August 2007 - Darwin-dev - lists.apple.com

Re: Poor performance of pthreads across cpu dies.
by Andrew Gallatin 29 Aug '07

29 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com driggett(a)mac.com writes: > The behavior is that if the threads that share common data are > running on the same physical cpu die then the performance is 3 to 4 > times greater than if they are split across two physical dies. I ran > the test shutting off one die per physical cpu and showed the same > results as if all cpus were active. > I've seen similar issues benchmarking 10GbE NICs, and I don't even need pthreads. The scheduler tends to run the user mode application on one core, the interrupt handler kernel thread (iokit "workloop") on another, and the network stack (dlil) kernel thread on yet another. I think I've actually seen worse performance than you because for me (on Tiger), I see up to a 25% increase on some benchmarks, and up to a few hundred percent on others, if I totally disable a CPU package on a dual dual core MacPro or Xserve. I think the fundamental problem is the scheduler doesn't have a clue about cpu affinity, and MacOSX is lacking any APIs or command line interfaces that would allow the app or admin to give it a clue (like you can on Linux, Solaris, etc). Good would be a scheduler with some notion of CPU affinity, and better would be a scheduler that allowed the user to give it some hints. Drew _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… This email sent to site_archiver(a)lists.apple.com

1 0

Re: Poor performance of pthreads across cpu dies.
by Terry Lambert 29 Aug '07

29 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Please do not post the same thing to multiple lists. -- Terry All, A = One worker thread, multiple support threads. B = two worker threads, multiple support threads. C = four worker threads, multiples support threads. If A runs on all cpus then the performance is 512 task units done. If A runs on 1 physical cpu (2 cores) the performance is 470. If B runs on all cpus the performance is 130. If B runs on one cpu (2 cores) the performance is 380. C has the same performance as B. Thanks, Chris Driggett driggett(a)mac.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/tlambert%40apple.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… You already posted this to the unix-porting mailing list, and it is being discussed there. Since it involves porting Linux code, that's the most appropriate list. On Aug 29, 2007, at 2:24 PM, driggett(a)mac.com wrote: I have a multithread application which has a number of threads. A certain number of these threads process data and share the same data structures. If I run this program on a 2 cpu dual core linux box I see no performance penalties. If I run the same program on a xserve with 2 xenon dual core cpus I see an 80% hit. So here is the scenario: The behavior is that if the threads that share common data are running on the same physical cpu die then the performance is 3 to 4 times greater than if they are split across two physical dies. I ran the test shutting off one die per physical cpu and showed the same results as if all cpus were active. So here is the question: Is there something special one needs to do using posix threads on Mac OS X Tiger vs Linux? The routines in question are pthread mutexes and cond_wait_timeout.. If you can point me to a technical doc or bulletin explaining the performance differences of pthreads on linux vs Mac OS X that would be great. This email sent to tlambert(a)apple.com This email sent to site_archiver(a)lists.apple.com

1 0

Poor performance of pthreads across cpu dies.
by driggett＠mac.com 29 Aug '07

29 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com All, A = One worker thread, multiple support threads. B = two worker threads, multiple support threads. C = four worker threads, multiples support threads. If A runs on all cpus then the performance is 512 task units done. If A runs on 1 physical cpu (2 cores) the performance is 470. If B runs on all cpus the performance is 130. If B runs on one cpu (2 cores) the performance is 380. C has the same performance as B. Thanks, Chris Driggett driggett(a)mac.com _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I have a multithread application which has a number of threads. A certain number of these threads process data and share the same data structures. If I run this program on a 2 cpu dual core linux box I see no performance penalties. If I run the same program on a xserve with 2 xenon dual core cpus I see an 80% hit. So here is the scenario: The behavior is that if the threads that share common data are running on the same physical cpu die then the performance is 3 to 4 times greater than if they are split across two physical dies. I ran the test shutting off one die per physical cpu and showed the same results as if all cpus were active. So here is the question: Is there something special one needs to do using posix threads on Mac OS X Tiger vs Linux? The routines in question are pthread mutexes and cond_wait_timeout.. If you can point me to a technical doc or bulletin explaining the performance differences of pthreads on linux vs Mac OS X that would be great. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Shutdown/Reboot/Logout problem.
by Uli Kusterer 27 Aug '07

27 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Cheers, -- M. Uli Kusterer http://www.zathras.de _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… On 27.08.2007, at 06:59, Draichis wrote: I'm not asking for your sympathy. The problem I have is a simple one. I am unable to shutdown/reboot or logout from the apple menu, and I'm able to via shutdown(8) in the terminal. I don't know the reasons for this. Linking me to the catb smart questions site. If there are any particular details you require, do say. Thanks for finally telling us what you're problem is. Though I'd like more detail: "Can't shut down" could be anything. Do you mean you choose the menu item and nothing happens, do you mean the menu item is disabled, do you mean all running apps quit but one doesn't, do you get any error messages like "application X canceled shutdown", do you mean you're left at an endlessly spinning cogwheel on a blue screen but the Mac never physically shuts down ??? Give us something to work with, already! So far one problem I see is that essentially you're saying that everything is fine at the Darwin level (which is what this list is for) but the Apple GUI level doesn't work (which is what this list isn't for). So that may be another reason why it's not too likely you'll get any useful answers here. Anyway, here are my guesses what could screw up your restarts: Have you checked whether you've installed any Input Managers ("SIMBL plugins") or APE haxies or so that could be buggy and cause problems? Do you have any apps that install Kernel Extensions that might have hung? Are there any applications that are still running, maybe showing error messages or asking you to save? This email sent to site_archiver(a)lists.apple.com

1 0

Re: Shutdown/Reboot/Logout problem.
by Draichis 27 Aug '07

27 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: a=rsa-sha1; c=relaxed/relaxed; d=gmail.com; s=beta; h=domainkey-signature:received:received:message-id:date:from:user-agent:mime-version:to:cc:subject:references:in-reply-to:content-type:content-transfer-encoding; b=IG65JzC6RRz79pFkKQ+l39fCn6fVCnGK2/o/ltibdnxIxeSCEfd+mSj0Tkmz5lBRmH3VbbvXM8p+lZDRexZuwb4KX1ft/bu2hKj41ALT7VS4kpUCFepiiTud44WJ6WWQO7DZP4KyVftNtGSen8gs2BlB3IfopScHq9vC39dDSFY= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:user-agent:mime-version:to:cc:subject:references:in-reply-to:content-type:content-transfer-encoding; b=A5GFEpumxy1tJ3CX4lHjkxZEBfmj0qProgHAylZ3qGW1NCXhrkJwttkKl4EFmkAm/76QnlXSnhZeGapSSUCkSCKBKXuP9cmwAJU/cHf78duyHceLFveCTyDl0Q10R3Khe5EPpjQ0WNGBAaMu+HhPwRfZ9ZIEf+aBEIY4Le21STk= User-agent: Thunderbird 1.5.0.10 (Macintosh/20070221) On Aug 26, 2007, at 10:32 PM, Draichis wrote: b) Why this would even occur? I can think of two reasons you are not getting any responses. See <http://catb.org/~esr/faqs/smart-questions.html>. — F Thankyou for posting the link, it's been a great help. -- --- Draichis draichis(a)gmail.com draichis(a)craplandia.org http://draichis.craplandia.org "I know that fewer people are won over by the written word than by the spoken word and that every great movement on this earth owes its growth to great speakers and not to great writers." _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… Fritz Anderson wrote: I think I may have, perhaps sent this earlier, but I didn't get any response. I have a problem with shutdown/logout/reboot on the apple menu. I've rebuilt the launchd services, and rotated my cron services. Still no avail; Could someone a) Tell me what the conventional means of fixing this would be and It's not a 'great' problem, but it's irritating.. 1: You posted this as a reply to a thread on a completely different subject. People who organize their mail readers by thread, and who might know the answer to your question, may never see it, because you buried it in a thread that says it's about launchd. Hitting "reply" and overwriting the subject line doesn't fool readers that follow In-Reply-To: headers. Use the Compose button and bear the burden of typing the list address yourself. 2: Your message fails to state a problem. You say, "I have a problem," but you don't say what the problem is. Nobody can answer you, except to offer sympathy, if you conceal what you mean to do, and exactly how you are prevented from doing it. Nobody can tell you "the means... of fixing this," when you don't say what "this" is. Fritz; I'm not asking for your sympathy. The problem I have is a simple one. I am unable to shutdown/reboot or logout from the apple menu, and I'm able to via shutdown(8) in the terminal. I don't know the reasons for this. Linking me to the catb smart questions site. If there are any particular details you require, do say. This email sent to site_archiver(a)lists.apple.com

1 0

Re: Shutdown/Reboot/Logout problem.
by Fritz Anderson 27 Aug '07

27 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 26, 2007, at 10:32 PM, Draichis wrote: b) Why this would even occur? I can think of two reasons you are not getting any responses. See <http://catb.org/~esr/faqs/smart-questions.html>. — F _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I think I may have, perhaps sent this earlier, but I didn't get any response. I have a problem with shutdown/logout/reboot on the apple menu. I've rebuilt the launchd services, and rotated my cron services. Still no avail; Could someone a) Tell me what the conventional means of fixing this would be and It's not a 'great' problem, but it's irritating.. 1: You posted this as a reply to a thread on a completely different subject. People who organize their mail readers by thread, and who might know the answer to your question, may never see it, because you buried it in a thread that says it's about launchd. Hitting "reply" and overwriting the subject line doesn't fool readers that follow In- Reply-To: headers. Use the Compose button and bear the burden of typing the list address yourself. 2: Your message fails to state a problem. You say, "I have a problem," but you don't say what the problem is. Nobody can answer you, except to offer sympathy, if you conceal what you mean to do, and exactly how you are prevented from doing it. Nobody can tell you "the means... of fixing this," when you don't say what "this" is. This email sent to site_archiver(a)lists.apple.com

1 0

Shutdown/Reboot/Logout problem.
by Draichis 27 Aug '07

27 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com Dkim-signature: a=rsa-sha1; c=relaxed/relaxed; d=gmail.com; s=beta; h=domainkey-signature:received:received:message-id:date:from:user-agent:mime-version:to:subject:references:in-reply-to:content-type:content-transfer-encoding; b=sDFxfbGal68VfHdFpsAm5OxOxP8r3JT55eCxa4By6m9Gukehy/gpCpsuoStN4Sw7Cag9lS2MGxe/2XuJBgPfQQZ24ihIcJ2f99IMZ8H9y8RfkvREZpg3L8sboyqW76OxpPVVyl7xwbO/Cq57mntWEwZU+sjljUOPeIqewp/SSxw= Domainkey-signature: a=rsa-sha1; c=nofws; d=gmail.com; s=beta; h=received:message-id:date:from:user-agent:mime-version:to:subject:references:in-reply-to:content-type:content-transfer-encoding; b=q/upAC9nA7FDhTYCD/Df0Xh1/an3dXU+K8c3Js5xTVgQiBUlLCvxwhHcXnhUbVw8YTZCmB/iF6mfC5y3CEYMqhFQlI/wtarq2h/V6G++SmuMJjtyEvBdEQiZnf2+dQ8pRdH8MVcGAltugLZ1zWXoADbzBcvCmyClf1aNh2rt804= User-agent: Thunderbird 1.5.0.10 (Macintosh/20070221) b) Why this would even occur? It's not a 'great' problem, but it's irritating.. Best regards. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… I think I may have, perhaps sent this earlier, but I didn't get any response. I have a problem with shutdown/logout/reboot on the apple menu. I've rebuilt the launchd services, and rotated my cron services. Still no avail; Could someone a) Tell me what the conventional means of fixing this would be and This email sent to site_archiver(a)lists.apple.com

1 0

Re: Questions about launchd
by Jordan K. Hubbard 26 Aug '07

26 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On Aug 26, 2007, at 12:44 PM, Terry Lambert wrote: - Jordan _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… But my point was that you could trust command line output a heck of a lot more than data interfaces. I more or less promise we won't intentionally break things unless the standards change or our upstream providers change things on us for no reason. That would be work. 8^). I'm really not sure what you mean by "data interfaces" in this context. but my point was that we don't promise anything with respect to "command API" (and never have), whereas we do make a substantial effort with actual APIs. Moreover, the fact that we discourage static linking also gives us a lot more leeway in providing API backwards compatibility via symbol decoration/versioning, app-specific conditional hacks, interpositioning tricks, etc. When it comes to the output of ps(1), however, we can't do a heck of a lot for you if you've buried it in some script and expect the contents of column 25 for arguments "-ef" to mean something very specific. Maintaining compatibility at that level has little to do with "avoiding work" and more to do with "avoiding any number of things which could potential perturb the output" (like, say, the PTY naming syntax changing). This email sent to site_archiver(a)lists.apple.com

1 0

Re: Questions about launchd
by Terry Lambert 26 Aug '07

26 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com -- Terry On 26 Aug 2007, at 20:44, Terry Lambert wrote: $ top top: could not open /dev/kmem: no such file or directory $ Cheers, Graham. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… We took away /dev/kmem for a reason. If you want it back, you can specifically ask for it using boot-args. But it's not there by default on purpose. On Aug 26, 2007, at 12:57 PM, Graham J Lee <leeg(a)thaesofereode.info> wrote: But my point was that you could trust command line output a heck of a lot more than data interfaces. I more or less promise we won't intentionally break things unless the standards change or our upstream providers change things on us for no reason. That would be work. 8^). If you absolutley must ask the OS for information you should already know, then the way to do it is via our tools that know how to ask. We won't be breaking our tools for no reason. That would be work. 8^). Unless someone changes the data interfaces those tools depend on, in which case getting your tool to stay the same involves work ;-). As an invented example: This email sent to site_archiver(a)lists.apple.com

1 0

Re: Questions about launchd
by Graham J Lee 26 Aug '07

26 Aug '07

site_archiver(a)lists.apple.com Delivered-To: darwin-dev(a)lists.apple.com On 26 Aug 2007, at 20:44, Terry Lambert wrote: $ top top: could not open /dev/kmem: no such file or directory $ Cheers, Graham. _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-dev mailing list (Darwin-dev(a)lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-dev/site_archiver%40lists.app… But my point was that you could trust command line output a heck of a lot more than data interfaces. I more or less promise we won't intentionally break things unless the standards change or our upstream providers change things on us for no reason. That would be work. 8^). If you absolutley must ask the OS for information you should already know, then the way to do it is via our tools that know how to ask. We won't be breaking our tools for no reason. That would be work. 8^). Unless someone changes the data interfaces those tools depend on, in which case getting your tool to stay the same involves work ;-). As an invented example: This email sent to site_archiver(a)lists.apple.com

1 0