User: Password:
|
|
Subscribe / Log in / New account

Rethinking power-aware scheduling

Rethinking power-aware scheduling

Posted Jan 12, 2012 21:41 UTC (Thu) by jzbiciak (subscriber, #5246)
Parent article: Rethinking power-aware scheduling

And then there's the fun AMD's new Bulldozer platform brings with its core/module structure. It's not exactly hyperthreading, but not exactly different either. You also get the excitement of being able to run some modules at a higher clock if others are powered down.

Speaking as an outsider that's not really familiar with the guts here, it does seem like the governor might be a good place to decide what CPUs the scheduler should play with, and maybe what order it should fill them in. It seems like that would potentially allow you to factor out the CPU / system specific heuristics for hyperthreading, core/module, NUMA, etc. from the actual process of scheduling.


(Log in to post comments)

Rethinking power-aware scheduling

Posted Jan 12, 2012 23:16 UTC (Thu) by dlang (subscriber, #313) [Link]

you can't abstract all of this out, for several of these things there are performance implications to how you distribute the work across the cores that you have, and only the scheduler should be involved in that discussion. this is mostly factors around shared resources (cores for hyperthreading, FPU for bulldozer, memory I/O for NUMA, etc)

but the governor can know that if it shuts down some cores it can increase the clock on the remaining cores, and as a result it may choose to shut down some cores even if the remaining cores couldn't _quite_ handle the expected load at the current clock speed.


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds