Xen devel - Sep 2008 - HVM kernbench performance on Linux (2.6.27-rc4 vs. 2.6.18.6-amd64)

Hi All,

Starting a new thread to narrow down the problem.

So it seems the problem doesn''t occur for me on Debian 4.0 HVM guest,
as suggested by Gianluca Guida.

The Debian 4.0 vanilla install has a  2.6.18.6-amd64 kernel.

The results are below. So it seems that the guest kernel makes a big difference.

The performance problem is not seen on the debian kernel.

HVM guest (Debian 4.0 vanilla 2.6.18.6-amd64)

shadow2
2.6.18.6-amd64 guest kernel
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 506.658 (4.13104)
User Time 364.536 (0.337757)
System Time 140.834 (4.04617)
Percent CPU 99 (0)
Context Switches 29803 (120.69)
Sleeps 37929.2 (355.421)

shadow3
2.6.18.6-amd64 guest kernel
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 489.58 (4.00024)
User Time 389.328 (0.614304)
System Time 98.838 (4.86796)
Percent CPU 99 (0)
Context Switches 29314.6 (196.596)
Sleeps 37773.6 (231.069)

vs. the same machine and guest config with a 2.6.27-rc4 guest kernel


shadow2
2.6.27-rc4 guest kernel
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 737.144 (5.19414)
User Time 498.508 (2.52895)
System Time 235.056 (2.71348)
Percent CPU 99 (0)
Context Switches 133127 (823.517)
Sleeps 36295.4 (124.088)


shadow 3
2.6.27-rc4 guest kernel
Average Optimal load -j 4 Run (std deviation):
Elapsed Time 2081.71 (34.0459)
User Time 617.36 (3.61771)
System Time 1430.36 (28.3309)
Percent CPU 98 (0)
Context Switches 331843 (5283.28)
Sleeps 37329.8 (91.538)

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

> So it seems the problem doesn''t occur for me on Debian 4.0 HVM
guest,
> as suggested by Gianluca Guida.
> 
> The Debian 4.0 vanilla install has a  2.6.18.6-amd64 kernel.
> 
> The results are below. So it seems that the guest kernel makes a big
> difference.
It sounds like the "find writeable mappings" heuristic might be broken
for newer linux 64b kernels. That will be painful for shadow2, but
catastrophic for shadow3.

I wonder if there''s a second 1:1 direct mapping added, or whether the
direct mapping has moved?

Ian


 
> The performance problem is not seen on the debian kernel.
> 
> HVM guest (Debian 4.0 vanilla 2.6.18.6-amd64)
> 
> shadow2
> 2.6.18.6-amd64 guest kernel
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 506.658 (4.13104)
> User Time 364.536 (0.337757)
> System Time 140.834 (4.04617)
> Percent CPU 99 (0)
> Context Switches 29803 (120.69)
> Sleeps 37929.2 (355.421)
> 
> shadow3
> 2.6.18.6-amd64 guest kernel
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 489.58 (4.00024)
> User Time 389.328 (0.614304)
> System Time 98.838 (4.86796)
> Percent CPU 99 (0)
> Context Switches 29314.6 (196.596)
> Sleeps 37773.6 (231.069)
> 
> vs. the same machine and guest config with a 2.6.27-rc4 guest kernel
> 
> 
> shadow2
> 2.6.27-rc4 guest kernel
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 737.144 (5.19414)
> User Time 498.508 (2.52895)
> System Time 235.056 (2.71348)
> Percent CPU 99 (0)
> Context Switches 133127 (823.517)
> Sleeps 36295.4 (124.088)
> 
> 
> shadow 3
> 2.6.27-rc4 guest kernel
> Average Optimal load -j 4 Run (std deviation):
> Elapsed Time 2081.71 (34.0459)
> User Time 617.36 (3.61771)
> System Time 1430.36 (28.3309)
> Percent CPU 98 (0)
> Context Switches 331843 (5283.28)
> Sleeps 37329.8 (91.538)
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel