Difference between revisions of "Performance/Meetings/2009 05"

Latest revision as of 02:04, 18 March 2010

Performance/Meetings/2009/05/22

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

Date: 2009/05/22
Time: 15:58– 17:02
Meeting No.:

Agenda:

(4:01:40 PM) odf-mib: Hi.
(4:01:50 PM) xiuzhi: hi all
(4:02:17 PM) odf-mib: Seems we are only a few project members today. Yesterday was a public holiday in Germany, so many from the Hamburg team are in vacation today.
(4:02:18 PM) liheng: Hi,all
(4:03:58 PM) liangjun: Hi,all
(4:04:06 PM) xiuzhi: odf-mib: really?I like vacation. :) next Friday is public holiday in China
(4:04:19 PM) liheng: odf-mib: Thanks, and next week is holiday in china ,the Dragon Boat Festival
(4:04:33 PM) Matthias: Hi all, sorry for being late
(4:04:50 PM) xiuzhi: mhu: hi. not too late
(4:05:05 PM) Matthias: xiuzhi: moin :-)
(4:05:18 PM) odf-mib: mhu: Most people from Hamburg seem to have vacation
(4:05:25 PM) liheng: Hi, Matthias:)
(4:05:27 PM) Matthias: so, I have not missed much, so far ?
(4:05:39 PM) Matthias: Hi Li Heng
(4:05:44 PM) odf-mib: mhu: We haven't started yet.
(4:05:58 PM) liheng: Matthias: No we are talking about holidays
(4:06:07 PM) liheng: next week is holiday in china ,the Dragon Boat Festival
(4:06:10 PM) Matthias: aha, that is a good topic
(4:06:49 PM) liheng: Let's update status of our works.
(4:06:59 PM) Matthias: ...and as we are at it, I will be away (on vacation) for the next 4 weeks (starting next wednesday)
(4:09:05 PM) liheng: 4 weeks, so long, and wish you have a very nice vacation!
(4:09:08 PM) Matthias: status: mostly debugging (with MAV) my osl/unx/file buffered I/O implementation (found hopefully last issue when building OOo from scratch)
(4:09:57 PM) Matthias: it is actually 3 weeks of journey plus a couple days before and after, thanks for the wishes
(4:10:58 PM) Matthias: I will be back for our meeting Friday June 26
(4:11:52 PM) mba: As Carsten, Oliver and Bj rn are on vacation today, let me give a short report of their work in the last week.
(4:12:29 PM) mba: Carsten is working together with Ingo Schmidt to make our setup on Windows able to support both rebased and non-rebased libraries
(4:13:01 PM) liheng: My Status:I collected the test cases and documents and make them systematical for Benchmark-System and deployed the OOo3.0 and OOo3.1 on the server, but I can't run them automatically in Java serivce ...
(4:13:36 PM) liangjun: status: refactor configmgr :)
(4:13:49 PM) mba: Oliver has done a first prototype for asynchronous saving in the background (working on a copy of the document created in memory). It worked surprisingly well, but it's not done yet, just a prototype as it does not care for threading problems. We will evaluate it further.
(4:14:23 PM) mba: Bj枚rn Michaelsen is still trying to squeeze out some CPU cycles in the SfxItemSet implementation
(4:14:59 PM) sb: ...and I am still busy with the new configmgr (and will be, for the foreseeable future...)
(4:16:24 PM) liheng: next week, I will do all benchmark test manually on Benchmark-System and get comparable raw data for showing the functions on Benchmark-System.
(4:17:15 PM) mba: I'm working on another performance optimization, not related to speed but to energy saving: I'm trying to get rid of the background processing in SFX to get the status of commands that can't be or are not invalidated correctly and so have to be polled with a 200ms timer (so called "volatile" slots/command). It's nearly finished now and waits for a missing implementation in the system layer...
(4:17:17 PM) mba: ...(notification for clipboard content changes).
(4:17:58 PM) zhangyuwei: my status: I'm fixing bugs about asynchronous load, and now take much less time all loading process.
(4:18:28 PM) Matthias: mba: system layer means VCL here, right ?
(4:19:15 PM) mba: That's up to Philipp to decide. Currently the clipboard notification is implemented on Windows only and it uses a UNO service. The implementation IIRC is outside of VCL
(4:20:15 PM) ***Matthias is always confused when someone uses "system layer" and is not talking about "sal/osl/rtl"
(4:22:17 PM) Matthias: sorry for interrupting, please go ahead ...
(4:25:24 PM) tora-jap`: javaldx - Can we prevent javaldx at the initial startup right after installation of OOo is finised?
(4:25:24 PM) tora-jap`: A user installs OOo into his/her PC.
(4:25:24 PM) tora-jap`: The user starts OOo.
(4:25:24 PM) tora-jap`: OOo first starts javaldx and it silently looks for JRE here and there.
(4:25:27 PM) tora-jap`: A user notices that a harddisk is making noises, but no information appears on the display.
(4:25:30 PM) tora-jap`: Eventutally, maybe several seconds later, the Welcome dialog appears to serve the user.
(4:27:36 PM) sb: tora: If OOo shall be able to instantiate JVMs in process, javaldx is needed (because LD_LIBRARY_PATH needs to be set, which is only read during exec and not again later on).
(4:28:20 PM) tora-jap`: Right. But is it needed to show Welcome dialog?
(4:29:23 PM) sb: The maintainer of javaldx is Joachim Lingner (not on this chat). Maybe we can discuss potential improvements on dev@udk.ooo or just dev@ooo mailing lists, so that Joachim can join in.
(4:29:45 PM) tora-jap`: I see.
(4:30:26 PM) tora-jap`: issue 88162 "Removing blank lines from a .xcu file to reduce its size"
(4:30:27 PM) tora-jap`: Does anyone have insterests on this issue?
(4:30:27 PM) tora-jap`: It is about configuration files.
(4:30:28 PM) sb: "is it needed to show Welcome dialog?": If Welcome dialog is shown from soffice.bin: yes, indirectly.
(4:31:38 PM) sb: tora: .xcu file size: I never found time to look at it (and improve my rudimentary xslt knowledge). If you want to do something, go ahead!
(4:34:05 PM) liheng: tora-jap:A people in RedFlag is just making some tools for rebuild all configure files including .xcu
(4:34:56 PM) tora-jap`: For Welcome dialog, regardless of with_JRE or whitout_JRE, the welcome dialog window appears. So I thought JRE is not required to show welcome dialog window. Anyway, its further discussion will be on the ml 'sb' suggests.
(4:35:55 PM) tora-jap`: liheng: Sounds nice!
(4:38:04 PM) sb: tora, re .xcu size: Also, depending on how I will move forward with my new configmgr (which I do not yet know), the .xcu files in the shared layer might or might not be replaced with some other representation, so that improvements on the .xcu file size might be mood in the long run. Just to let you know...
(4:38:06 PM) Matthias: liheng: do we somehow coordinate who does what on configmgr / xcu files / ...
(4:38:09 PM) liheng: sb:For UX performance, to show Welcome dialog maybe really not with JRE
(4:38:50 PM) sb: liheng: Sure, that would mean moving Welcome dialog from big soffice.bin to some earlier process.
(4:39:42 PM) sb: mhu: I never heard back from any other groups trying to improve configmgr, so decided to start work on it myself.
(4:40:38 PM) liheng: sb:liangjun has been working on it for one year.
(4:40:40 PM) Matthias: oh yes, I know; it is just that liheng , his team and I discussed somethings before this team got so large
(4:41:35 PM) Matthias: ...and of course, everyone should coordinate with sb (he maintains configmgr when all else fails)
(4:42:07 PM) xiuzhi left the room (quit: Read error: 60 (Operation timed out)).
(4:43:00 PM) liangjun: liheng: sb: http://wiki.services.openoffice.org/wiki/Configmgr_Refactoring
(4:43:06 PM) tora-jap`: sb: For the issue, it is not about the amount of information. It is about how to remove does-not-match-localized-text. Original source and localized.sdf will be combined at compilation and be applied the XSLT. Unwanted tranlations are currently left as a blank line. The unwated one could be simply disregard.
(4:43:18 PM) liangjun: http://wiki.services.openoffice.org/wiki/Configmgr_Refactoring/Design
(4:44:36 PM) sb: re configmgr: liangjun, I know that wiki page, but did not see any progress (and no working code for a long time). I will put my own ideas into the wiki soon, so we can see how to continue.
(4:46:34 PM) sb: tora: not "amount of information," but "amount of redundant whitespace," right?
(4:46:53 PM) liheng: sb:Thanks, and we are still working no it,but we start that work based on a earlier RedOffice version and we must transform some work to openoffice.org 3.x
(4:47:24 PM) liheng: Sorry we are still working on it :)
(4:47:36 PM) tora-jap`: sb: Exactly! :-)
(4:49:12 PM) sb: liheng: I see. It is hard for me to think about improvements in the configmgr area without actually trying them out, and it is extremely hard for me to try out anything with the existing configmgr code base. Hence, I got dragged into redoing the code base alongside producing ideas for improvement.
(4:49:28 PM) liangjun: sb: :), yes We are send source code on svn://svn@svn.services.openoffice.org/ooo/cws/configtune00 . but it's not final.
(4:51:32 PM) sb: liangjun: OK, will look at it.
(4:52:01 PM) liheng: sb:We will discuss all ideas about configmgr again, via email :)
(4:52:48 PM) sb: liheng: Stick to the dev@performance.ooo mailing list, please.
(4:53:10 PM) tora-jap`: sb: With current implementation at the level of file I/O, every such as get-the-next-line() triggers a read() system call. It is too costly. So removing redundant whitespace will reduce the number of system calls.
(4:53:10 PM) tora-jap`: Meanwhile, for redundent system calls around get-the-next-line(), 'mhu' has been working on.
(4:53:10 PM) liangjun: sb::) Thanks.
(4:54:09 PM) liheng: tora-jap:If you have some suggestion for tools of config, please send your requirments to dev@perfomance.openoffice.org
(4:54:48 PM) Matthias: indeed, for line based IO the new impl of osl_readLine (then buffered) should help reduce IO system calls drastically
(4:54:56 PM) liheng: sb: Yes, of course
(4:59:35 PM) ***Matthias will probably not finish 'buffered I/O' in the next few days, and will be back only Fri, June 26; 'buffered I/O' will need to wait until then
(4:59:42 PM) tora-jap`: liheng: Well, I am trying for it. Requirements for configuration manager is not so simple. Even a user changes a location of a window, information of the window will be stored to the configurtion data. Data of configuraion might be kept in LDAP, file server over the network, local disk, local removal device, ...
(5:02:01 PM) liheng: Okay,time up,have nice weekend
(5:02:17 PM) tora-jap`: Cheers!
(5:02:31 PM) Matthias: okay, bye; have a nice weekend everyone
(5:02:43 PM) liheng: Bye,all !
(5:02:49 PM) odf-mib: Have a nice weekend! Bye!

Go back

Performance/Meetings/2009/05/15

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

Date: 2009/05/15
Time: 16:00– 17:26
Meeting No.:

Agenda:

(4:00:25 PM) arwe: Hi everyone ☺
(4:00:41 PM) yugq: hi all
(4:01:07 PM) Matthias: hi
(4:02:07 PM) cd_oo: Hi all
(4:02:27 PM) liheng: Hi,all!
(4:04:01 PM) liheng: Let's update our status! :)
(4:06:16 PM) sb: my status: working on configmgr redesign/reimplementation, mainly to improve performance (by touching way fewer files during startup), but also to finally address some other shortcomings
(4:06:40 PM) Malte: Sounds good! :)
(4:08:39 PM) ***Matthias continues with buffered file I/O; unix implementation ready for developer review (currently MAV is looking into it); starting with windows implementation.
(4:09:11 PM) liheng: My status:I just finished all works to transform Benchmark-System to Glassfish/MySQL and clear commercial code, but I can't complete manual and commit all helper code into my CWS.
(4:09:11 PM) os_ooo: Hi,
(4:09:12 PM) os_ooo: the cws os128 that contains the fixes related to SfxItemPropertySets, frame access in Writer and some others in Writer is on it's way to QA now.
(4:09:14 PM) os_ooo: Next areas are asynchronous storing in AutoSave and changes in the implementation of SfxItemSet where Bjoern Michaelsen is currently working on.
(4:10:03 PM) odf-mib: I have updated the rtllog2odf tool
(4:10:19 PM) Matthias: liheng: you cant commit into your CWS?
(4:11:19 PM) liheng: Matthias: :), I need more time to do that.
(4:11:30 PM) Matthias: liheng: oh, I see :-)
(4:13:28 PM) liheng: I changed the record format in help code,at last week, because I find my new analyzer need more information. ;)
(4:13:30 PM) arwe: I have now added the DrawingLayer ModelLocking performance improvement to CWS aw065; no more expensive BoundRect calculations when there is not yet a view.
(4:14:07 PM) arwe: Also looking with IHA for some superfluous chart actions with Drawingayer
(4:15:04 PM) arwe: Also tinking about a 'Meta-Object' for chart, e.g. 5000 data points currently create 5000 SdrObjects, change to one MetaObject with 5000 transformations...
(4:15:52 PM) liheng: Matthias:If you done your works,which improvments will be found in general Hotspots (like, startup, load ,save) ?
(4:17:03 PM) Matthias: liheng: I expect improvement for all: startup (mainly cold), load and save
(4:17:31 PM) Matthias: ...but of course, when there is a build to test, we will see more (4:17:57 PM) Matthias: ...details
(4:19:08 PM) cd_oo: My status: I am currently working with the setup team on non-rebased libraries and how to get it into installation sets without breaking anything. Especially to not break Windows patching is difficult as we can have two possible installations (rebased/non-rebased) but just one patch.
(4:20:43 PM) liheng: Ok,as I see, I think I can use your CWS to test our Benchmark-System, because it may improve all number of Hotspot and user experience index
(4:22:31 PM) Matthias: liheng: okay, no problem; just give me some time to commit all changes to the repository (it will be in cws mhu20).
(4:24:01 PM) liheng: If that CWS is ready for QA please tell me, I will checkout and create a report
(4:25:42 PM) Matthias: yes, sure
(4:27:27 PM) liheng: It seems that we have no more topic today, break up?
(4:29:06 PM) liheng: Hi,enoki_
(4:30:36 PM) Matthias: liheng: I think, the status round table has finished; and I dont have more for today; maybe others ?
(4:31:45 PM) liheng: Yes,I think so.:)
(4:33:18 PM) Matthias: any comments from our japanese / goodday attendees ?
(4:35:22 PM) makoto_goodday: no thing thanks
(4:35:46 PM) Matthias: you're welcome, greetings to tora
(4:37:58 PM) liheng: Matthias: I think no more topic today :)
(4:38:10 PM) ***Matthias now needs to leave; have a nice weekend everyone ...
(4:38:21 PM) Matthias: liheng: yes, I think so too :-)
(4:38:51 PM) Matthias: bye all ...
(4:39:32 PM) liheng: That all ;)
(4:40:03 PM) makoto_goodday left the room.
(4:40:10 PM) cd_oo: Bye and have a nice weekend.
(4:40:35 PM) liheng: bye all
(4:42:34 PM) odf-mib: Bye all
(4:42:38 PM) xiuzhi: bye

Go back