From: Tim W. <ti...@sp...> - 2000-12-22 06:48:12
|
Interesting... The paper seems to have been quite thoroughly and carefully prepared. However... 1) It applies only to single-CPU systems. 2) It applies to the 8259-PIC - not exactly a high-performance part, and hopefully not relevant to performance-critical drivers on an SMP system. If this sort of thing is the rationale for eschewing interrupt priority levels then that would seem to be flawed in the same way as slavishly implementing them in that they may or may not help depending on the relative performance characteristics of the system. Judging from the paper, it certainly didn't make sense to use them on systems employing the 8259, but that doesn't mean that it doesn't on other (arguably better-designed) systems. x86 systems with an IO-APIC may or may not make the cut :-) Tim On Fri, Dec 22, 2000 at 02:38:23AM +0100, Andi Kleen wrote: > On Thu, Dec 21, 2000 at 05:13:08PM -0800, Tim Wright wrote: > > On Thu, Dec 21, 2000 at 11:51:39PM +0100, Andi Kleen wrote: > > > I doubt that there would be many chances to get generally visible spl levels into > > > Linux -- Torvalds et.al. have a near religious aversion against them. > > > > > > > Yes, I sort of gathered that. I've not seen any explanation of this antipathy. > > One argument is e.g. > http://www.usenix.org/publications/library/proceedings/ana97/full_papers/small/small.html > > > -Andi -- Tim Wright - ti...@sp... or ti...@ar... or tw...@us... IBM Linux Technology Center, Beaverton, Oregon "Nobody ever said I was charming, they said "Rimmer, you're a git!"" RD VI |
From: Tim W. <ti...@sp...> - 2000-12-21 21:16:39
|
Interesting. Has there been any discussion recently, and do you think that the attitudes may have softened in the interim ? Regards, Tim On Thu, Dec 21, 2000 at 11:36:52AM -0800, Tim Hockin wrote: > Nick Pollitt wrote: > > > > Here are some various code snippets for doing process pinning. We use 'runon' > > which has the following syntax: > > runon processor command [args] > > or > > runon processor -p pid > > Please note that I did a port of the sysmp() system call and pset tools > to linux quite some time ago - http://isunix.it.ilstu.edu/~thockin/pset. > > Linus and Alan generally rejected the patch as "bloat". > > -- > Tim Hockin > Software/OS Engineer > Sun Microsystems, Cobalt Server Appliance Business Unit > th...@co... > > _______________________________________________ > Lse-tech mailing list > Lse...@li... > http://lists.sourceforge.net/mailman/listinfo/lse-tech -- Tim Wright - ti...@sp... or ti...@ar... or tw...@us... IBM Linux Technology Center, Beaverton, Oregon "Nobody ever said I was charming, they said "Rimmer, you're a git!"" RD VI |
From: Andi K. <ak...@su...> - 2000-12-21 21:21:01
|
On Thu, Dec 21, 2000 at 01:16:21PM -0800, Tim Wright wrote: > Interesting. > Has there been any discussion recently, and do you think that the attitudes > may have softened in the interim ? The magic word is Tux -- the SMP tux performance is only possible with thread/cpu binding and manual irq->cpu APIC tuning (already in 2.4) -Andi |
From: Tim W. <ti...@sp...> - 2000-12-21 21:26:55
|
Good ! that means we have at least one real-world example that verifies the validity of pursuing this area. Tim On Thu, Dec 21, 2000 at 10:20:58PM +0100, Andi Kleen wrote: > On Thu, Dec 21, 2000 at 01:16:21PM -0800, Tim Wright wrote: > > Interesting. > > Has there been any discussion recently, and do you think that the attitudes > > may have softened in the interim ? > > The magic word is Tux -- the SMP tux performance is only possible with > thread/cpu binding and manual irq->cpu APIC tuning (already in 2.4) > > > > -Andi -- Tim Wright - ti...@sp... or ti...@ar... or tw...@us... IBM Linux Technology Center, Beaverton, Oregon "Nobody ever said I was charming, they said "Rimmer, you're a git!"" RD VI |