Re: SMP & Interrupt Affinity
When it comes to the BFS scheduler, the term "snake-oil"
comes to mind. For starters the BFS scheduler really doesn't post any gains under system benchmarking when compared to the CFS scheduler
. Typically the lack of benchmarks to show that BFS actually does anything is met with responses that BFS is "about the feel of the system"
rather than any specific response time or measurable priority usage. However I've never been able to create any situation with any x86 processor; including Atom, Phenom AM2, Phenom II AM3, I7, Celeron D, A64 754, A64 939; where a kernel using BFS provided any noticeable improvement over the same kernel using CFS. The few times I've run a "blind"
test where the person I grabbed to sit at the keyboard and do stuff, they thought the kernel with the CFS was more responsive.
The impression I get with BFS is that people claim BFS is more responsive or better than CFS because they are -told-
it's more responsive or better than CFS.
I suspect that might also be why the main author behind BFS, Con-Man Kolivas, is extremely disinterested in having the scheduler looked at by Ingo, or any of the CFS scheduler contributors at IBM or Google.