Date: Mon, 5 Apr 1999 12:08:31 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: Julian Elischer <julian@whistle.com> Cc: Alan Cox <alc@cs.rice.edu>, current@FreeBSD.ORG, luoqi@FreeBSD.ORG Subject: Re: SMP users (important) Message-ID: <199904051908.MAA12939@apollo.backplane.com> References: <Pine.BSF.4.05.9904042312550.282-100000@s204m82.isp.whistle.com>
next in thread | previous in thread | raw e-mail | index | archive | help
:> :julian
:>
:> I'd like to see this too. I will soon have two SMP boxes of my own to play
:> with for my own personal use and for an upcoming project, and at least one
:> will be available for SMP life-testing purposes for several months.
:> I really want to see two things: (1) Actual sharing of the physical pmap
:> between rfork(RFMEM|RFPROC)'d processes, and (2) Avoiding the %cr3 reload
:> ( which clears the TLB ) when switching between such processes.
:
:This would also suggest shceduler changes that would increasr the
:likelyhood of 'related' processes being scheduled together.
:
:One scheme I remember id the A-list/B-list scheme, where
:each scheduling priority has two list that are alternated.
:whichever list is being drained is not eligible for receiving new items.
:They must go to the other list. When the active list is drained and theya
:are switched, the new list is first sorted accoring to affinity
:related effects.
:
:julian
This can get real tricky. I think the scheduler will almost do this
anyway, just due to the way threads tend to go to sleep and get woken
up, so adding a lot of sophistication here may not help.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199904051908.MAA12939>