Difference between revisions of "Performance/Meetings/2009 08"

Latest revision as of 23:11, 17 March 2010

Performance/Meetings/2009/08/28

Meeting Minutes
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000
Performance Project

Date: 2009/08/28
Time: 15:58– 16:50
Meeting No.:

Agenda:

(4:02:25 PM) liheng: Hi,all
(4:02:50 PM) liheng: Good morning/afternoon
(4:03:09 PM) yugq: hi all
(4:03:16 PM) yujinliang: hi
(4:03:44 PM) cd_oo: hi all :-)
(4:04:07 PM) liangjun: Hi all :P
(4:05:32 PM) erAck: good morning / afternoon
(4:08:34 PM) liheng: Hi, all, Let's update our status.
(4:08:47 PM) Matthias: Hi all, sorry for being late
(4:09:10 PM) liheng: Hi, mhu, we are just beginning to update our status.
(4:09:39 PM) Matthias: liheng: hi, thanks
(4:13:03 PM) liheng: I have just arranged all UOF works and prepared to put them into OOo, but they are not for performance.
(4:14:00 PM) sb: I haven't got anything new to share relative to last week...
(4:14:11 PM) liangjun: my status: now I had finished the SD styles's chage . next I will continue the sd shape and object.
(4:15:35 PM) liangjun: improve ODF performance .
(4:16:10 PM) ***Matthias is finalizing the buffered file I/O cws (mhu20); resync'ed to dev300m54, built 4 platforms; needs final developer test (smoketest); then ready for QA
(4:17:42 PM) cd_oo: My status: One update regarding current start up performance. Skotti made first test with DEV300m55 under Windows. Due to unknown reasons he could only see 10-20% performance increase where we expected ~30% for cold start up (CWS sb110 is missing will be part of DEV300m57). We are currently looking for the root cause of this unexpected effect. I don't know the state of the Linux improvements...
(4:17:44 PM) cd_oo: ...on the master builds.
(4:18:40 PM) sb: cd_oo: nor do I
(4:20:57 PM) skotti: The numbers are based on one single measurement currently. So we can have wrong values here. Currently i'm rerunning the cold start tests to get statiscally relevant data.
(4:21:09 PM) sb: cd_oo: I mean I do not know whether start up performance degraded again on Linux for the last few builds.
(4:21:46 PM) liheng: I will restart the work that put checkpoints into some CWS and trace the changes of performance in Benchmark System.
(4:25:18 PM) erAck: nothing special in Calc, CWS calcsheetdata is awaiting QA, AutoTests ran.
(4:25:35 PM) liheng: skotti:Do you need we do cold start tests here and send result to you?
(4:26:11 PM) skotti: liheng: Thank you for your offer, but tests are running, i'll have the results later today.
(4:27:10 PM) skotti: erAck: QA status for CWS calcsheetdata: Automation done, OC verifies the issue, probably today.
(4:28:58 PM) liheng: skotti:You are welcome, if you expect more results and more people to run test, you can give a mail,i can do them in our company.
(4:30:52 PM) skotti: liheng: Thank you. Whem m57 is there it could be nice to have independent results, i'll get back to you then
(4:32:40 PM) liheng: skotti:You are welcome, anytime :)
(4:47:27 PM) liheng: I think we can close this meeting earlier today, anyone have more topic?
(4:48:12 PM) Matthias: I am fine with finishing early; maybe I can finalize my cws today, then :-)
(4:48:21 PM) erAck: fine with me
(4:49:11 PM) liangjun: and me
(4:49:20 PM) liheng: Okay,bye all, nice weekend!
(4:49:34 PM) yujinliang: bye
(4:49:36 PM) odf-mib: Bye! Have a nice weekend!
(4:49:42 PM) erAck: bye
(4:49:42 PM) Matthias: yes, bye all; have a nice weekend
(4:50:22 PM) liangjun: bye

Performance/Meetings/2009/08/21

Meeting Minutes
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000
Performance Project

Date: 2009/08/21
Time: 15:58– 17:13
Meeting No.:

Agenda:

(4:02:22 PM) liheng: Hi,all
(4:02:41 PM) erAck: moin
(4:02:52 PM) cd_oo: hi all :-)
(4:02:59 PM) liheng: Good morning / afternoon
(4:03:12 PM) yugq: hi all:)
(4:03:23 PM) yujinliang: hi
(4:03:30 PM) liangjun: :) hi all
(4:05:03 PM) odf-mib: Hi!
(4:05:37 PM) liheng: Hi odf-mib:)
(4:07:01 PM) Matthias: Hi all, sorry for being late ...
(4:07:50 PM) liheng: Just in time , Let's update our status :)
(4:08:08 PM) Matthias: so, you have not yet started ?
(4:09:47 PM) ***Matthias is currently testing / verifying the buffered file I/O implementation for windows platform; a (bootstrap) build from scratch already uses so much of osl/file functionality that once a build plus smoketest succeedes, almost all should be fine :-)
(4:10:39 PM) liheng: No, we are just beginning
(4:10:42 PM) cd_oo: My status: Nothing new from my side. Library restructuring is on hold
(4:11:08 PM) cd_oo: cd_oo: Currently I am working on docking windows for extensions
(4:15:28 PM) sb: My status is that first start-up measurements for my new configmgr look promising (see <http://wiki.services.openoffice.org/wiki/Performance/Configuration#Work_in_Progress>) and I am currently working on a hand-crafted XML reader to improve those numbers.
(4:18:05 PM) liheng: sb:What is the hand-crafted XML reader? please explain to us :)
(4:22:04 PM) sb: liheng: It turned out my current implementation burned too many processing cycles in libxml2's xmlreader (which I am currently using to read the configuration data XML files), and I think I can drastically reduce the number of cycles burned by writing my own code to read the XML files, tailored especially to the specific needs of those .xcs/.xcu files. (I know that you are in a state of sin...
(4:22:05 PM) sb: ...if you roll your own XML parser, which will most certainly not be 100% compliant, but I think timing constraints warrant this here. Plus, the old configmgr also had its own XML reader...)
(4:25:22 PM) liheng: sb:Thank you, I see, you make special reader for .xcs/.xcu only check specific items, right?
(4:28:28 PM) sb: liheng: it is more the passing of data between XML reader and the code that consumes it, and things like e.g. not making sure in the reader that what I parse as an XML tag name is indeed well-formed, as the consuming code will detect the error, anyway. Thinks like that. (4:31:36 PM) liheng: sb: Thank you, exactly clear!
(4:36:52 PM) liangjun: sb: if we replace the xml reader to binary reader ? Our new configmgr is binary reader :)
(4:50:48 PM) ***Matthias wonders whether the discussion is already finished; if yes, I would propose to close the meeting for today ...
(4:51:13 PM) sb: liangjun: Yes, but I would like to avoid that (I still think reading/writing XML can be fast enough, and XML data has advantages over binary data).
(4:57:48 PM) liheng: sb: In our works to merge all configure file into only file is major target.
(5:00:24 PM) liheng: Time up.
(5:00:43 PM) ***erAck gotta go
(5:00:46 PM) erAck: bye
(5:01:10 PM) cd_oo: bye and have a nice week
(5:01:15 PM) Matthias: bye all, have a nice weekend
(5:01:25 PM) liheng: Bye all, have a nice weekend.
(5:01:34 PM) yujinliang: bye all
(5:01:49 PM) liangjun: Bye all,
(5:05:20 PM) sb: liheng: Merging files is also important to me (I just merge XML files instead of binary files).
(5:05:42 PM) sb: ...sorry for responding late, I had been distracted...
(5:13:39 PM) sb: ...and bye

Performance/Meetings/2009/08/07

Meeting Minutes
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000
Performance Project

Date: 2009/08/07
Time: 15:49– 16:38
Meeting No.:

Agenda:

(3:58:44 PM) erAck: good morning / afternoon
(4:00:04 PM) cd_oo: hi all :-)
(4:00:17 PM) erAck: weather at Hamburg: a nice sunny day, blue sky, not one cloud, already
(4:01:12 PM) liheng: Hi,all Good morning / afternoon
(4:02:42 PM) liheng: Yuguoqiang & Matthias Huetsch can't attends this meeting.
(4:03:18 PM) liheng: Let's start to update our status :)
(4:03:53 PM) odf-mib: Hello!
(4:04:00 PM) liangjunzeng: hi all :-)
(4:05:42 PM) liheng: Hi,odf-mib
(4:06:02 PM) liheng: We just start to update our status :)
(4:06:46 PM) arwe: Hi all :-)
(4:07:43 PM) cd_oo: my status: Nothing new from my side as I am currently helping to finalize an important OOo 3.2 feature.
(4:08:18 PM) liheng: My start: Finished the auto-test module based test-tools, and put helper code into m44 and OOo3.2.
(4:09:21 PM) sb: My status is that I am still busy with the new configmgr, doing some first improvements (of the kind like replacing STLport std::hash_map with faster std::map, or switching from DOM-based XML processing to SAX-based).
(4:09:40 PM) sb: s/improvements/performance improvements/
(4:12:35 PM) liheng: Please you all give me the name of your CWS about performance, I hope add some check-point in OOoBenchmarkSystem for our current works
(4:16:12 PM) erAck: Niklas continues work on saving only modified sheets. CWS is 'calcsheetdata'
(4:20:20 PM) liheng: erAck:Thank you:)
(4:22:18 PM) erAck: liheng: np :)
(4:22:49 PM) sb: liheng: CWS sb110 "Windows: move all DLLs into brand program directory", to be integrated into DEV300 soon, will improve performance on Windows (only)
(4:35:53 PM) liheng: Okay, I think we have no more topic, can we close the meeting ahead of time.
(4:36:30 PM) erAck: I'm fine with it.
(4:36:49 PM) arwe: Okay
(4:36:56 PM) odf-mib: Okay. Thank you. bye
(4:36:59 PM) sb: bye
(4:37:06 PM) liheng: Bye all:)
(4:37:08 PM) erAck: bye
(4:37:16 PM) liheng: Have nice weekend.

Performance/Meetings/2009/08/04

Meeting Minutes
IRC Meeting of Sun Microsystems (StarOffice) with RedFlag2000
Performance Project

Date: 2009/08/04
Time: 15:28– 17:12
Meeting No.:

Agenda:

(16时03分05秒) erAck: good morning / afternoon
(16时03分07秒) mhu: Hi all
(16时03分16秒) yugq: hi all
(16时03分28秒) Malte: Hi - back from vacation :)
(16时04分13秒) yugq: Malte, me too:)
(16时04分14秒) liangjun: Hi all:P
(16时04分46秒) yugq: All: please update your status.
(16时06分38秒) mhu: yugq: how about liheng, will he not attend today ?
(16时07分39秒) ***mhu is finalizing the buffered file I/O windows implementation, plus resync cws to newer milestone, etc ...
(16时07分49秒) sb: I haven't got much to share, as this week I was mainly busy with other tasks than the configmgr rewrite.
(16时07分56秒) yugq: mhu: He will be right here soon:)
(16时08分07秒) mhu: yugq: okay, thanks :-)
(16时09分24秒) liheng: Hi,Sorry I'm late:(
(16时09分54秒) liheng: Hi, all good morning / after
(16时10分06秒) mhu: liheng: we've just started, yugq already asked us for a status update
(16时10分33秒) mhu: liheng: ...and yes, good afternoon, of course ... :-)
(16时12分31秒) liheng: Matthias:My network has some troubles.
(16时12分33秒) yugq: CWS configtune02 for configmgr refactor was created and it is a workable edition now. :-D
(16时14分20秒) liheng: Let's update our status :)
(16时14分27秒) erAck: Niklas continued work on saving changed sheets only, cws calcsheetdata.
(16时15分43秒) ***mhu and sb already reported their status, yugq should have it in his logfile
(16时16分38秒) liangjun: my status: now I know the OOo style. And continue my odf loading performance. wish I 'm right.:P
(16时16分43秒) yugq: ***mhu is finalizing the buffered file I/O windows implementation, plus resync cws to newer milestone, etc ...
(16时16分43秒) yugq: sb: I haven't got much to share, as this week I was mainly busy with other tasks than the configmgr rewrite.
(16时17分18秒) mhu: yugq: thanks; can you also remind me of what you did in cws configtune02, please ?
(16时17分45秒) sb: yugq: "CWS configtune02 [...] was created" -- why based on very old DEV300m33?
(16时18分14秒) yugq: mhu: yes...
(16时19分34秒) yugq: mhu: before configtune02, after installing OO, we have to do some mannual work to make OO work. Now, it doesn't need.
(16时19分54秒) mhu: yugq: I'm asking to prevent / correct (try to...) too much collision between your and sb's work on configmgr, if at all possible
(16时20分37秒) mhu: ah, yes; I remember, that was what discussed one of the last times.
(16时21分23秒) yugq: mhu: sorry. I just have repeat two lines of the work status of you and sb. Not my status.
(16时21分52秒) mhu: yugq: yes, I understand
(16时22分39秒) liangjun: sb: now we work on m33. ant test it is runing good. (16时23分03秒) mhu: ...that your changes could not be tested without manual work after installation.
(16时23分03秒) liangjun: ant / and
(16时23分39秒) yugq: mhu: but unfornately, the CWS configtune02 is still based on a lower edition which is DEV300_m33.
(16时24分16秒) mhu: yes, understood, but sb asked why still that old milestone ?
(16时25分11秒) sb: liangjun/yugq: OK, be that basement on DEV300m33 as it is. I will see if I can build your CWS and have a look at it. (I still hope that, in the end, we can get away with sticking to the good old xml data formats, but verifying that with my own CWS will still take some time.)
(16时25分11秒) mhu: (...which is approximately 6 to 9 months old, right ? )
(16时25分47秒) liangjun: mhu: so we n't update it to new milestone.
(16时25分58秒) yugq: this is because our developers are working on that old milestone for time. And they didn't used to the community develop way. And I think they will be switch it to latest milestone soon.
(16时26分23秒) liangjun: yugq: yes
(16时26分29秒) mhu: yugq: okay, understood; thanks for the explanation.
(16时27分03秒) liangjun: sb: :P
(16时27分42秒) lihen1: My status:Update first page of wiki and fill up IRC Logs
(16时29分41秒) yugq: mhu: we rewrite the whole configmgr module. so it takes long time now.
(16时30分42秒) mhu: yugq: that's why I asked how your and sb's work fit together, I hope the collision is not too hard
(16时31分31秒) yugq: mhu: yes. we hope so:)
(16时31分33秒) mhu: ...and yes, I remember when we first discussed your rewrite.
(16时32分31秒) mhu: so, probably you should keep close contact to sb
(16时33分23秒) liangjun: mhu: I think so. because it is base on OOo's source.
(16时33分53秒) yugq: mhu, sb: yes. We can contact with email. And we also have many problems to ask your help:)
(16时34分04秒) lihen1: liangjun: I think you change lots of objects and config
(16时35分02秒) sb: yugq,mhu: To re-state what I already wrote some time ago: I have various motivations to do the configmgr rewrite, only one of them being performance (I never saw working code from anywhere else, so had to start this myself). I hope we can integrate ideas -- maybe even some code -- from yugq's work, but we'll have to see.
(16时35分24秒) liangjun: lihen1: yes , but I dn't change the base data and config UNO
(16时35分26秒) mhu: liangjun: yes, of course; OOo's configmgr source; and two "groups" now working on the same code, you and sb
(16时36分04秒) mhu: sb: yes, I know. That is also what I am trying to explain to yugq and liangjun.
(16时36分29秒) liangjun: mhu: :)
(16时36分38秒) mhu: :-)
(16时37分18秒) yugq: mhu, sb: yes. liangjun wrote most of the code, and we can share some, also ideas.
(16时38分46秒) lihen1: yugq,liangjun:We can also share code of config tools :) (16时39分22秒) yugq: liheng: of course;-)
(16时39分42秒) mhu: okay, so we now all understand that there are two sets of changes to configmgr, not based on each other; but sb will have a look how to best approach the integration of ideas / code.
(16时40分10秒) ericb2: hello
(16时40分18秒) mhu: [that should have been a question, ending with ?]
(16时40分28秒) mhu: Hi Eric
(16时41分09秒) liangjun: lihen1:yugq:mhu:sb: yes , I think our current design and source code has any problems. wish it will be better :P
(16时41分35秒) sb: mhu: Yes, I'm fine with that approach.
(16时42分19秒) mhu: okay, thanks
(16时42分24秒) yugq: mhu, sb: I think we can exchange some docs or emails for the main idea, design or other, and that will have better understanding.
(16时42分31秒) liangjun: mhu: yes we can get more helper :)
(16时43分38秒) mhu: liangjun: best advise: keep in touch with sb :-)
(16时46分36秒) mhu: ericb2: you wanted to comment ? please don't hesitate to do so
(16时46分57秒) liangjun: mhu: I know, I 'm waiting today :)
(16时47分30秒) ericb2: mhu: thanks. My case is special : I'm working with students, who discover OOo source code, and profiling
(16时47分55秒) ericb2: but we can do some basics tasks, like help to adapt cws to more recent milestones, and some other basic tasks
(16时48分13秒) ericb2: sb: I'm just discovering configmgr, but for another reason :)
(16时48分28秒) sb: ericb2: for what reason?
(16时49分03秒) ericb2: sb: I implement a menu entry, and I'm trying to understand the complete travel of the information, starting from the mouse click .. vcl .. desktop and so on :)
(16时49分29秒) ericb2: sb: and I'm currently learning what does configmgr. The basics, of course
(16时51分47秒) sb: ericb2: But IIUC, menu data is not stored in the configmgr-controlled registry, but rather in its own XML files. Anyway, the configmgr's UNO interface is IMO a bad thing, as it re-uses too generic interfaces and tries to support every conceivable combination of those generic interfaces that only looked remotely usable (like supporting all of XPropertySet, XMultiPropertySet, etc.). That makes...
(16时51分49秒) sb: ...comprehension much harder than necessary. Good luck. ;)
(16时53分08秒) ericb2: sb: thanks :)
(16时53分27秒) ericb2: erAck: one user reported 2 problematic (enormous) files
(16时53分31秒) ericb2: erAck: http://www.mydisk.se/coco/?user=cubytus
(16时54分20秒) ericb2: erAck: do you think it is possible to trace, or study what happens, to improve things ?
(16时54分42秒) erAck: ericb2: please, this is a meeting, use #dev...
(16时55分06秒) ericb2: erAck: my question was about performance
(16时55分13秒) ericb2: erAck: but sorry if I'm off topic
(16时58分25秒) erAck: ericb2: the usual approach is to run the office in valgrind with callgrind, wait a few hours and then analyze the output in kcachegrind.
(16时58分47秒) ericb2: erAck: ok, we'll do. Thanks :)
(16时59分15秒) erAck: ericb2: see http://wiki.services.openoffice.org/wiki/Callgrind
(16时59分38秒) mhu: ericb2: yes, I guess there is no better advise than what erAck gave you.
(17时01分09秒) mhu: ericb2: also, the documents that you mentioned, do you think they could serve as "standard" benchmarks for doc. load performance ?
(17时01分30秒) ericb2: mhu: I don't know yet, but this could be
(17时02分02秒) ericb2: mhu: they are really big. Maybe something is wonrg in them, I don't know yet
(17时03分14秒) mhu: so, maybe you contact erAck and Niklas offline to discuss these docs, Niklas also takes care of calc doc load performance (I think)
(17时03分31秒) ericb2: mhu: I'm sorry if my questions look dumb, but I think if we can document more how to proceed, and provide "standard' document (very good idea), more people will join and help
(17时03分54秒) ericb2: mhu: yes, we'll do
(17时04分27秒) mhu: yes, of course; it is just that even in this context here, Niklas would be the one to work with us on calc doc load performance improvements
(17时05分34秒) mhu: ericb2: and, no I don't think your questions are dumb :-) But please accept the answers :-)
(17时05分43秒) ericb2: mhu: I agree, but my question was not only focused on Calc in fact. Was just the example I had in mind
(17时06分08秒) ericb2: mhu: no problem. The first step is always to most difficult :)
(17时06分26秒) mhu: yes :-) lets do more steps then.
(17时07分51秒) ***mhu looks at his calendar, sees more meetings scheduled for today, and would like to finish this one ...
(17时08分38秒) erAck: yep, bye
(17时08分45秒) lihen1: Time up :)
(17时09分07秒) yugq: bye all:)
(17时09分21秒) liangjun: bye all:P
(17时09分30秒) ericb2: mhu: more precisely, and since I'm working with student, I really need to define a methodology. Use tools is imho not sufficient for me, but I'll search by myself
(17时10分08秒) mhu: ericb2: methodology for doing what ?
(17时10分20秒) ericb2: mhu: learning how to profile e.g.
(17时10分46秒) ericb2: mhu: but I'll think on my side, and I'll stop boring you with such questions
(17时11分31秒) mhu: aha, but again, as erAck said you may find more responses on #dev than here, as this is just a weekly meeting focussed on specific issues
(17时11分43秒) ericb2: mhu: ok
(17时11分48秒) mhu: and, no it is not boring :-)
(17时12分05秒) mhu: bye all have a nice weekend

Go back