ) and getting an unexpected (~5x) performance decrease in domU when specifying more than 1 processor in the application.   dom0 seems to handle this fine.
 
Specificly I am seeing this on the water-nsquared app.  I am not going to give my machine config here (if you need them please ask) just the differences in runtime and the parameters of the app.
 
 
all are running single processor & 1GB RAM
(times reported as real/user/system, in seconds)
 
water-nsquared 64 processors 8000 molecules
 
baseline (non-xen) 72.20/78.80/0.17 
Dom0                   77.76/77.17/0.16
DomU                   357.24/357.04/0.018
 
water-nsquared 1 processor 8000 molecules
 
baseline (non-xen) 73.16/72.75/0.15 
Dom0                   72.75/72.37/0.02
DomU                   66.07/66.04/0.02
 
 
I have disabled the /lib/tls.
 
I am hoping that someone here with a little more experience with Xen can help me try to explain this.  Any input would be greatly appreciated.
 
Jessie
jessiesmart354@xxxxxxxxx