Re: Help low-jitter man.

[Ove Karlsen <ove.karlsen@paradoxuncreated.com>]

On Tue, 06 Nov 2012 17:11:02 +0100, Ove Karlsen  
<ove.karlsen@paradoxuncreated.com> wrote:

> Den 06.11.2012 16:40, skrev Ove Karlsen:
>> Heya, it`s me again. I am getting my low-jitter E5 computer, soon, and I
>> have done all the tweaks I can think of with the kernel.
>>
>> I was wondering if anyone had anything to add, any good technical
>> opinion on reducing jitter (for workstation desktop/OpenGL).
>>
>> I have a kernel with "low latency desktop" enabled. And edited
>> Kconfig.hz to 90.
>> idle=poll and tick_skew=1 in boot options.
>> Compile a shaved local kernel by doing "make localmodconfig" from a full
>> distro kernel.
>> (Also turned of high_res_timers, to get 90hz)
>>
>> CFS granularity 1158500nS. (Current setting is a bit high)
>> Reniced X to -20, to prevent X being a bottleneck.
>> Remove 60hz limit in doom 3, to avoid timing jitter.
>> http://paradoxuncreated.com/tmp/AutoExec.cfg
>>
>> Some nvidia-specific stuff on my blog aswell :
>> http://paradoxuncreated.com/Blog/wordpress/?p=2268
>>
>> A lot of people think SLAB has lower jitter than SLUB etc, aswell, and
>> some tuning with regards to what kernel options execute the least code,
>> least background activity can be done. Opinions?
>>
>> I use SLUB currently, and have turned off mem-compaction.
>> And probably all kinds of debugging, counters and tracers.
>>
>> RCU boost to 99.
>>
>> I think I have collected most the relevant data for a low-jitter linux
>> config. but in case there is additional information, I would love to
>> hear it.
>>
>> Also ACPI on or off seems to be varying opions on. Some state ACPI
>> removes SMI`s? I don`t know what is true here, and knowing that some
>> recommend 10000hz timer for "low-jitter" (which is going the completely
>> wrong way) I would like to hear your opinion on this aswell.
>>
>> Also approaching a transparent OS, with realtime-threads is interesting.
>> Such as running maintask (optionally + depencies) as realtime.
>> Unfortunately it doesn`t seem to work so well with standard FIFO. Do you
>> think deadline could help this? Accurate 0.95ms of 1ms would be
>> interesting to try.
>>
>> Also if anyone knows a linux-distro with the same philosophy throughout
>> the OS, I would be interested to know about it.
>>
>
> Also have anyone given any thoughts to vsynced HZ? And also even  
> possibly dividing subhz off that, or more hz counters if neccesary.

Ok, I think after this, simple renicing seems to reduce significant  
amounts of jitter. So then the question becomes, how far can you take  
renice.

Peace Be With you.