|
|
Subscribe / Log in / New account

Facebook and the kernel

Facebook and the kernel

Posted Apr 10, 2014 20:08 UTC (Thu) by dchichkov (guest, #90878)
In reply to: Facebook and the kernel by zblaxell
Parent article: Facebook and the kernel

I'd love to say a bit more to the kernel: "Stay out of the way for that particular tight-loop user-space process that is running on an isolated CPU. Do not call it, it will call you. Do not try to steal or remap physical memory from it. Do not do numa rebalancing, it will cause soft page faults. Try to accelerate and extend RCU grace periods as much as you can. Do not interfere with L1,L2 and try not to pollute L3. If I'm looking on 'perf top', for that particular isolated CPU, you shouldn't be nowhere near the top for both cache and CPU counters. Nor should you interrupt that process for more than 100,000 CPU cycles."


to post comments

Facebook and the kernel

Posted Apr 10, 2014 20:21 UTC (Thu) by mpr22 (subscriber, #60784) [Link]

I believe Linux already supports a sizeable slice of that list thanks to mlockall(), sched_setscheduler(), and sched_setaffinity().


Copyright © 2025, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds