Difference between revisions of "Performance/Meetings/2008 01"
|Line 1:||Line 1:|
Latest revision as of 23:13, 17 March 2010
|About this template|
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000 and the Community ??
Time: 15:43– 17:29
(15:54:39) kuangliang: hello guys
(16:04:13) kuangliang: it's time to start!
(16:09:22) LiHeng: where is liangjun?
(16:09:44) yugq: just a minute
(16:10:01) liangjun: hello
(16:11:44) LiHeng: i just sell the alarm clock for you , later !!:(
(16:12:22) LiHeng: are you ready for this meeting?
(16:12:30) yugq: i'm sorry. i'll remember this from now on.
(16:13:44) LiHeng: please check agenda in mailbox
(16:14:16) yugq: maybe i can say more later when i did do many things. but now, i have little to say.
(16:14:42) yugq: :-(
(16:16:31) LiHeng: ok,i direct meeting this time
(16:16:54) LiHeng: please, talk about you idea
(16:19:02) LiHeng: i want to create independent CWS for performance project , do you think it's a right idea
(16:19:25) yugq: i think it's nessisary.
(16:19:31) LiHeng: why?
(16:19:46) LiHeng: in your points
(16:20:29) liangjun: :)
(16:20:58) yugq: because when you do some optimizations on OO, you are not sure the statable of OO.
(16:21:55) yugq: we need provide thease performance versions to test stuff and let them do some test.
(16:22:29) yugq: then we can put this optimization wins to release versions.
(16:23:02) LiHeng: ok,please show these words in meeting soon
(16:23:27) yugq: ok
(16:26:14) yugq: and another reason is: performance versions is not synchronous with OO developing version.
(16:29:06) yugq: we can not put optimize achievements to new version immediately. so we need performance version to record our achievements.
(16:34:42) mhu: Hi all, sorry for being late
(16:34:42) LiHeng: hello, mhu!
(16:35:08) yugq: hi all
(16:35:22) liangjun: hello :-D
(16:35:41) LiHeng: mhu:xiuzhi can't attend this meeting ,
(16:36:02) mhu: LiHeng: yes, I've seen his email
(16:36:28) LiHeng: mhu: do you have some tipoc for this time?
(16:37:09) mhu: no nothing more than you have in your agenda
(16:37:47) LiHeng: ok , we are beginning :)1.Independent CWS for performance project on OO3.0 and subsequent versions.
(16:38:23) LiHeng: mhu: we want to create Independent CWS for performance project on every version
(16:39:15) mhu: yes, no problem at all. you can create as many cws'ses as you like, for what purpose you like
(16:39:44) mhu: and yes, creating a cws for performance improvements surely makes sens.
(16:40:45) LiHeng: mhu: thank you, but we need combine some effectual change to major version,
(16:41:37) mhu: LiHeng: can you explain the "but" ?
(16:42:52) LiHeng: LiHeng: i can't be sure the time to check out dev-version for the CWS
(16:43:11) LiHeng: mhu: i can't be sure the time to check out dev-version for the CWS
(16:43:44) LiHeng: mhu:you know the problem of performance is dynamic
(16:44:35) LiHeng: mhu:if we checkout early, some change would be void
(16:44:43) mhu: I'm not yet sure, I understand where you see what problem, sorry.
(16:46:11) mhu: ...or, are you saying that other changes might influence your improvements?
(16:46:28) LiHeng: mhu:yes,:) i mean that we need you help for decided the time we checkout the
(16:46:32) mhu: ...if development happens in parallel
(16:46:48) mhu: okay, now I seem to understand
(16:46:57) mhu: ...but...
(16:47:53) mhu: ...I think it is unavoidable that multiple developments happen in parallel. that is the purpose of working with CWS'ses.
(16:49:05) LiHeng: yes ,but performance maybe changed after change for function
(16:49:14) mhu: so, there will surely be other projects targeted at 3.0 that would wish others would not always change so much (the packaging restructuring project, e.g.)
(16:50:36) LiHeng: Is there a period that function not be changed be for release?
(16:50:36) mhu: yes, I understand. But how can parallel changes be avoided? In fact, that's what we want: many developers working simultaneously on the code.
(16:51:26) mhu: yes, there is a period that no changes are accepted: that is after "code freeze", and your changes also would be rejected :-)
(16:53:23) LiHeng: mhu:so , if we have some change can improve more speed for OO, do you think we can have a period , maybe "pre-code freeze"?
(16:53:58) mhu: I still have the idea, that our performance improvement changes can happen in parallel with other changes...
(16:54:29) mhu: ...and I don't think that we need a period where no other change goes in.
(16:55:06) LiHeng: ok, we try to do it so :)
(16:56:38) mhu: if there is an area which you find critcial, we can (a) either monitor that area for other changes and / or (b) talk to the developer(s) that are reponsible for that other area to help us and avoid changes that we think are counter productive.
(16:57:10) LiHeng: thank you!
(16:57:23) mhu: maybe, what I think is, we work with them and not against them?
(16:57:48) LiHeng: ok, we try first ;)
(16:58:23) LiHeng: go next
(16:58:33) mhu: okay. and as I said, if there's something really critical, we'll find a way.
(16:59:11) LiHeng: Put all suggestions and issues in order, make cases for them , and public on wiki
(16:59:41) mhu: yes, sounds like a plan :-)
(17:01:07) mhu: do you already have details?
(17:01:13) LiHeng: mhu:yes , when we changed wiki we find many suggestions and issues for performance, we want to put them in order, and make standart cases
(17:01:37) mhu: ah, yes. that's good I think.
(17:02:15) LiHeng: do you have some idea for this?:)
(17:04:20) yugq: there are issue categories in wiki page before. i think those categories are ok.
(17:04:40) mhu: sorry, i've just looked over the "performance" wiki page. No, I don't have an immediate idea.
(17:04:48) yugq: we may change some if we need.
(17:05:43) mhu: yes, the categories look generally okay. If you find something to add / change, please do so.
(17:05:55) yugq: the most big thing is test case.
(17:06:10) LiHeng: mhu:we can rebuild wiki first ,and I hope you can help us to check them
(17:06:46) mhu: LiHeng: yes, of course. just point me to something, and I'll check it.
(17:06:48) yugq: categories look good, maybe contents mess a little.:P
(17:07:19) LiHeng: yugq:it is just our jobs
(17:07:28) mhu: yes, there's some "mess", mostly remaining from some expermients
(17:07:37) yugq: LiHeng, yes.
(17:07:55) yugq: :)
(17:08:45) LiHeng: kuangliang:are you there?
(17:09:10) kuangliang: I am back
(17:09:32) LiHeng: mhu:can we go to next topic ?
(17:09:39) mhu: yes, please.
(17:09:46) LiHeng: 3.Further discussion of OO compiling and profiling on Solaris and SunStido Perfermance Analysis.
(17:09:56) mhu: yes...
(17:10:15) LiHeng: kuangliang please?
(17:10:28) mhu: I think, we've not had any repsonse to the email on compiler switches?
(17:11:50) mhu: I think, we may try to cross-post to email@example.com, where most experts read email, and the ask them to join the firstname.lastname@example.org list for performance related questions.
(17:12:14) LiHeng: yes, i just send a email to matle, maybe he know something about it
(17:12:17) mhu: s/the ask them/then ask tehm/
(17:12:32) mhu: s/tehm/them/
(17:12:57) LiHeng: mhu: i will send the mail again
(17:13:28) mhu: I was actually thinking of our Hamburg release engineers, which should be more familiar with the optimization compiler switches on all platforms
(17:14:07) mhu: ...and those release engineers do read (and answer) email@example.com
(17:14:35) mhu: ...I think we can also convince them to read firstname.lastname@example.org
(17:15:19) LiHeng: mhu:okay , i try to do it
(17:15:36) LiHeng: and you must help me:)
(17:15:41) mhu: LiHeng: I will also talk to them, and ask.
(17:19:50) LiHeng: ok, for SunStudio Perfermance Analysis, we will document to explanation how to using it for OO, and workout we report
(17:20:30) mhu: okay, sounds good. Have you been able to compile on Solaris now?
(17:21:12) LiHeng: yes , we aleady have a release vesion :)
(17:22:01) LiHeng: and that job we may start on April, after testing OO 2.3.1
(17:22:50) LiHeng: and 3.0
(17:24:14) LiHeng: mhu: another topic temporary
(17:24:29) LiHeng: We have 3Weeks holidays for Spring Festival, Feb 2nd to Feb 17 , so we have to cancel 2 time IRC.
(17:25:19) mhu: LiHeng: sorry, someone just disturbed me at my desk here :-(
(17:25:29) mhu: but now I'm back.
(17:26:03) mhu: I'm fine with your solaris approach (start after 2.3.1)
(17:26:35) LiHeng: and~:)We have 3Weeks holidays for Spring Festival, Feb 2nd to Feb 17 , so we have to cancel 2 time IRC.
(17:26:48) mhu: yes, Peter Junge already told me, that you are all out on Holidays...
(17:27:07) mhu: I wish you all the best, and have a good time.
(17:27:26) LiHeng: thank a lots
(17:27:39) liangjun: :)
(17:27:45) LiHeng: and anybody have some words?
(17:28:02) yugq: no
(17:28:02) kuangliang: no
(17:28:08) mhu: no
(17:28:10) liangjun: no
(17:28:24) LiHeng: mhu:okay, that's all
(17:28:46) mhu: okay, then good bye, see you in 3 weeks.
(17:28:56) LiHeng: see you in 3 weeks :)
(17:29:10) mhu: bye
(17:29:10) LiHeng: bye
(17:29:13) liangjun: bye
(17:29:16) kuangliang: bye
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000
Time: 16:25– 17:33
(16:25:49) LiHen1: hi xiuzhi
(16:25:58) xiuzhi: hi testing....
(16:26:09) LiHen1: it's so difficult to connect
(16:26:18) xiuzhi: hi LiHen1,
(16:26:43) LiHen1: i'm ok
(16:26:47) LiHen1: ;)
(16:26:57) xiuzhi: I am fine
(16:27:09) LiHen1: did you receive the agenda
(16:27:24) xiuzhi: yee
(16:27:33) xiuzhi: it is OK for me
(16:28:13) LiHen1: ok, Mattiaths have a little time for this ,so i reduce the agenda
(16:28:58) xiuzhi: LiHen1:ok
(16:29:20) xiuzhi: mhu:hi
(16:29:29) mhu: Hi LiHeng, xiuzhi
(16:29:43) xiuzhi: LiHen1: where are other members?
(16:29:45) LiHen1: mhu:hi
(16:30:59) LiHen1: network is with some trouble
(16:31:01) mhu: I can wait a few minutes (need to have a quick look into my email first)
(16:31:45) LiHen1: xiuzhi:our gateway have some trouble
(16:31:47) LiHen1: :(
(16:32:08) LiHen1: but, they will connect soon,
(16:32:10) xiuzhi: LiHen1: it is often...
(16:33:26) yugq: Hi
(16:35:15) mhu: okay, back again.
(16:35:39) liangjun: hello:)
(16:35:49) LiHen1: only one last
(16:36:47) LiHen1: mhu: did you receive the agenda
(16:37:01) mhu: LiHen1: yes, looks good to me.
(16:38:07) yugq: mhu: I sent a mail about "Change Performance Project Wiki", did you receive it?
(16:38:47) mhu: yugq: Yes, I have seen and read it.
(16:39:20) yugq: mhu: OK
(16:39:36) mhu: yugq: I think it is a good plan. I'll post a comment to the mailing list.
(16:40:16) yugq: mhu: OK. Thanks!
(16:41:04) LiHen1: we go first, kuangliang can't connect to IRC
(16:41:15) mhu: okay
(16:41:39) LiHen1: 1/Progress Report of OO 2.3.1 Evaluation
(16:42:07) LiHen1: yugq:please talk about you works?
(16:43:34) kuangliang: sorry ,I'm late
(16:44:22) yugq: OK. I mainly test some profile and tuning tools thease days. They are IBM Rational Purifyplus 7.0 & Intel VTune Performance Analyzer.
(16:46:13) mhu: yes
(16:46:15) yugq: Thease tools have good GUI to show the profile data. VTune can also provide a tool to give some optimize suggestion.
(16:47:06) LiHen1: mhu:we will make some analysis reports based on those datas generated by tools,
(16:47:37) yugq: And thease tools seem working fine, easy to use and analyze the data.
(16:47:43) LiHen1: mhu: do you think we will public all analysis on wiki?
(16:49:02) mhu: I think we can publish as much of the analysis as you see fit. It is the best way for others (and ourselfs) to follow the work done.
(16:50:11) yugq: The main problem is that the tool itself will influence the OO startup and other operation and how much influence is hard to know.
(16:50:45) mhu: If some results can only be obtained by using confidential documents (e.g.) then we can say so and publish incomplete data.
(16:51:21) mhu: yugq: I see.
(16:51:45) LiHen1: mhu:thanks,i see, then we will send all reports to you and Matle with e-mail
(16:51:49) yugq: Another problem is thease tools are commercial.
(16:52:13) yugq: They are not free. And I used just trial version.
(16:53:02) mhu: yugq: yes, on windows that is simply so. No idea how to change that. But others also ca use the trial versions.
(16:53:02) xiuzhi: yugh: that would be an issue IMHO
(16:53:43) yugq: mhu: yes.
(16:54:03) mhu: Maybe we can also learn from tools on Linux (valgrind et al) and Solaris (dtrace) which are freely available.
(16:54:39) yugq: mhu: we will also use some tools in Linux and other platforms such as valgrind.
(16:55:14) mhu: yugq: okay.
(16:55:25) yugq: I tried valgrind on Linux, it worked well.
(16:55:42) LiHen1: mhu:yes, we already to study some tools on Linux , so all performance reports will based on Linux version of OO
(16:56:09) kuangliang: I compiled OOo2.3.1 with -O2 optimize compiling on windows and I found this will improve cool startup speed.
(16:56:30) yugq: By the way, the two tools I said before have Linux version either.
(16:56:46) kuangliang: But I don't know ,why OOo don't use this?
(16:56:50) LiHen1: kuangliang:wait a moment
(16:56:58) kuangliang: OK
(16:57:48) LiHen1: yugq:do you have more topics?
(16:57:54) mhu: yugq: yes, I've heard of Intels VTune for Linux, but never used it myself
(16:58:18) LiHen1: mhu:okay, we will try it soon :)
(16:58:38) mhu: LiHen1: okay :-)
(16:58:46) yugq: mhu: yes. It also has GUI like it works on Windows.
(16:59:21) yugq: So I won't talk about detail in the tools now.
(16:59:52) yugq: Another topic is about the Change of Wiki.
(17:00:21) LiHen1: yugq:wiki it's next , kuangliang have some words
(17:00:22) LiHen1: :)
(17:00:29) yugq: OK.
(17:00:33) LiHen1: kuangliang:your turn;)
(17:00:37) yugq: :P
(17:00:42) kuangliang: ok
(17:01:04) kuangliang: I compiled OOo2.3.1 with -O2 optimize compiling and I found this will improve cool startup speed.
(17:01:28) kuangliang: But I don't know, why OOo doesn't use this?
(17:01:51) kuangliang: At most on windows?
(17:01:58) mhu: kuangliang: okay, interesting...
(17:02:28) LiHen1: mhu:do you think will try this method on 3.0?
(17:02:40) mhu: ...it is some time ago, that I looked into windows compiler optimization.
(17:03:52) mhu: ...I think, the current optimization is -Oxs, yes? (optimize as much as possible = x, but favor size = s)
(17:04:45) mhu: ...I think it depends on the actual compiler version, which optimization is best (speed vs size)
(17:04:55) yugq: -Od I think.
(17:05:07) mhu: ... and needs to be verified with each each new compiler version.
(17:05:47) kuangliang: IMHO,it is Ob1 in OOo2.3.1
(17:06:06) mhu: yugq: that should be CFLAGSNOOPT
(17:06:57) kuangliang: CFLAGS +=-Ob1 in OOo2.3.1
(17:07:40) kuangliang: sorry CFLAGS+= -Ob1
(17:07:52) LiHen1: kuangliang:can you put list of your compile-options and result into mailinglist , and that all people can check it
(17:08:23) kuangliang: OK
(17:08:29) mhu: yes, that would be good.
(17:09:49) yugq: mhu: yes. -Ob1
(17:10:04) LiHen1: kuangliang:How about your compiling and testing on Solaris?
(17:10:21) kuangliang: Now I'm compiling OOo2.3.1 on OpenSolaris,I hope to analysis OOo2.3.1 by SunStido Perfermance Analysis
(17:10:47) mhu: kuangliang: oh, that is good.
(17:12:13) kuangliang: But I have some matter,Could Anybody help me?
(17:12:53) mhu: yes?
(17:12:55) LiHen1: mhu:Do Matle know something about it?
(17:13:14) mhu: LiHen1: about what?
(17:14:02) kuangliang: mhu,I finished compiled use dmake debug=true,and setup ,but I can't run it!
(17:14:37) mhu: kuangliang: you have compiled everything with debug=true ?
(17:14:56) LiHen1: maybe yes;)
(17:15:00) kuangliang: No,I run dmake debug=true in src root
(17:15:21) mhu: this will probably need more than 4GB of memory (too much for a 32bit process)
(17:16:06) LiHen1: mhu:thank ,maybe i know how to deal it:)
(17:16:40) kuangliang: I finished compiled it in a long time ,and setup it,:),but OOo2.3.1 can't run:(
(17:17:11) mhu: have you had a look at e.g. the writer library (libsw680...so)? it is probably larger 500MB already, then all the other libs...
(17:18:58) LiHen1: kuangliang:you can run OO under consle, and copy the output information to me, we think i can deal it
(17:19:38) mhu: OOo cannot be completely compiled with debug, it does not fit into memory.
(17:19:38) kuangliang: LiHen1:OK
(17:19:48) LiHen1: kuangliang:also you can compile a release version first
(17:20:36) kuangliang: mhu:Did you mean I can't use debug=true in OpenSolaris?
(17:20:41) LiHen1: mhu:i see, and that we will focus on a part of modules
(17:21:14) mhu: there is something that we call a "non product" version (as opposed to the release or "product" version). it omits the "product=full" argument from the dmake command line.
(17:21:45) mhu: kuangliang: debug=t can be used, but not for all libraries at the same time.
(17:22:20) kuangliang: mhu:thank you
(17:22:58) mhu: ...the "non product" version builds with symbols, but w/o full debug info.
(17:23:25) LiHen1: mhu:thanks,
(17:23:32) mhu: ...please ask Martin Hollmichel how to build such a "non product" version for OOo
(17:23:43) mhu: okay, you're welcome.
(17:23:55) LiHen1: we go to next topic
(17:24:24) LiHen1: mhu:do you think we can remove and rebuild wiki page right?
(17:24:28) yugq: mhu: we need your suggestion about the Change of Wiki.
(17:25:46) mhu: if anything looks somehow important, just move it somewhere safe. Otherwise feel free to change what you want. I think those pages are not maintained.
(17:26:29) mhu: ...some of the info on those pages might have historic value (documenting what has been done in the past).
(17:27:21) mhu: yugq: I think your email already contained a good plan how to handle those pages.
(17:27:47) yugq: mhu: ;-)
(17:28:17) yugq: mhu: Thanks for the suggestion.
(17:28:49) LiHen1: okay ,we will discuss all CAT in mailinglist, ....
(17:28:58) mhu: yugq: no problem, you are doing most of the work (I'm only suggesting)
(17:29:38) yugq: aha
(17:30:03) yugq: LiHeng: I finished my topic.
(17:30:09) LiHen1: mhu:if you have some suggestions ,please reply us whatever you mind
(17:30:25) mhu: LiHen1: yes, will do.
(17:30:51) LiHen1: okay, that all for me
(17:31:07) mhu: okay, then we are finished for today?
(17:31:41) LiHen1: yes, ;) , it's on time today
(17:31:59) mhu: okay, then see you on the mailing list, or next Thursday on IRC.
(17:32:08) LiHen1: bye
(17:32:14) kuangliang: bye
(17:32:14) mhu: bye all
(17:32:16) yugq: bye everyone.
(17:32:34) liangjun: :) bye