| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Sun, 16 Dec 2018 17:53:41 -0800
From: "Doug Smythies" <dsmythies@...us.net>
To: "'Rafael J. Wysocki'" <rjw@...ysocki.net>,
"'Giovanni Gherdovich'" <ggherdovich@...e.cz>
Cc: "'Srinivas Pandruvada'" <srinivas.pandruvada@...ux.intel.com>,
"'Peter Zijlstra'" <peterz@...radead.org>,
"'LKML'" <linux-kernel@...r.kernel.org>,
"'Frederic Weisbecker'" <frederic@...nel.org>,
"'Mel Gorman'" <mgorman@...e.de>,
"'Daniel Lezcano'" <daniel.lezcano@...aro.org>,
"'Chen, Hu'" <hu1.chen@...el.com>,
"'Quentin Perret'" <quentin.perret@....com>,
"'Linux PM'" <linux-pm@...r.kernel.org>,
"Doug Smythies" <dsmythies@...us.net>
Subject: RE: [RFC/RFT][PATCH v8] cpuidle: New timer events oriented governor for tickless systems
On 2018.12.11 03:50 Rafael J. Wysocki wrote:
...[snip]...
> With this version of the TEO governor I'm observing slight, but consistent
> performance improvements in some different benchmarks on a few different
> systems with respect to menu
Same here.
> and the corresponding power draw differences
> appear to be in the noise. The idle power doesn't seem to change either.
Same here.
> Also the "above" and "below" metrics (introduced by a separate patch under
> discussion) indicate changes in the "good" direction.
My graphs now include the "above" and "below" metrics.
In particular see Idle State 1 "above" (was too deep) graphs in the links below.
However, performance is up and power about the same, so O.K.
> Overall, I like this one, so I may consider dropping the RFC/RFT tag from the
> next submission. :-)
>
> v7 -> v8:
> * Apply the selection rules to the idle deepest state as well as to
> the shallower ones (the deepest idle state was treated differently
> before by mistake).
> * Subtract 1/2 of the exit latency from the measured idle duration
> in teo_update() (instead of subtracting the entire exit latency).
> This makes the idle state selection be slightly more performance-
> oriented.
I cherry picked a couple of the mmtests that Giovanni was doing:
Test kernels:
"stock" kernel 4.20-rc5 + a couple of rjw patches. Specifically:
2a110ed cpuidle: poll_state: Disregard disable idle states
8f09875 cpuidle: Add 'above' and 'below' idle state metrics
d6851a5 Documentation: admin-guide: PM: Add cpuidle document
2595646 Linux 4.20-rc5
"teov6" above + teov6 patch
"teov7" above + teov7 patch
"teov8" above + teov8 patch
1.) mmtests - netperf-unbound test (UDP):
4.20-rc5 4.20-rc5 4.20-rc5 4.20-rc5
stock teo6 teo7 teo8
Hmean send-64 129.64 ( 0.00%) 132.45 * 2.17%* 130.55 * 0.71%* 132.87 * 2.49%*
Hmean send-128 259.53 ( 0.00%) 264.90 * 2.07%* 261.61 * 0.80%* 264.94 * 2.09%*
Hmean send-256 515.24 ( 0.00%) 525.41 * 1.97%* 517.41 * 0.42%* 524.88 * 1.87%*
Hmean send-1024 2041.01 ( 0.00%) 2079.22 * 1.87%* 2045.03 * 0.20%* 2077.25 * 1.78%*
Hmean send-2048 3980.04 ( 0.00%) 4071.09 * 2.29%* 4041.15 * 1.54%* 4057.09 * 1.94%*
Hmean send-3312 6321.90 ( 0.00%) 6460.23 * 2.19%* 6395.71 * 1.17%* 6409.09 * 1.38%*
Hmean send-4096 7695.18 ( 0.00%) 7882.81 * 2.44%* 7813.72 * 1.54%* 7832.77 * 1.79%*
Hmean send-8192 13920.53 ( 0.00%) 14146.47 * 1.62%* 13986.72 * 0.48%* 14079.07 * 1.14%*
Hmean send-16384 24714.99 ( 0.00%) 25225.95 * 2.07%* 24896.10 * 0.73%* 25131.52 * 1.69%*
Hmean recv-64 129.64 ( 0.00%) 132.45 * 2.17%* 130.55 * 0.71%* 132.87 * 2.49%*
Hmean recv-128 259.53 ( 0.00%) 264.90 * 2.07%* 261.61 * 0.80%* 264.94 * 2.09%*
Hmean recv-256 515.24 ( 0.00%) 525.41 * 1.97%* 517.41 * 0.42%* 524.88 * 1.87%*
Hmean recv-1024 2041.01 ( 0.00%) 2079.22 * 1.87%* 2045.03 * 0.20%* 2077.25 * 1.78%*
Hmean recv-2048 3980.04 ( 0.00%) 4071.09 * 2.29%* 4041.15 * 1.54%* 4057.09 * 1.94%*
Hmean recv-3312 6321.88 ( 0.00%) 6460.23 * 2.19%* 6395.71 * 1.17%* 6409.09 * 1.38%*
Hmean recv-4096 7695.15 ( 0.00%) 7882.81 * 2.44%* 7813.72 * 1.54%* 7832.75 * 1.79%*
Hmean recv-8192 13920.52 ( 0.00%) 14146.43 * 1.62%* 13986.72 * 0.48%* 14079.07 * 1.14%*
Hmean recv-16384 24714.99 ( 0.00%) 25225.90 * 2.07%* 24896.07 * 0.73%* 25131.49 * 1.69%*
Graphs: http://www.smythies.com/~doug/linux/idle/teo8/net-pref-udp-unbound/index.html
(note: slow upload speed from my server)
2.) mmtests - sockperf-udp-throughput test:
4.20-rc5 4.20-rc5 4.20-rc5 4.20-rc5
stock teo6 teo7 teo8
Hmean 14 24.57 ( 0.00%) 25.91 * 5.46%* 25.99 * 5.78%* 25.73 * 4.75%*
Hmean 100 175.37 ( 0.00%) 185.09 * 5.54%* 185.89 * 6.00%* 184.48 * 5.19%*
Hmean 300 523.81 ( 0.00%) 553.47 * 5.66%* 554.70 * 5.90%* 550.16 * 5.03%*
Hmean 500 870.08 ( 0.00%) 918.88 * 5.61%* 924.33 * 6.24%* 914.53 * 5.11%*
Hmean 850 1449.44 ( 0.00%) 1530.84 * 5.62%* 1535.40 * 5.93%* 1522.53 * 5.04%*
Graphs: http://www.smythies.com/~doug/linux/idle/teo8/sockperf-udp-throughput/index.html
(note: slow upload speed from my server)
The above results tables are also here:
http://www.smythies.com/~doug/linux/idle/teo8/index.html
I wanted to also do the tbench on loopback test, but have not been able to get it working on my system yet.
I'll supply more test results at a later date.
... Doug
Powered by blists - more mailing lists