[PATCH v2 0/2] credit2 runqueue_per_core code

[PATCH v2 0/2] credit2 runqueue_per_core code

[Uma Sharma]

Hi everyone,

This series of patch inserts runqueue_per_core code and creates a boot parameter to choose which type of runqueue mapping to perform i.e core or socket.
Core is used by default as runqueue_per_core scheduler performs better than runqueue_per_socket.

This series mainly performs the following tasks :
-Provide helpers to access the core and socket
-Add runqueue_per_core code
-Adding boot parameter to decide the runqueue

This is relevant to credit2 scheduler only.

Performance Analysis :

Hardware Used: 4 core 2 HT per core machine.

Guest 1:
vcpu=4

Core 4 vcpu:
make -j  1 82.183
make -j  2 39.488
make -j  4 26.568
make -j  6 24.749
make -j  8 24.835

Socket 4 vcpu:
make -j  1 100.044
make -j  2 49.912
make -j  4 31.28
make -j  6 29.058
make -j  8 35.152

credit 4 vcpu:
make -j  1 87.466
make -j  2 42.159
make -j  4 26.922
make -j  6 25.638
make -j  8 25.619

Guest 1:
vcpu=8

Core 8 vcpu:
make -j  1 95.075
make -j  2 45.346
make -j  4 28.775
make -j  6 23.475
make -j  8 21.887

Socket 8 vcpu:
make -j  1 103.118
make -j  2 51.412
make -j  4 29.612
make -j  6 24.112
make -j  8 22.521

credit 8 vcpu:
make -j  1 90.104
make -j  2 44.932
make -j  4 27.168
make -j  6 24.423
make -j  8 22.109

So according to these analysis core scheduler is working better than socket.
And in the case of guest with 4 cores runqueue_per_core credit2 scheduler is working slightly better than credit scheduler.

Signed-off-by: Uma Sharma <uma.sharma523@gmail.com>

Uma Sharma (2):
  [Xen-devel] x86: identifying the boot cpu
  [Xen-devel] sched_credit2.c : runqueue_per_core code

 docs/misc/xen-command-line.markdown |  7 +++++++
 xen/arch/x86/setup.c                |  7 +++++--
 xen/arch/x86/smpboot.c              |  3 ++-
 xen/common/sched_credit2.c          | 27 ++++++++++++++++++++++-----
 xen/include/asm-x86/processor.h     |  2 ++
 5 files changed, 38 insertions(+), 8 deletions(-)

-- 
1.9.1