Benchmarking The Ubuntu "Low-Jitter" Linux Kernel

[Paradox Ethereal]

edit: outdated

[Paradox Ethereal]

Imagine if you could use a million z80s. 1ms osjitter, and suddently a million ms lost. Well figuratively speaking. One needs to have a good multicpu mainboard design aswell. But still. That is what liked about BeOS in it`s day. Designed from the ground of for multiple cpus. Instead of 1 or 2 expensive cpus, 8 cheap. Give the ghetto people doom 3.

As many obscure things, good or bad, it faded.

[Paradox Ethereal]

My kernels are faster and more responsive ... and energy consumption + my script APM at most 40-60% of standard solutions

http://ubuntu.pl/forum/viewforum.php?f=216

https://www.youtube.com/watch?v=JDav...1&feature=plcp

https://www.youtube.com/watch?v=eWqP...2&feature=plcp

http://dl.dropbox.com/u/72313101/kub...M-redehead.png

greetings

e X t 7 3

If you have anything to contribute to the os-jitter discussion, please do. However jitter seems often to be compromised by wattage.

Peace Be With You.

[Paradox Ethereal]

Thanks at least for this full config.
Some list of the specific options you think must be set would be nice, e.g. a diff to the distro config.
Also, I wonder where do you get the CONFIG_HZ_90 option, when vanilla kernel has a 100Hz option. Also I wonder why the lowest value is best, when the kernel config says the highest (1000Hz) option is for low-latency.

You can edit Kconfig.hz, in kernels, kernel directory. When I am done tweaking, and feel I have sufficently tried and gathered all information on this, I will do some patches.

A lot of people do believe 1000hz is the best, and I was recently told to use 10.000hz and BFS for "lowest latency". This is not true at all, infact a setting of 20hz with BFS, will not be noticably different from 10.000 except cpu-usage. With CFS though, 90hz gives the least jitter. Also the value may not scale, as the kernel code says the scaling is a bit unpredictable. It gives the lowest jitter on my dualcore though. I would strongly advise against 1000hz on the desktop, even though many believe that to be the best.

Peace Be With You.

[JanC]

Noob. Beer is very good at preserving itself without any additives and refridgeration, which is exactly why it is not kept in a freezer in the super market. I know because I tried one - even a unfiltered one where the guaranteed time is shorter, but still in the range of months - about two years after that date and it was still good as ever.
Prerequisites:
- bottled in glass bottle, not aluminium can (though that should also work), or worst: PET
- not opened
- preferably strong (the stronger the better, obviously, as less germs survive)
- brewer has to know what he was doing

... so in short, US beer is out

The oldest beer I drank was > 20 years old. And it was unpasteurised & unfiltered (a flemish "oud bruin"). It was a little "flat" tasting after so much time, but still tasted okay, and I didn't get sick...

And the "Abt" trappist ale from West-Vleteren is often at its best at 6-8yo (bottle refermentation and an artisanal brewing process make the result slightly unpredictable, but it's very unlikely it will be bad after 8 years).

[JanC]

PS: Even games were made in basic on c64, that had no slowdowns. Imagine interpreted basic, on a 1mhz CPU, can do lower-jitter than a modern pc. (well on many PCs )

The OS on the C64 didn't support multitasking, so there weren't other processes competing for CPU time. That allows the application/game developer to have complete control over the CPU usage.

The same was mostly true on a PC with DOS (see all the amazing things people from the demoscene did on that platform).

[JanC]

(...)
Throughput benchmarks should only be included as an afterthought, if at all. They don't measure the important variable.

Throughput benchmarking is important too, but mostly to make sure that after confirming that the changes give you low jitter, it doesn't impact throughput too much. You don't want nearly no jitter @ 1 FPS...

(It's like after finding a new medicine, you want to make sure it doesn't have too many side-effects.)

[Paradox Ethereal]

The OS on the C64 didn't support multitasking, so there weren't other processes competing for CPU time. That allows the application/game developer to have complete control over the CPU usage.

The same was mostly true on a PC with DOS (see all the amazing things people from the demoscene did on that platform).

The "OS" in the c64? It had no OS packaged, but it did have GEOS etc. Not so graceful in the low c64 res though. Opening a whole windowing system and moving a pointer, just to launch an app on the c64 made it much a waste for it.. With a bit of resolution and resources it gets good though

The whole idea with low-jitter is to get that "no OS" experience. Making the OS completely transparent to the user.

Remember that "multi-tasking" are high-level concepts related to programming paradigms, context switches etc. You CAN run "multiple apps" on a c64 aswell, but you have to write them like that.

Peace Be With You.

[V10lator]

Kano, I do, and you should try Mega Apocalypse to get an idea of how fun a low-jitter game can be. And then boot back in to high-level OS paradigms where there is latency, and get an idea of what I think of those.



I post a lot there. Was there anything in specific? I can post the whole .config for 3.6.2 shaved local, which you can take inspiration from. PS, not full distro config.

Peace Be With you.

http://paradoxuncreated.com/tmp/_config3_6_2local
(I also did some edits, but they are documented on the blog)

Could you diff this with a config without your low-jitter changes?

infact a setting of 20hz with BFS, will not be noticably different from 10.000 except cpu-usage.

I tested exactly this some days ago! I didn't do much testing, just booted with a kernel with 1000 hz and one with 100 hz and uses the same games. Games with slight input lat at 1000 hz had high input lag at 100. Note that this test was done with BFS, but I can't really believe that the result should turn around by using CFS...

[Paradox Ethereal]

Input lag? Are you aware that my da converter has no clicks on my machine at 0.3ms? How do you relate that to "hz"? When I type, things are FAST. The OS is ultrasnappy.

Ofcourse when you optimize for a high-performance bus, like a graphics bus, used in OpenGL, you can be sure that good things are happening.

Peace Be With You.