[Xen-devel] Regressions with BVT

["Diwaker Gupta" <diwaker.lists@xxxxxxxxx>]

Hi everyone,

I understand BVT is not the default scheduler anymore, but perhaps
this would be of interest to some people. Some changes in the current
BVT scheduler compared to the BVT code in Xen 3.0 has affected its
correctness.

In the original (Xen 3.0) BVT scheduler the weights resulted in
correct CPU allocations. However, the current BVT scheduler (CS
10647:8e55c5c11475)  does not do it properly.

Here is the outcome of the weights for CPU allocation in the original
(Xen 3.0) BVT scheduler and the current BVT scheduler when Dom0 and
Dom1 run pure CPU intensive tasks (slurp program).

             Xen 3.0 BVT                 Current BVT

           Dom 0   Dom 1                 Dom 0   Dom 1
---------------------------------------------------------------
weights     10      10                    10      10
CPU Util    50%     50%                   50%     50%

---------------------------------------------------------------
weights      10     20                    10      20
CPU Util    66.6%  33.3%                  60%     40%

---------------------------------------------------------------
weights     10      30                    10      30
CPU Util    75%     25%                   67%     33%

---------------------------------------------------------------
weights     10      40                    10      40
CPU Util    80%     20%                   70%     30%

---------------------------------------------------------------
weights     10      50                    10      50
CPU Util    83%     16.6%                 72.7%   26.6%

---------------------------------------------------------------
weights     10      60                    10      60
CPU Util    85.6%   14.3%                 75%     25%

---------------------------------------------------------------
weights     10      70                    10      70
CPU Util    87.4%  12.5%                  77%     23%

---------------------------------------------------------------

CPU utilization were obtained using 'xentop'.

Does anyone know which new changes might have caused this anomaly? I
ran 'hg annotate' on sched_bvt.c and it looks like most of the
algorithm specific code has not been touched in a while.

Thanks,
Diwaker
--
Web/Blog/Gallery: http://floatingsun.net/blog

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