A New BFS "Smoking" Scheduler For Linux 3.3

[Paradox Uncreated]

dude, those who treat others with disrepect should NOT expect it back. Sorry, if your feelings are hurt, but i don't care.

and just because you write 'peace' at the bottom of each message doesn't mean anything, as the rest of your post is usually pretty brutal and ignorant.

and if you want me to greet you with peace, i suggest not being an asshole in your posts.

Lol, I have a real life. If that makes me an asshole to you, then it be so. Go with your fervent tweaking, and analog distortion. And people who know better, will follow their own bliss.

Peace.

Note: Just in case you think your NEXT post will somehow persuade me more, no. It won`t. Just kill that idea completely dead right there.

[ninez]

Lol, I have a real life. If that makes me an asshole to you, then it be so. Go with your fervent tweaking, and analog distortion. And people who know better, will follow their own bliss.

who said anything about you not having a life, and what does that have to do with anything?

As i said, i tweak my systems on installation, so there is no 'fervent' tweaking. I have that all down. run a few tests, change a few values and done. As for why your going on about analog distortion, i have no idea ~ but clearly you've never worked in a Studio with high-end gear, something tells me you also never went to school (college/university) for either music and/or production, so you probably aren't one of these 'people who know better' that you speak of, and instead are just a pretentious jackoff.

Note: Just in case you think your NEXT post will somehow persuade me more, no. It won`t. Just kill that idea completely dead right there.

I could care less, i don't expect to persuade you of anything. it's pretty obvious that you don't really know what your talking about.

[kraftman]

https://lkml.org/lkml/2012/3/25/38

I think those benchmarks really give a good indication of the current state of CFS vs BFS.

CFS is actually more fair now, at the cost of worst case latency. It does have higher overall throughput.

BFS is somewhat unfair (but still generally within a percent or two of being completely fair) and yet has much lower latency in some test (as much as 100% less latency).

What test are you talking about? What I'm seeing latency is about the same in both.

[liam]

analog gear does sound great, and typically much warmer than digital (that's a well-known FACT) and i don't play video games. so no cigar.
As for being a 'hyterical enthusiastic tweaker' ~ seriously dude, you have your pretentious little head rammed so far up your ass, all i can do is laugh at you, at this point. WTF are you even going on about? oh ya, i forgot this is a prime example of your 'ranting'.



you're just trolling, and clearly it is you who is hyterical - i'm just logged into work, relaxing, getting a few things done before tomorrow,

nice try though, it's pretty obvious at this point that you are just trolling. You really haven't said anything relevant to this post in your last few posts, and didn't even bother to reply to most of what i have said.
grow up.

Hey ninez,

Why are you debating with this guy? He doesn't seem like the type to listen. You provided a pretty detailed response and then he ad hom'd you.
Anyway, you should check out Monty's (of xiph.org and various codecs) rather long post about sampling above 44k and analog vs digital. It is quite an interesting read and has lots of references to his assertion that it never makes sense to sample at 192k or at 24bit depth.
If you read it let me know what you think.

Best/liam

[evolution]

Well, I'm going to give my personal experience of BFS:

For gaming/compiling and video playback (in a 9 year old P4C), BFS is slightly better then CFS. For instance, if I'm loading a web page and watching a HD-video on mplayer-vaapi, while the browser loads the page, the audio of my video stutters a lot with CFS, whereas with BFS, audio stutters much less.

Compiling the linux kernel for my P4 architecture: On average, with CFS, it takes about 35 minutes, whereas with BFS takes about 30/32 minutes.

Gaming: CFS gives me better glxgears results (900 FPS vs 750 FPS on BFS), but gaming on nexuiz and xonotic, I have, on average, +/- same fps with BFS (with medium details).

I/O workload: Here CFS is better (I get on average +4/5 MB/s of throughput in my ext4 linux partition (Copying/Moving data from/to a Windows NTFS partition)).

So, all in all, BFS is better suited for my (basic desktop) needs, whereas CFS might be better suited for server workloads...

My 2c, cheers!

[russofris]

Well, I'm going to give my personal experience of BFS:

For gaming/compiling and video playback (in a 9 year old P4C), BFS is slightly better then CFS. For instance, if I'm loading a web page and watching a HD-video on mplayer-vaapi, while the browser loads the page, the audio of my video stutters a lot with CFS, whereas with BFS, audio stutters much less.

How did BFS compare to CFS + nice?
How did BFS compare to CFS + a decreased sched_granularity_ns

I understand that setting the niceness of a browser-flash-plugin may be...... problematic (there's a better word than problematic, but it escapes me).

F

[Merkil]

Yo, peace mah man. Me an' mah homies are chillin' an' readin' thah forums. Yo, check it out man, people are writin' "peace" cuz it makes 'em look cool.

You made my day twice on this thread

Anyway, I would enjoy seeing some benchmarks about the BFQ disk scheduler. I've always wondered how it performs comparing to the default CFQ on recent kernels.

[evolution]

How did BFS compare to CFS + nice?
How did BFS compare to CFS + a decreased sched_granularity_ns

I understand that setting the niceness of a browser-flash-plugin may be...... problematic (there's a better word than problematic, but it escapes me).

F

Well, the results I'm mentioning here are with kernel "defaults" (I didn't change any of those paramenters in "/proc/sys/", but I can test them...), the kernel compilation was done with "time make -j2" (P4 with HT) and time "make -j2 install".

I was talking about doing playback of a full-HD video in mplayer-vaapi with 5.1 software mixing (I've a HD4650AGP that supports it) and, at the same time, doing some basic browser navigation with Opera+Flash...

Cheers

[del_diablo]

If the latency is 0.33ms, max time for short tasks = 0.33ms, and it is displayed the next frame. You cannot feel the difference here.

But we can measure 16 milliseconds of time. Which is roughly the time it takes between frames on a "normal" display. Well, 8 is starting to get more common as the 3D rage pushes up the monitors framerates.
A 0.33 ms advantage per task, perhaps 40-50 times a second, that will after some time produce really large numbers. And lets for arguments sake say we play some game, perhaps a emulator, where after some time the difference in troughput for just that application will be felt measurable. Uneven audio playback, cutscenes takes longer to finish, loading takes longer.
Even if what we have at hand can not be measured by us directly, we do feel it. Just like that we can perceive the difference of 58-62 uneven framerate vs 60 stable frames.
The reason BFS is better than CFS is because we do not care about all those background applications that is running when we are doing something, what we care about is not them getting troughput, we care about the pointer and the application window getting updates and redrawed exactly as many times as the desktop refreshes each second. Even post 12 points at this important aspect.