[Xen-ia64-devel] Benchmarking [RFC] performance tuning TAKE 3

[Ken Hironaka <hironaka.ken@xxxxxxxxxxxxxxxx>]

I did a crude benchmark on the following patch that tries to reduce
costs with vcpu migration and physical-machine address conversion. The
benchmark was a kernel compilation. The results are by no means
definitive and serious testing requires 

patch used:
http://lists.xensource.com/archives/html/xen-ia64-devel/2006-08/msg00157.html

conditions:
-machine:
see file:dom0_settings

-running with 2 domUs(domU1,domU2)
settings:
memory=1024
vcpus=3

UNPATCHED
changeset:   10937:15498beef5d8
user:        awilliam@xxxxxxxxxxx
date:        Tue Aug 08 14:42:34 2006 -0600
summary:     [IA64] Remove scheduler heartbeat

PATCHED
changeset:   11054:c69df7b381cd
tag:         tip
user:        yamahata@xxxxxxxxxxxxx
date:        Thu Aug 17 13:39:47 2006 +0900
summary:     p2m exposure test module

test:
linux 2.6 kernel compile

$make clean
$time make -j4 >& testlog.txt

results:
1) testing only on Dom0(DomUs idle) with/without patch

UNPATCHED:      real    19m37.632s user    32m59.270s  sys     2m10.280s

PATCHED:        real    19m27.442s user    32m48.890s  sys     2m2.450s

2) testing on the two DomUs simultaneously(Dom0 Idle)
UNPATCHED:
domU1
real    10m13.558s
user    22m48.310s
sys     1m30.440s

domU2
real    13m59.841s
user    22m21.620s
sys     1m22.750s

PATCHED:
domU1
real    13m45.952s
user    22m13.430s
sys     1m22.090s

domU2
real    13m44.554s
user    22m16.800s
sys     1m20.500s

Input on test suggestions are welcome

Best Regards,
Ken Hironaka

Attachment: dom0_settings
Description: Text document

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel