Has Anyone Tried FreeBSD BFS scheduler?

I stumbled across a blog, on which a user has made an attempt to port the Brain *beep**beep**beep**beep* Scheduler to FreeBSD, also known as FBFS! There was some nice benchmarks and videos showing the scheduler in use, and tested against 4BSD and ULE.

Now the only machine that seems to benefit from FBFS is a dualcore machine, but on an 4core plus machine, ULE seems to be dominant. While 4BSD seems to win over ULE and on similar grounds to FBFS on dualcore or less machines:

Sysbench Postgresql benchmark on a 2 CPU machine

plot_sysbench_2cpu.gif


Sysbench Postgresql benchmark on a 8 CPU machine

plot_sysbench_8cpu.gif


Sysbench Postgresql benchmark on a 4 CPU virtual machine

plot_sysbench_4_virtual.gif


So I figure, each scheduler benefits different machines. Seems like ULL is best for multiple cpu environments, while FBFS and 4BSD better for single core environments.

I hope you find this information useful: The project of fBFS
 
Interesting, but I'd suggest it is of limited usefulness. Given that we have mobile phones and tablets starting to ship with 4 core CPUs now it's probably of limited use to go chasing single or dual core scheduler optimizations.

2c.
 
throAU said:
...Given that we have mobile phones and tablets starting to ship with 4 core CPUs now it's probably of limited use to go chasing single or dual core scheduler optimizations.

The cheap embedded market is still pretty strong.
 
I have used bfs on an old p4 machine and I can tell a very obvious difference in responsiveness, but not in speed.

For older, single-core hardware, I tend to believe that it does indeed help, especially under heavy load.

You can sacrifice some speed for less latency, and so on. When used in the right situation, I believe it can provide some benifit. As in the case of the single-core p4 desktop, for example.

I'm just speculating however. I don't have any actual data to back this up.

Gentoo has a "ck-sources" kernel with BFS.
 
Back
Top