00:32:21 -!- sellout [n=greg@63.118.137.211] has quit [] 01:54:03 sellout [n=greg@12.130.118.13] has joined #ccl 01:58:19 -!- alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has quit [] 02:19:07 alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has joined #ccl 03:06:49 -!- alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has quit [] 03:08:52 bfulgham_ [n=brent@adsl-69-234-109-59.dsl.irvnca.pacbell.net] has joined #ccl 03:13:04 alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has joined #ccl 03:13:50 -!- alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has quit [Client Quit] 03:18:58 alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has joined #ccl 03:34:57 -!- rme [n=rme@70.104.116.109] has quit [] 05:33:26 -!- bfulgham_ is now known as bfulgham|zzz 06:46:14 -!- sellout [n=greg@12.130.118.13] has quit [] 10:31:03 -!- jauaor [n=araujo@gentoo/developer/araujo] has quit [] 11:54:07 segv [n=mb@p4FC1BE70.dip.t-dialin.net] has joined #ccl 13:11:28 -!- vy [n=user@nbvyazici.cs.bilkent.edu.tr] has quit [Remote closed the connection] 15:00:02 anRch [n=markmill@nmd.sbx07269.sauguma.wayport.net] has joined #ccl 15:00:37 milanj [n=milan@93.86.186.18] has joined #ccl 15:21:29 -!- alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has quit [] 15:22:28 vy [n=user@88.227.48.158] has joined #ccl 15:24:08 alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has joined #ccl 15:47:25 Hi, I'm using PATRON (cliki.net/patron) and experiencing serious thread related problems on CCL platform. (Same performance gotchas don't appear on SBCL.) Here [http://www.students.itu.edu.tr/~yazicivo/tmp/patron-test.lisp] is a sample code with profiling output attached. The problem is, despite the oprofile output doesn't change that much, performance degrades nearly 40x while threading is turned on. Any ideas? 15:51:46 sellout [n=greg@75-25-126-88.lightspeed.sjcpca.sbcglobal.net] has joined #ccl 16:07:16 -!- alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has quit [] 16:17:22 -!- pem [n=pem@159.226.35.246] has quit [Remote closed the connection] 16:24:49 vy: yeah, ccl has really slow bignums. We have some ideas to improve them, but nobody's had the time to do so yet... 16:26:43 I don't know what threading would have to do with it, except perhaps if there is some attempt at synchronization. 16:32:05 alms [n=alms@146-115-42-237.c3-0.bkl-ubr1.sbo-bkl.ma.cable.rcn.com] has joined #ccl 16:33:06 -!- anRch [n=markmill@nmd.sbx07269.sauguma.wayport.net] has quit [] 16:33:38 gz: Setting worker-capacity to 1 (instead of 2), i.e. no threading, results in a 40x performance increase. I don't think it has anything to do with bignums. 16:37:16 I see. It's hard to say what'd be going on without understanding more about patron. But one somewhat suspicious thing I see is CCL::GAG. at 4% and GAG-TWO-ARG at 1%, these are part of gf dispatch and that seems really high. 16:39:35 Also, almost 1% for %clone-x86-function, I *think* that's only done when adding/removing methods. Does patron do something very dynamic with clos? 16:44:36 gz: That's actually impossible, since I don't know clos in that much depth. :-) 16:45:08 Does it redefine classes? 16:45:15 No. 16:46:46 Perhaps the oprofile includes the loading up of your system? Because there is stuff like UPDATE-OBSOLETE-INSTANCE in there that I don't think happens normally. 16:47:37 I start the oprofile via "opcontrol --start" and then run the program, as instructed in the CCL profiling manual. 16:48:20 The interesting thing is that there is nothing related with semaphores, locks and/or threading in the profiler output. Is that something expected? 16:50:06 gz: I simplified the test in http://www.students.itu.edu.tr/~yazicivo/tmp/patron-test.lisp, check it out. 16:50:16 We're getting to the limit of what I know here. But I believe for semaphores and such we're pretty much just calling into the OS, and the oprofile output may not include os time. 16:51:35 Well, %ptr-store-conditional is a locking thing. 16:53:41 Should I move the discussion to the mailing-list? 16:53:45 And again with the %clone-x86-function high up there. Hmm, maybe you can just run your test, without oprofile, and break at ccl::%clone-x86-function E.g. (trace :break-before t ccl::%clone 16:54:52 gz: But %clone-x86-function appears to be consuming %1.5 of the execution time. 16:55:29 yes, but it's an anomaly, at least as far as my understand of its use goes. Perhaps it's nothing, but seems worth investigating. 16:55:31 That doesn't sound like it could possibly account for the described degredation of performance. 16:55:49 When running the benchmark, what % of CPU does top report that ccl is consuming? 16:56:13 chandler: %100. 16:56:25 rme [n=rme@pool-70-104-116-109.chi.dsl-w.verizon.net] has joined #ccl 16:56:34 -!- sellout [n=greg@75-25-126-88.lightspeed.sjcpca.sbcglobal.net] has quit [Read error: 110 (Connection timed out)] 16:56:58 In threaded case, both cores are 100% busy; in single threaded case, just a single core appears to be busy. 16:57:24 In the threaded case, can you tell what % of time is spent in kernel time vs user time? 16:58:32 Can you do an oprofile run with the single threaded case? 17:00:41 chandler: http://www.students.itu.edu.tr/~yazicivo/tmp/timing.txt 17:00:47 gz: http://www.students.itu.edu.tr/~yazicivo/tmp/patron-1t.oprofile.txt 17:01:41 Well, there's where your time is going. 17:02:02 Now, the question is how to get oprofile to tell you why and where it's spending all its time in system calls. 17:04:46 I'm totally clueless. Any suggestions? 17:05:32 No help here. 17:14:59 jauaor [n=araujo@gentoo/developer/araujo] has joined #ccl 17:39:22 milanj- [n=milan@93.87.168.184] has joined #ccl 17:47:23 -!- milanj [n=milan@93.86.186.18] has quit [Read error: 110 (Connection timed out)] 18:13:43 -!- vy [n=user@88.227.48.158] has quit [Remote closed the connection] 18:40:16 -!- bfulgham|zzz is now known as bfulgham_ 19:10:53 -!- segv [n=mb@p4FC1BE70.dip.t-dialin.net] has quit ["quit"] 19:32:27 vy [n=user@88.227.48.158] has joined #ccl 20:24:56 -!- jauaor [n=araujo@gentoo/developer/araujo] has quit [Read error: 54 (Connection reset by peer)] 20:25:01 jauaor_ [n=araujo@190.38.50.207] has joined #ccl 20:25:19 gbyers: I still couldn't figure out how to decompress a vmlinuz, do you mind giving some pointers? 20:27:07 -!- jauaor_ [n=araujo@190.38.50.207] has quit [Client Quit] 20:27:27 jauaor [n=araujo@gentoo/developer/araujo] has joined #ccl 20:46:44 -!- jauaor [n=araujo@gentoo/developer/araujo] has quit [] 20:47:33 jauaor [n=araujo@gentoo/developer/araujo] has joined #ccl 20:48:05 -!- jauaor [n=araujo@gentoo/developer/araujo] has quit [Client Quit] 21:50:39 anRch [n=markmill@nmd.sbx07258.melroma.wayport.net] has joined #ccl 21:57:09 jauaor [n=araujo@gentoo/developer/araujo] has joined #ccl 22:01:49 -!- milanj- [n=milan@93.87.168.184] has quit ["Leaving"] 22:11:51 -!- anRch [n=markmill@nmd.sbx07258.melroma.wayport.net] has quit [] 22:15:16 -!- vy [n=user@88.227.48.158] has quit [Remote closed the connection] 22:53:24 cl-newb [n=john@c-67-183-22-64.hsd1.wa.comcast.net] has joined #ccl 23:17:59 How do I suppress the compiler warning that I have an unused lexical variable?