Attachments
No attachments.
Duplicates
Comments
|
viraptor
wrote on the 12 Sep 09 at 10:58
|
|
|
Isn't it a bit early to talk about it? Sure - it's a very nice scheduler and some people may have fun trying it out.
But others report system not booting, filesystem corruptions, etc. Con's patches are not even close to "stable". If anyone still remembers about his patches at the end of this year, it might be a good time to ask that...
|
|
|
I'm excited to see interest in Desktop responsiveness, and I applaud Con's efforts to provide an alternative to the status quo, which assumes I have better hardware than I do.
That said, Con has explicitly said that he has no plans to support this code, and has acknowledged that it isn't ready for production use. Ubuntu would be wise to get its ducks in a row before committing to support this.
According to the Swap FAQ (link below), the default swappiness setting is 60, but this was virtually unusable on my desktop system. For desktop users interested in responsiveness wins, changing desktop Ubuntu's default to be lower would probably be a much more straightforward solution.
https://help.ubuntu.com/community/SwapFaq#Performance%20tuning%20with%20''swapp iness''
|
|
cos
wrote on the 13 Sep 09 at 23:19
|
|
|
Of course, our one-to-rule-them-all wonder-scheduler that runs on routers, phones, PCs, servers and supercomputers. Ridiculous!
Con's work is experimental, and is meant to help Linus et al understand that the scheduling needs of a desktop system are very different from those of a server. Desktop Linux is in dire need of a custom scheduler, but it doesn't sound like the kernel team will provide that. Development on that front will probably need to come from the distros and pushed back up into the kernel.
|
|
cos
wrote on the 13 Sep 09 at 23:22
|
|
|
|
Oh, of course -- GoogleOS! They might come up with something (don't know if they already have done anything about this with Android that we haven't noticed).
|
|
|
Desktop responsiveness is the only thing missing in Linux.
+1000 for this
|
|
adotei
wrote on the 19 Sep 09 at 12:53
|
|
|
Bfs should certainly be looked at. The android developers have started playing around with it.
http://en.wikipedia.org/wiki/Brain_Fuck_Scheduler
I have recompiled my 2.6.31 kernel with the bfs scheduler and it seems to be more responsive.
I haven't run any scientific test but if anyone has any suggestions, please let me know. Bfs should certainly be looked at.
|
|
Djhg2000
wrote on the 23 Sep 09 at 08:10
|
|
|
It would be insane to implement BFS as default at this stage!
Perhaps a check box after initial reboot like this:
"Enable BFS?
Please note while BFS may improve system performance, it will break some applications and/or decrease stability of your system. You can change this option later in the [insert menu here]"
|
|
Djhg2000
wrote on the 23 Sep 09 at 23:39
|
|
|
Sorry AndrewLuecke, but I meant we should NOT use BFS unless the user know what he/she is doing. :D
And I agree; prompting after initial boot is a really bad idea.
Seriously, what was I thinking?
|
|
Edman274
wrote on the 26 Sep 09 at 18:41
|
|
|
|
BFS doesn't even improve performance in regular desktops! On multicore systems BFS slows down requests and makes responsiveness *slower*! And I disagree that responsiveness is reduced by a bad scheduler: I've noticed that overall responsiveness is best when video drivers are optimized for the platform they're running on.
|
|
tdomhan
wrote on the 29 Sep 09 at 19:18
|
|
|
the kernel developers already make changes to the upstream kernel as a reaction to BFS.
the average user doesn't know what a scheduler is. he probably would be confused about an option.
|
|
cos
wrote on the 9 Oct 09 at 14:34
|
|
|
I agree that optimised video drivers can make more of a difference in the short-term -- most notably, ATI drivers introduce various delays when minimising and maximising windows with compiz, which are very annoying.
That aside, however, in the long run desktop Linux will need a desktop scheduler.
|
Post your comment
|