Friday, 2010-01-08

« Thursday, 2010-01-07 Index Saturday, 2010-01-09 »
../irclogs/#mantishelp.2010-01-08.log
--- scribe started ---00:00
dhx_mpaul_____ & nuclear_eclipse: I've optimised view_all_bug_page.php to reach 14.5-15 requests/second (dual core)03:35
dhx_mjust by redoing the way we handle file inclusions03:35
dhx_mfor comparison, I was getting 4 requests/second (dual core) before changes03:36
nuclear_eclipsedhx_m: with or without something like php-apc?04:28
dhx_mnuclear_eclipse: with xcache (before and after)04:28
nuclear_eclipsewhat's xcache? :P04:29
dhx_mnuclear_eclipse: not sure what would happen with xcache off, I'll try :)04:29
dhx_mit's another PHP optimiser written by the lighttpd people04:29
nuclear_eclipseok04:29
nuclear_eclipseI already get pretty good results with php-apc on my vps install04:29
nuclear_eclipsedhx_m: how are you benchmarking?04:31
dhx_mab2 (comes with apache2)04:31
dhx_mseeing how many times per second I can load some MantisBT pages04:32
dhx_mto see if my changes are good/bad :)04:32
nuclear_eclipsemind running that against my install at leetcode.net/mantis?04:32
dhx_mI don't have ab2 on my server, so it'll be bandwidth bound (not CPU bound)04:32
dhx_mas my home connection is slooow04:32
nuclear_eclipseok, can you paste your ab2 command?04:32
dhx_mab2 -n 100 -c 2 http://localhost/mantis/view_all_bug_page.php04:33
dhx_mchange -c 2 to match the number of cores (or CPU threads, if you have an Intel CPU)04:33
dhx_m-n 100 is how many times to request the page04:33
dhx_mrun it a few times to get an average IMO04:34
dhx_mthe first one will be skewed as your PHP accelerator starts caching things04:34
nuclear_eclipseI'm getting about 25 req/s, with 4 cores on a small vps04:35
dhx_msounds about right, that's ~6 requests/second/CPU04:37
nuclear_eclipseI'm actually topping 30 req/s on my other install at bugs.steelwow.com, which only uses one plugin, so that's probably a big difference04:37
dhx_mhmm :)04:37
nuclear_eclipseleetcode.net uses quite a few plugins for vcs, irc, etc04:38
dhx_myeah that'd slow things down a bit04:38
nuclear_eclipsestill performs reasonably well, considering memory and cpu constraints of running on a virtualised system04:39
nuclear_eclipseanywho04:40
nuclear_eclipseI gotta get some sleep04:40
nuclear_eclipsecheers04:40
dhx_mok cya later :)04:41
dhx_mhmmm so there is no performance improvement10:58
dhx_mit's about a 40-50ms hit from what I'm seeing10:59
dhx_mbut the upshot is that it puts us in a position to make decoupling easier (and the results much faster to see)10:59
paul_____dhx_m: so slower?13:08
paul_____i'm confused:P13:08
paul_____is it faster or slow with require_api?13:08
dhx_mslower for me13:09
dhx_mbut my OS caches stat() calls13:09
dhx_mhowever given that Mantis isn't going to run any more than 50 times per second or so13:09
dhx_ma few stat() calls won't hurt I don't think ;)13:10
dhx_mI'm sticking with require_api for now though13:10
dhx_mactually correction...13:10
dhx_mrequire_api is faster than require_once in a benchmark (10000 iterations)13:10
dhx_mbut there really isn't anything in it13:10
dhx_mI suspect that benchmark is flawed like all others though ;)13:11
dhx_mI'll see if I can fix it13:11
paul_____patch file ready soon?13:23
paul_____bbiab13:23
dhx_mcommitting now ;)13:27
dhx_mjust writing a message13:27
dhx_m 263 files changed, 6778 insertions(+), 3138 deletions(-)13:32
dhx_mpaul_____: do you use APC... if so, I suspect lazy loading could have a large benefit?13:36
CIA-22Mantisbt: hickseydr * r3ffc2fa4b174 / (202 files in 2 dirs): require_api/require_lib13:37
dhx_mDAMNIT13:37
dhx_mso it commits the wrong thing :(13:37
CIA-22Mantisbt: hickseydr * rcf559e0d009b / (202 files in 2 dirs): Revert "require_api/require_lib"13:40
dhx_mthat's better now13:40
CIA-22Mantisbt: hickseydr * r2d5455cce290 / (263 files in 2 dirs): Use require_api/require_lib to pull in dependencies13:40
CIA-22Mantisbt: hickseydr * r77f2e2ae91b8 / (348 files in 17 dirs): Fix capitalisation in header license comment13:48
paul_____dhx_m: I thought you was gonna show me patch :P13:51
paul_____as if i need to merge in comments stuff ;/13:52
dhx_mI just did :p13:52
paul_____:(13:52
dhx_mcherry-pick?13:52
paul_____i.e. my approach last night was to revert your patch13:52
paul_____run patch <13:52
paul_____fix comments13:52
dhx_mcomments don't need fixing13:52
paul_____depends if phpdoc whinges13:53
dhx_mcommenting every require_once is stupid13:53
paul_____if phpdoc whinges, comments get fixed13:53
dhx_mI don't even think that is what phpdoc was complaining about13:53
dhx_m?13:53
paul_____nah it is ;p13:53
paul_____as i've had this before :P13:53
paul_____on a different note13:53
paul_____-# MantisBT - a php based bugtracking system13:53
paul_____+# MantisBT - A PHP based bugtracking system13:53
paul_____I think we should replace 'bugtracking' with 'issue tracking'13:54
dhx_mI thought it was complaining about not knowing whether the master file comment belonged to the file or to the first line of code (in this case, require_once)13:54
dhx_min time, I agree13:55
dhx_mwe're not really ready to call MantisBT a proper "issue tracker" though IMO13:55
paul_____we changed source from bug to issue ages ago13:56
paul_____for whats visible to users ;p13:56
paul_____MantisIT.org13:56
paul_____heh13:56
paul_____mm13:56
paul_____hmmm13:56
dhx_m:)13:58
CIA-22Mantisbt: hickseydr * rffb2a55f27f9 /proj_doc_page.php: Fix #11361: Allow 'bytes' to be localised on proj_doc_page.php14:00
CIA-22Mantisbt: hickseydr master-1.2.x * r7b27c02aa7d2 /proj_doc_page.php: Fix #11361: Allow 'bytes' to be localised on proj_doc_page.php14:00
CIA-22Mantisbt: hickseydr * rb9396c147802 /core/string_api.php: Fix #11358: Sort by dropdown in view filters shows column IDs, not names14:24
CIA-22Mantisbt: hickseydr master-1.2.x * r6e8d2751ae30 /core/string_api.php: Fix #11358: Sort by dropdown in view filters shows column IDs, not names14:25
paul_____so15:49
paul_____dhx_m: *Sigh*15:57
paul_____dhx_m: we need to call set_include_path....15:57
« Thursday, 2010-01-07 Index Saturday, 2010-01-09 »

Generated by irclog2html.py