Re: [PATCH] sched: prevent high load weight tasks suppressing balancing

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Siddha, Suresh B wrote:
This breaks HT and MC optimizations.. Consider a DP system with each
physical processor having two HT logical threads.. if there are two runnable processes running on package-0, with this patch scheduler will never move one of those processes to package-1..

Is this an active_load_balance() issue?

If it is then I suggest that the solution is to fix the active_load_balance() and associated code so that it works with this patch in place.

It would be possible to modify find_busiest_group() and find_busiest_queue() so that they just PREFER the busiest group to have at least one CPU with more than one running task and the busiest queue to have more than one task. However, this would make the code considerably more complex and I'm reluctant to impose that on all architectures just to satisfy HT and MC requirements. Are there configuration macros or other means that I can use to exclude this (proposed) code on systems where it isn't needed i.e. non HT and MC systems or HT and MC systems with only one package.

Personally, I think that the optimal performance of the load balancing code has already been considerably compromised by its unconditionally accommodating the requirements of active_load_balance() (which you have said is now only required by HT and MC systems) and that it might be better if active load balancing was separated out into a separate mechanism that could be excluded from the build on architectures that don't need it. I can't help thinking that this would result in a more efficient active load balancing mechanism as well because the current one is very inefficient.

Peter
PS I don't think that this issue is sufficiently important to prevent the adoption of the smpnice patches while it's being resolved.
--
Peter Williams                                   [email protected]

"Learning, n. The kind of ignorance distinguishing the studious."
 -- Ambrose Bierce
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux