http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=717

------- Comment #2 from inak...@jp.fujitsu.com 2006-10-24 04:59 ------- In the previous report, the vcpu configuration was: dom0: #vcpu=1, pinned on pcpu#0 domU: #vcpu=1, any cpu (e.e.

Comparison was not fair. I would report the case where the domU vcpu is pinned so that a pcpu is dedicated to it. In addition, I added standard deviations over each 10 runs.

The conclusion was;

1/ considering the efficient buffered I/O ratio and the large standard deviation of cached I/O throughput, we can say that the performance was almost ideal in the latest C/S, meaning that cached I/O ratio was similar to buffered I/O ratio. 2/ There was no significant difference between the results with and without vcpu-pin. I confirmed this by t-test with 5% significance level.

Results:

Xen : ia64-unstable C/S 11810 Native: rhel4u4 Xen0 : linux-2.6.16.29-xen XenU : linux-2.6.16.29-xen

o hdparam -tT /dev/sda Value Format: mean+/-stdev (normalized by native throughput)

buffered IO: native 93.29+/-0.05 dom0 92.96+/-0.62 (0.99) domU (without vcpu-pin) 92.74+/-0.92 (0.99) domU (with vcpu-pin) 91.78+/-2.50 (0.98) cached IO: native 5959.69+/-1897.60 dom0 5481.48+/- 722.56 (0.91) domU (without vcpu-pin) 5037.76+/- 597.84 (0.85) domU (with vcpu-pin) 5788.94+/- 788.11 (0.97)