| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
| |
BUG=579813
Review URL: https://codereview.chromium.org/1669493005
Cr-Commit-Position: refs/heads/master@{#379669}
|
|
|
|
|
|
|
|
|
| |
BUG=138542
TBR=tsepez@chromium.org
Review URL: https://codereview.chromium.org/1546533002
Cr-Commit-Position: refs/heads/master@{#366527}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
lock most threads of ipc_perftest and ipc_mojo_perftest
to a single core.
This is important because the different threads and
processes naturally end up on different cores and these
cores end up idle about 40% of the time, due to each
process waiting on the others. This confuses the power
management systems in Windows and Linux such that they
don't reliably ramp up the CPU speeds.
In extreme cases this can make these tests run ~3x slower
(at 1.2 GHz instead of at ~3.8 GHz).
In addition to running the tests more quickly the thread
affinity makes them much more consistent, which makes
measuring performance more scientific.
Setting the OS power plan to high-performance can also
help maximize performance and minimize variation.
Review URL: https://codereview.chromium.org/914403002
Cr-Commit-Position: refs/heads/master@{#316066}
|
|
This change adds ipc_mojo_perftests that runs the same benchmark
as of ipc_perftests. Now head-to-head comparison becomes possible.
For this change, whole ipc_perftests logic is extracted to
ipc_perftest_support.cc to make it reusable by ipc_mojo_perftests.
TEST=none
BUG=none
R=jam@chromium.org, darin@chromium.org, yuzhu@chromium.org
Review URL: https://codereview.chromium.org/536213002
Cr-Commit-Position: refs/heads/master@{#293988}
|