Not a member yet? Why not Sign up today
Create an account  

  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
 
Brainstorming for XanMod Kernel

#1
This new topic has been opened to replace the former one, lost on a server failure. This is an open space to discuss all ideas and questions about development trends only, not for testing purposes nor benchmarking results of any kind -- there is other topic located at the general section of this forum to post results from benchmarking procedures. Please, feel free to post every useful suggestion to contribute to XanMod Kernel development. Smile

Here is my first question: cpu 400hz timer is useful? (1/400= exact jiffy of 0,0025).
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#2
About CFS tuning:
https://blog.acolyer.org/2016/04/26/the-...ted-cores/
'Those wastedcore patches from the git link are even faster than BFS.' -- user comment.
Bug fixing patches?
https://github.com/jplozi/wastedcores/tr...er/patches
https://lkml.org/lkml/2016/4/23/135
Shy

Edit: thanks Xan for fixing attachment issue.


Attached Files
.pdf   eurosys16-final29.pdf (Size: 364.45 KB / Downloads: 126)
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#3
Deadline+MuQSS?
http://ck-hack.blogspot.com.es/2016/10/m...duler.html
http://ck-hack.blogspot.com.es/2016/10/l...v0116.html
http://ck-hack.blogspot.com.es/2016/10/i...s-116.html
Shy
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#4
I liked the topic title. Wink

Will soon be in the folder /testing a build of 400Hz.

MuQSS, bad is the delay to follow the development of the mainline and other problems that can come later. I can test too.
Reply

#5
You're rigth, MuQSS has collateral issues -- I haven't noticed. I like 'Dragonk' name. Thanks Xan! Wink
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#6
NOOP, CFQ and Deadline under Linux 4.9.0-999:
http://openbenchmarking.org/result/16102...INUX49SC61
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#7
(31-10-2016, 01:22 PM)tropic Wrote: NOOP, CFQ and Deadline under Linux 4.9.0-999:
http://openbenchmarking.org/result/16102...INUX49SC61

Default Deadline It's fine! Cool
Reply

#8
(01-11-2016, 11:32 AM)Xan Wrote:
(31-10-2016, 01:22 PM)tropic Wrote: NOOP, CFQ and Deadline under Linux 4.9.0-999:
http://openbenchmarking.org/result/16102...INUX49SC61

Default Deadline It's fine! Cool

Yes!  Cool
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#9
Is kernel.sched_migration_cost_ns=0 good for throughout with ondemand governor? -- if more CPUs/tasks per cluster are wanted, shouldn't be the interactive governor the choice, as it was intended for?
Edit: would setting the same value of sched_min_granularity_ns balance the response times?
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#10
Is UKSM still merged? New patches: https://github.com/dolohow/uksm
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#11
Artem S. Tashkinov's article: http://itvision.altervista.org/why.linux...rrent.html
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#12
Would it be a good idea to release a RPM kernel flavour of XanMod?
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#13
Consider to release an experimental version with lowlatency from mainline at 400Mhz and CFS latency/min_granularity at 350/175µs -- with benchmarking tests on several machines.

Edit: just to see how it works vs the current releases, to valuate new ways of development.
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#14
DADS: Dynamic and Automatic Disk Scheduling -- aka automatic selection of I/O scheduler for better Linux performance.


Attached Files
.pdf   sac2012_dads.pdf (Size: 719.68 KB / Downloads: 827)
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#15
System seems to work better under tuned deadline I/O scheduler if vm.dirty_ratio=40-60 range and vm.dirty_background_ratio=10-20 range -- benchmarking with good machines is needed.

Edit: probably these ranges can be increased +10 in powerful systems.
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#16
Also, system seems to work better under tuned deadline I/O scheduler if vm.overcommit_ratio=60-80 range -- benchmarking with good machines is needed.

Edit: probably this range can be increased +10 in powerful systems.
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#17
Would it be a good idea to include the 'schedulit' governor? 4.9-rc6-mainline includes this governor with good results on average performance -- observed with both generic and lowlatency flavours.
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#18
I know this will sound weird at this point of development, but I have a doubt after wide tested 4.9-rc6-generic/lowlatency+CFQ/Deadline under very heavy workload: Does lowlatency+CFQ have more 'fair access to storage' and more 'efficient sustained CPU usage' than generic+Deadline on multicore systems?

Edit: probably the answer is 'yes' but tuned deadline elevator still seems to be a good choice.
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#19
Zen I/O scheduler * Primarily based on Noop, Deadline and SIO I/O schedulers:
https://github.com/codediablos/io-schedu...-iosched.c
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply

#20
Desktop latency is getting bigger and bigger -- are current values 800/400µs at CFS?
"(...) the grandest occasion the past or present has seen, or the future can hope to see." -- Cervantes.
Reply



[-]
Quick Reply
Message
Type your reply to this message here.

Image Verification
Please enter the text contained within the image into the text box below it. This process is used to prevent automated spam bots.
Image Verification
(case insensitive)

Forum Jump:


Users browsing this thread:
1 Guest(s)