BFQ Still Trying To Replace The CFQ I/O Scheduler In Linux

[FLHerne]

Pretty much anything will give you better worst-case responsiveness than CFQ (which is what matters - it's the 1% of multi-second stalls that you notice, not a few ms here and there).

At some point in the chain, yes. BFQ is as good as CFQ there, BFS is aimed at low core-counts with theoretical disadvantages on huge things with dozens of cores. Those don't seem to be a factor on realistic machines anyway.

[teresaejunior]

For some reason I found that no-op was providing me more responsive experiences on android smart phones than CFQ.

But I thought bfq was more centered around like unicore setups?... was that misinformation?

From a research I did months ago, flash storage in general won't benefit from I/O schedulers, the best would be noop for a phone.

[liam]

Following the previous ml thread was headache inducing.
Tejun just wasn't getting the difference between time vs bandwidth based scheduling (well, he thought, iirc, that bandwidth allocation ~ time-based allocation).
Aside from that (long) thread, there were at least a few cases (involving cgroups, again, iirc) where bfq causing severe regressions.

[crowen]

Thanks for posting this, I was stuck on Fedora with Linux 4.5 because I'm using BFQ and it only supported version 4.5. If you're running Linux on an HDD you gotta try BFQ. No glitches found so far.

Apparently your link is broken:
http://lkml.iu.edu/hypermail/linux/k...7.3/02456.html

You can just use the copr repo here: https://copr.fedorainfracloud.org/co...ter/kernel-pf/

[stan]

Michael, how come Phoronix never has any benchmarks with BFQ? I'd be very interested in some objective values.

[jwilliams]

Michael, how come Phoronix never has any benchmarks with BFQ? I'd be very interested in some objective values.

The benchmarks that Phoronix uses will not be very interesting with BFQ. Where BFQ makes a difference is when you have several different kinds of IO going on at the same time, and at least one of them is "interactive", meaning that the user will notice if there are large latency pauses.

Most of Phoronix's tests just have one type of IO going on, and mostly they depend on throughput, not latency. So you would probably see BFQ performing similar to CFQ for most of the benchmarks that Phoronix might do, since BFQ behaves like any other good IO scheduler in such situations.

[Axl_Mas]

Great news! I am using BFQ from years with a sensible difference in performances in desktop use, I really hope it will be integrated in vanilla kernel asap

[dimko]

CFQ will be transformed into BFQ to be exact.

He has a point there. BFS is best thing there is for desktop. So far it has been very stable for me.(untill my Asus mobo burned out a month ago)
Linus kicked arse of guy who is mantaining it, which is SHAME...
Oh well...

[spirit]

The lkml article have a link to have youtube video comparaison benchamark between differents io scheduler (on ssd)

https://youtu.be/1cjZeaCXIyM

[cj.wijtmans]

The lkml article have a link to have youtube video comparaison benchamark between differents io scheduler (on ssd)

https://youtu.be/1cjZeaCXIyM

Seems like i should use no-op till bfq.