Hi Quentin and EAS team,
As we discussed offline with Quentin my issues with EAS (a bit old version)
and tasks migration, here is the patch set that I have with some fixes.
The version that I have been working on is:
- kernel: 4.14.24 with some driver fixes for Exynos SoC
(it was close to eas/next/integration_base_20171211_0913)
- EAS: eas/next/integration_20171211_0913
http://linux-arm.org/git?p=linux-power.git;a=shortlog;h=refs/heads/eas/next…
This patch set tries to solve:
- issue with one little core not used in 1st phase,
- issues with not used LITTLE cores in the second phase,
The test script uses sysbench in 2 phases. First it starts two task sets:
4 tasks on big cores and 4 will be on little. The big tasks finish earlier
and little should be migrated there. Then there is the second phase. The set
of 4 new tasks is started and should be placed on little cores.
------------8<-------------------
REQUESTS=20000
BIGS="4 5 6 7"
LITTLES="0 1 2 3"
PATH="$PATH:/root/devlib-target/bin"
# Don't care about the score for those, just keep the bigs busy
for i in $BIGS; do
taskset -c $i sysbench --max-requests=$((REQUESTS / 4)) \
--test=cpu run &>/dev/null &
done
for i in $LITTLES; do
$(sleep 25 && sysbench --max-requests=$((REQUESTS / 8)) --test=cpu run &>/dev/null &) &
done
for i in $LITTLES; do
sysbench --max-requests=$REQUESTS --test=cpu run \
| grep "total time:" &
done
wait
----------->8--------------------
It is a bit old EAS but maybe you can find something useful in this fixes
or test scenario.
Regards,
Lukasz Luba
Lukasz Luba (5):
arm64: exynos: Add support of exynos5433 to EAS with energy model v1
trace: sched: add new trace events for tracking migrations
sched/fair: drop aggressive migration
sched/fair: change migration destination based on CPU utilization
sched/fair: change finding idle group
arch/arm64/boot/dts/exynos/exynos5433.dtsi | 8 ++
arch/arm64/kernel/energy_model.h | 119 +++++++++++++++++++
include/trace/events/sched.h | 130 +++++++++++++++++++++
kernel/sched/fair.c | 112 +++++++++++++++---
4 files changed, 351 insertions(+), 18 deletions(-)
--
2.17.1
(Post the questions here on behalf of Yun Hsiang)
Hi all,
I’d like to ask discuss some questions about uclamp in linux-power.git<http://www.linux-arm.org/git?p=linux-power.git;a=summary>.
Code is based on the following version.
linux-power.git<http://www.linux-arm.org/git?p=linux-power.git;a=summary>
tag: topic-util_clamp-20190412
1. If a task clamped utilization is higher than a cpu capacity, should EAS bypass that cpu?
kernel/sched/fair.c
static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
{
…
/* Skip CPUs that will be overutilized. */
util = cpu_util_next(cpu, p, cpu); <= Should this util use task clamped utilization?
cpu_cap = capacity_of(cpu);
if (cpu_cap * 1024 < util * capacity_margin)
continue;
…
}
1. Will uclamp add a kernel-space API to set per-task utilization clamping?
Drivers/kernel module can use this API for precise performance/power tuning.
Best regards,
Yun
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
Hi Linaro,
I am really intrested in LISA... and may I ask some question about it?
on github https://github.com/ARM-software/lisa
and PDF
https://developer.arm.com/-/media/developer/developers/open-source/energy-a…
P.43
I just read and download the source code and try to test my android phone...
however, I could not understand "Slack"'s definition in code.
===
Thus two types of test failures can result from these tests:
1. test_task_placement asserts that the placement was energy efficient. In
a big.LITTLE system, if a low-utilisation task was placed on a big CPU then
this test should be expected to fail.
*2. test_slack asserts that the required bandwidth was provided. In a
big.LITTLE system if a high-utilization task was placed on a LITTLE CPU
then this test should be expected to fail.*
===
I run generic.py and can see sometimes "Slack" is negative and sometimes is
positive...
Negative > test failed, positive > test pass...
===
How to define this SLACK + or -...? I cannot understand...
====
Besides, Could I judge my algorithm is worse or better about power
efficient by hikey_energy.py
&
est_power = self.get_power_df(experiment)
est_energy = area_under_curve(est_power.sum(axis=1), method='rect
(I don't have energy meter.... so.. I can just est by EM. Ofcause, I will
design a power for xxx_energy.py)
Really appreciate for your help.
I want to write a report and see this module is so excited.
Best,
CCC
<div dir="ltr"><div dir="ltr"><div>Hi,</div><div><br></div><div>I guess
this is the appropriate time for you to target Competitors IoT Install Base
with Service Providers/Partners/Resellers as for your strategic marketing
and channel opportunities for 2019.</div><div><br></div><div>We also have
distinctive Types Of development customers like: IBM IoT Users, AWS IoT
Users, Cisco IoT Users, Prophet IoT Users , SAP IoT Users , Siemens IoT
Users , PTC IoT Users, Autodesk IoT Users, Salesforce IoT
Users </div><div><br></div><div>Kindly review and let me know if we can
connect or let me know if I can share more insights on
this.</div><div><br></div><div>Await your
response!</div><div><br></div><div>Warm Regards</div><div>Angie
Campos</div><div>Demand
Generation</div><div><br></div><div><br></div><div><br></div><div>If you
don't want to include yourself in our mailing list, please reply back
“Leave Out"" in a subject
line"</div><div><br></div></div></div>
<p> </p><a style='display: block; margin: 32px 0 40px 0; padding:
10px; font-size: 1em; text-align: center; border: 0; border-top: 1px solid
gray; '
href='chrome.google.com/webstore/detail/gsm-mailmerge/fjimadonilpeknihlnnkbgmlmjn…'>powered
by GSM. Free mail merge and email marketing software for Gmail.</a>
Dear Sir,
With reference to your PO NO. FINS/1819/NSK/PO438, the material is ready at our work.
Please find attached herewith Proforma invoice for your reference. You are requested to please make the payment so that we can proceed further.
Thanks & Regards,
Ms. Meena Chavan,
Account’s Manager - 8097070830
WESMEC ENGINEERING PVT.LTD