BFS 426 Scheduler Released For Linux 3.7 Kernel

[phoronix]

Phoronix: BFS 426 Scheduler Released For Linux 3.7 Kernel

Version 426 of the BFS scheduler was released this weekend for the Linux 3.7 kernel...

http://www.phoronix.com/vr.php?view=MTI1NDE

[talvik]

I've started using it recently on my desktop (i5 2500 kernel 3.5). The difference to CFS is absurd, with the standard CFS and running with light iowait the desktop starts becoming unresponsive. If I start something like a file transfer or a backup, the desktop becomes virtually unusable.

With BFS I don't even notice something like a backup running. BFS makes Firefox almost seem responsive

[signals]

I've started using it recently on my desktop (i5 2500 kernel 3.5). The difference to CFS is absurd, with the standard CFS and running with light iowait the desktop starts becoming unresponsive. If I start something like a file transfer or a backup, the desktop becomes virtually unusable.

Just out of curiosity, what distro are you running? I've never noticed a problem with CFS, and I've never been able to see a difference with BFS on my i7 3930k or my i7 920. But, I'm running a fairly lightweight desktop (awesome) under Gentoo, and I'm using SSDs for the root partition on both machines. I also haven't looked into BFS since the 2.6.X days, so my experience may be out of date.

I suppose I should emerge one of the kernel sources that includes BFS and give it another shot...

[nej_simon]

I've started using it recently on my desktop (i5 2500 kernel 3.5). The difference to CFS is absurd, with the standard CFS and running with light iowait the desktop starts becoming unresponsive. If I start something like a file transfer or a backup, the desktop becomes virtually unusable.

With BFS I don't even notice something like a backup running. BFS makes Firefox almost seem responsive

A lot of people seems to have that problem with some kernels / hardware configurations. I'm not sure the design of CFS is to blame. For ex. on my old slow single core Atom N280 netbook, running a 2.6.35 kernel with CFS, I could run apt-get dist-upgrade and the system would still be responsive. But on my current much more powerful dual core AMD E350 laptop with current 3.x kernels stuff like apt-get, update-apt-xapian-index and even dropboxd can easily cause some serious lag.

[disi]

A lot of people seems to have that problem with some kernels / hardware configurations. I'm not sure the design of CFS is to blame. For ex. on my old slow single core Atom N280 netbook, running a 2.6.35 kernel with CFS, I could run apt-get dist-upgrade and the system would still be responsive. But on my current much more powerful dual core AMD E350 laptop with current 3.x kernels stuff like apt-get, update-apt-xapian-index and even dropboxd can easily cause some serious lag.

I think the problem is reoccurring, all the same hardware (i7 with SSD etc.) with Fedora 16/17/18.
copy ~20GB from A to B locally -> freezes video playback for 1-2 seconds and/or input devices now and then
copy ~2GB over the network (NFS) -> freezes video playback for 1-2 seconds and/or input devices now and then
Then a couple of updates later the problem disappears. A few updates later or distro upgrade and the problem is back...

The i7 is not as good in multitasking as the A10 for example, but this shouldn't happen IMHO. It doesn't happen all the time, but comes and goes...

[bwat47]

I think the problem is reoccurring, all the same hardware (i7 with SSD etc.) with Fedora 16/17/18.
copy ~20GB from A to B locally -> freezes video playback for 1-2 seconds and/or input devices now and then
copy ~2GB over the network (NFS) -> freezes video playback for 1-2 seconds and/or input devices now and then
Then a couple of updates later the problem disappears. A few updates later or distro upgrade and the problem is back...

The i7 is not as good in multitasking as the A10 for example, but this shouldn't happen IMHO. It doesn't happen all the time, but comes and goes...

I've never had any of these problems and I have a laptop with an i5 and a hybrid hdd...

[varikonniemi]

In addition to the primary design goals of the bfs, increased desktop
interactivity and responsiveness, kernels patched with the ck1 patch set
including the bfs outperformed the vanilla kernel using the cfs at nearly all the
performance-based benchmarks tested.

I think the BFS scheduler has proven its merit in desktop usage, and using CFS for anyhing less than multiprocessor environments seems counterintuitive. I don't know enough about the inner workings of CFS to give more critic about it, but usually KISS applies.

[halo9en]

It would be interesting to see how BFS compares to the latest CFS now that the fixes & io patches have been added in 3.9. The only benchmark I could find @ Phoronix is dated Sep 2009.

[RealNC]

It would be interesting to see how BFS compares to the latest CFS now that the fixes & io patches have been added in 3.9. The only benchmark I could find @ Phoronix is dated Sep 2009.

BFS has not been ported yet to 3.9. This usually happens about a week or two after the final comes out.

[kernelOfTruth]

It would be interesting to see how BFS compares to the latest CFS now that the fixes & io patches have been added in 3.9. The only benchmark I could find @ Phoronix is dated Sep 2009.

I compared 3.8* with BFS to 3.9-rc4/rc5 with CFS and it seemed that 3.8* with CFS and following patch (sched: Fix select_idle_sibling() bouncing cow syndrome) was better

at least in my opinion


it might have gotten better now - didn't follow if additional patches got added


3.8.* with BFS and BFQ is current best for me in terms of interactivity

3.9 with CFS had still noticably longer delays and stuttering - but it's getting significantly better

so it might only take some more kernel releases and it's "acceptable"