Difference between revisions of "Performance/Meetings/2008 11"

Revision as of 02:43, 15 December 2009

Performance/Meetings/2009/11/27

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

Date: 2008/10/16
Time: 16:31– 17:16
Meeting No.:

Agenda:

(09:31:26 AM) kuangl: Hi Dieter
(09:31:31 AM) kuangl: hi mhu
(09:31:33 AM) mhu: Hi all
(09:31:39 AM) yugq: Hi, mhu
(09:32:02 AM) lihen1: hi,all
(09:33:02 AM) mhu: Hmm, I was supposed to excuse Dieter and Malte because they are busy with another meeting... ?
(09:33:31 AM) liangjunzeng: hello all
(09:33:49 AM) mhu: Hi liangjun
(09:33:51 AM) Malte: We decided to attend to this meeting instead of the other :)
(09:33:57 AM) liangjunzeng: :)
(09:34:08 AM) lihen1: Malte:thank you
(09:34:17 AM) mhu: okay
(09:34:21 AM) Dieter_: ...because this a very important project for us
(09:34:30 AM) Dieter_: lihen: Thank you for your very detailed document. I think it would be good to go through the document to get a common understanding.
(09:35:12 AM) lihen1: Dieter_:It's my pleasure
(09:35:52 AM) lihen1: another part for time-binder will be finish in a few days
(09:39:23 AM) mhu: sorry, I have just skipped over the document (read the first half) ... looks good so far
(09:40:12 AM) mhu: shall we quickly go through the document, as Dieter suggested ?
(09:43:45 AM) mhu: ping ... am I still online ... ?
(09:45:12 AM) mhu: ...anybody out there ?
(09:48:16 AM) lihen1: it's important unit of our new tools
(09:48:16 AM) lihen1: it can make it easy to measure effects of some key points
(09:48:16 AM) lihen1: Okay, can we continue to discuss the web server?
(09:48:16 AM) Malte: Can you please explain "time-binder"?
(09:48:16 AM) lihen1: Okay, we will put some native code for trace all node of Performance Roadmap, in every node we can add some time when it is executed, and that if we add 2 second time to a node ...
(09:48:16 AM) Dieter_: lihen: in your document the word "performance map" means a "sequence map" or "process flow map", right?
(09:48:16 AM) lihen1: and find the 10 seconds delay in macroscopical value, we can know how many time we can save
(09:48:16 AM) lihen1: Dieter_:Yes, we splited a macroscopical value to several key nodes
(09:48:16 AM) lihen1: Malte: can you understand me?
(09:48:18 AM) Malte: Next try ;)
(09:48:48 AM) yugq: yes
(09:49:05 AM) yugq: mhu, I can see you online.
(09:50:01 AM) mhu: hi all, Dieter Malte and I just found out that we have massive network delay in our chat connection (looks like 2-3 minutes delay) , sorry for that
(09:51:39 AM) mhu: yugq: thanks for your response, now I can hear you again; but we some seriously network delay ...
(09:52:30 AM) lihen1: Matthias:Okay, we can through it now, and I will explain some key point to Malte via email,
(09:52:30 AM) lihen1: Malte:Do you think so?
(09:52:30 AM) lihen1: Malte: Do you think so?
(09:52:30 AM) [1]Malte: Damn, I got disconnected again...
(09:52:30 AM) [1]Malte: Was there some comment on my naming suggestion?
(09:52:32 AM) lihen1: Matthias:can you receive my word?
(09:53:18 AM) lihen1: Dieter_:Yes, we splited a macroscopical value to several key nodes
(09:53:59 AM) mhu: LiHeng: yes I can read your words, but they seem to arrive blockwise with 2-3 minutes delay (after you wrote them)
(09:54:15 AM) lihen1: yugq:are you there?
(09:54:15 AM) yugq: yes?
(09:54:15 AM) yugq: liheng, I'm here.
(09:54:38 AM) lihen1: yugq:but I can't talk with Mhu,and the others
(09:54:39 AM) lihen1: :(
(09:54:47 AM) lihen1: network is worry?
(09:55:16 AM) yugq: I think there is some problem with their networks.
(09:58:47 AM) yugq: liheng, are you online now?
(09:58:47 AM) lihen1: Matthias:Do you think network is well enough to continue this meeting?
(09:58:47 AM) lihen1: yugq:I'm here but some trouble with network, ...
(09:58:47 AM) lihen1: :(
(09:59:49 AM) mhu: lihen1,all : sorry, I have just sent email explaining (our Hamburg) network problem; we see your words, but with minutes delay; yes I am patient enough to try to continue the meeting if you like :-)
(10:00:00 AM) [1]Malte: I think continuing IRC doesn't make sense... :(
(10:00:10 AM) [1]Malte: Could switch to email...
(10:00:23 AM) yugq: Agree
(10:00:42 AM) mhu: yugq: agree to what ?
(10:00:42 AM) [1]Malte: mailing-list, or direct email?
(10:01:38 AM) lihen1: direct email first
(10:01:45 AM) yugq: mhu, Malte said: "I think continuing IRC doesn't make sense... :(Could switch to email..."
(10:02:08 AM) [1]Malte: so who as all emails adresse at hand, and will start the first mail?
(10:02:09 AM) mhu: ah okay, I am only testing the reponsiveness :-)
(10:02:12 AM) lihen1: Matthias:We should switch to email
(10:02:38 AM) mhu: okay, we can continue via email then ...
(10:03:02 AM) [1]Malte: I suggest Li Heng sends out a first mail to get things started.
(10:03:05 AM) yugq: It seems the network is better now:)
(10:03:35 AM) [1]Malte: So we try to stick with IRC now?
(10:03:49 AM) mhu: yugq: that is what I was testing ...
(10:04:03 AM) [1]Malte: Seems to work now for me...
(10:04:29 AM) lihen1: [1]Malte:Seems better
(10:04:30 AM) lihen1: :)
(10:04:46 AM) mhu: I have sent out an email 7 minutes ago; has anyone of you received it ?
(10:05:08 AM) [1]Malte: yes
(10:05:36 AM) lihen1: Matthias:Yes
(10:05:49 AM) yugq: yes, received
(10:05:53 AM) mhu: okay, then at least the email fallback still works :-)
(10:08:10 AM) lihen1: but i think we ave lost many words
(10:08:10 AM) lihen1: but i think we have lost many words
(10:08:10 AM) lihen1: Malte:Did you receive the explaination of Time-binder?
(10:09:12 AM) mhu: I think, I have a log of all our conversation; I can send that via email when we are finished for today.
(10:10:15 AM) mhu: ...and, maybe email really is the better medium than IRC today; I think we have these delays again ...
(10:16:15 AM) mhu: okay, I am giving up for today, sorry. Bye for now, see you next week, or via email ...

Performance/Meetings/2008/11/20

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

Date: 2008/11/20
Time: 16:28– 17:24
Meeting No.:

Agenda:

(4:28:55 PM) Malte: Morning :)
(4:29:38 PM) liangjunzeng: hello all
(4:30:39 PM) kuangl: hi malte
(4:30:47 PM) Malte: Hi :)
(4:31:09 PM) mhu: Hi all ...
(4:31:22 PM) DL: Hi all!
(4:31:29 PM) kuangl: hi ,mhu
(4:31:32 PM) kuangl: hi DL
(4:31:49 PM) yugq: hi, all
(4:31:50 PM) lihen1: hi,mhu:)
(4:32:07 PM) DL: Hi Xiuzhi!
(4:32:13 PM) mhu: As we spoke in Beijing, I have invited Dieter Loeschky and Malte Timmermann to join our meeting.
(4:32:16 PM) lihen1: hi DL
(4:32:28 PM) Malte: Who is moderating this meeting? And is there an agenda?
(4:32:30 PM) mhu: liheng, xiuzhi: hi and moin
(4:32:31 PM) DL: hi lihen1
(4:32:35 PM) xiuzhi: DL:moin
(4:32:43 PM) xiuzhi: mhu: Hi,moin
(4:32:56 PM) DL: xiuzhi :-)
(4:33:26 PM) mhu: usually, Li Heng is moderating and sends out an agenda; I have just forwarded it to Dieter and Malte
(4:34:12 PM) lihen1: mhu:thank you,
(4:35:10 PM) Malte: OK, seems we have 2 points:
(4:35:12 PM) Malte: 1) Which operating systems and tools to use
(4:35:14 PM) Malte: 2) How to get some hardware in place...
(4:35:34 PM) lihen1: Hi, Malte:)
(4:35:42 PM) Malte: Did someone already play with automatic performance data gathering on different platforms?
(4:35:49 PM) Malte: My experience is mostly VTune...
(4:36:34 PM) lihen1: yugq and I have used Valgrind
(4:37:12 PM) Malte: Do you record ALL data with Valgrind, or with some filters and/or start/end sequences?
(4:37:22 PM) lihen1: and we used Valgrind also for RedOffice but only on memery leak
(4:37:51 PM) mhu: the other approach we discussed is to use internal instrumentation (RTL_LOG_*)
(4:38:17 PM) mhu: ...and possible extend that where necessary or appropriate.
(4:38:22 PM) lihen1: Malte:but we need also to discuss the new tools!
(4:38:24 PM) lihen1: :)
(4:38:44 PM) Malte: What are the "new" tools?
(4:38:52 PM) yugq: Malte, we used Valgrind mainly for find hotspots. Most of the thing is about configmgr.
(4:39:45 PM) lihen1: Performance Roadmap and a set of Tools for Benchmark hotspot and tracing
(4:40:43 PM) lihen1: we discuss them in Beijing, and we need to establish a server to keep our work in long-term
(4:41:40 PM) Malte: Understood - that's why I asked for the data collection tooling...
(4:42:11 PM) Malte: RTL_LOG_* is probably more lightweight than real profile data
(4:42:42 PM) Malte: Has advantages and disadvantages...
(4:42:54 PM) lihen1: yes,but is good for trace all hotspots though several version of OOo
(4:43:18 PM) Malte: Pro: Much less data
(4:43:36 PM) Malte: Con: You will have to find MANY places to add special RTL_LOG statements
(4:43:40 PM) mhu: ...and none of the tools is to be used exclusively, but they shall provide additional aspects.
(4:43:47 PM) lihen1: For comparison it is better than real profile data
(4:43:54 PM) Malte: Agree
(4:44:28 PM) Malte: Hint: RTL_LOG is not available in product builds, except special macros which I created for automatic performance tests
(4:44:45 PM) lihen1: Malte:not MANY but FIXED items in Performance roadmap
(4:44:47 PM) Malte: So currently you only have 6 marks in startup + load/save
(4:45:03 PM) mhu: yes, those special macros is what is interesting here.
(4:45:37 PM) Malte: Makes sense then...
(4:45:55 PM) Malte: But we need to define more marks then...
(4:45:57 PM) lihen1: did you read the document in attachment? It's a summary for new tools
(4:46:19 PM) mhu: yes, I have read it
(4:46:46 PM) Malte: I must admit that's it's not very self explaining for me.
(4:46:56 PM) Malte: But maybe it's just because of different wording
(4:46:58 PM) mhu: ...but I may need more explanation.
(4:47:02 PM) Malte: Will discuss with MHU later
(4:47:34 PM) Malte: OK, seems MHU won't be much help for me then ;)
(4:47:38 PM) Malte: Li - can you explain?
(4:47:56 PM) lihen1: Malte:On the other side, the real profile get only calling information, we need real Benchmark to trace time
(4:48:14 PM) lihen1: like you do that, RTL_LOG...
(4:48:31 PM) yugq: Malte, the roadmap is used to help to make more marks.
(4:49:34 PM) lihen1: yugq:you can describe your new roadmap of "loading" simply
(4:49:49 PM) yugq: First, we need a define of performance, and then insert the test code in the right place.
(4:50:00 PM) yugq: OK ...
(4:52:00 PM) yugq: The "load" is a sequence work flow. We seperate it to sevral parts. And test them in some way, and then find the slowest part.
(4:53:03 PM) yugq: The most import part is obvious the XML load part.
(4:53:49 PM) yugq: But document is composite with 4 basic xml files and more other files.
(4:54:17 PM) yugq: Load content xml is the most import part I think.
(4:56:14 PM) yugq: And about the road map, the "load" is a performance we defined. The "load xml" is an action during perform the load function. So we just need test all these kind of actions and get the result calculated to describe the performance.
(4:56:43 PM) yugq: I don't know if I explain it clearly:P
(4:56:57 PM) Malte: It's clearer now, thanks :)
(4:57:05 PM) Matthias: yugq: well, for me that was okay :-)
(4:57:27 PM) yugq: OK, thanks:)
(4:57:37 PM) Matthias: the point being, that we try to approach the identified problems systematically.
(4:58:43 PM) Matthias: ...and do a "mapping" of the issues (which code does what and contributes what) in a way that helps our understanding of the problems
(4:59:52 PM) lihen1: ...and trace all problems in following working
(4:59:55 PM) lihen1: :P
(5:00:34 PM) Matthias: yes
(5:01:30 PM) Dieter L: So, what are our action items for the next meeting?
(5:01:32 PM) yugq: Understanding what is performance is important! And it can help to classify the problem and help decide what we need do first.
(5:02:39 PM) lihen1: DL: how to create a server to store the benchmark data
(5:02:56 PM) Malte: One side note, which just came in my mind: With the RTL_LOG approach, we should agree to collect data only starting sometime in the future
(5:03:04 PM) Dieter L: lihen1: agree
(5:03:09 PM) Malte: For "historical" data, it would be difficult to get the marks in old versions...
(5:03:49 PM) Malte: To the server:
(5:03:59 PM) Malte: IMHO the software part is not very difficult to achieve
(5:04:05 PM) lihen1: Malte:If we have the stable performance roadmap we can store them on server and tracing them
(5:04:06 PM) Malte: Problem is getting the hardware
(5:04:29 PM) lihen1: I can getting the machine
(5:04:34 PM) Malte: Li: That's what I said - but probably then not for OLD OOo versions
(5:04:45 PM) Malte: Li - great :)
(5:05:12 PM) Matthias: [As a side note: I think, both Malte and I would probably appreciate some more explanations to the "Benchmark Tracer" document, maybe more examples or some other formulation that helps explain the used terms (like "numerical case-id" for example).]
(5:05:42 PM) Matthias: [...but we do that via email, I think]
(5:05:50 PM) lihen1: yes, I am working for it,...:P
(5:05:56 PM) Malte: Will it be a stable machine in your network, or available in the OOo network?
(5:05:56 PM) yugq: :)
(5:06:57 PM) lihen1: DL: last meeting in beijing we talk about storage method, but do you have some suggestion ,DB,CVS or ..
(5:08:02 PM) lihen1: Our company is an ISP in china, i think we can make server stable.
(5:08:04 PM) Malte: Li: We can thinks about that and discuss later
(5:09:44 PM) Matthias: previously, we discussed using a database to store the collected performance data; that should be easier to retrieve back and reorganize
(5:09:46 PM) Malte: Li: It would probably help if you start setting up a machine, and then piece by piece we work on the tooling
(5:09:49 PM) lihen1: ALL: I think the server can accept test application and do benchmark automic
(5:10:07 PM) Dieter L: lihen1: yes, let's discuss these technical issues (DB, CVS) later
(5:10:26 PM) lihen1: Dieter L:okay
(5:11:31 PM) Matthias: liheng: "can accept test application": you mean a developer can submit his test case to that machine and get back the results, right ?
(5:11:46 PM) lihen1: yes, i think so
(5:12:01 PM) Matthias: okay, that sounds good.
(5:12:29 PM) lihen1: it will be a powerful support to other developers
(5:13:00 PM) Matthias: and you are already working on setting up such a machine ?
(5:13:14 PM) lihen1: Matthias:so, I just make some web application for it
(5:13:40 PM) lihen1: Yes,
(5:13:59 PM) Matthias: okay, good. Maybe Malte is right, that we can add piece by piece once we have started.
(5:15:05 PM) lihen1: agree!
(5:15:23 PM) Malte: So should we end this meeting now, with the action item for Li to get and setup the machine? More discussions on email then...
(5:16:10 PM) lihen1: Is first step to run a case on this machine?
(5:16:21 PM) Matthias: okay; have you contacted Martin Hollmichel as where to commit the new tools in CVS / Subversion (i.e. which subdirectory of the "tools" project) ?
(5:17:35 PM) lihen1: No, Peter said he want apply to make performance project to be a accept project
(5:17:35 PM) Matthias: maybe, a first step could be for yugq and Malte and me to work on our understanding of the benchmark tooling, and commit it to CVS / SVN ?
(5:18:10 PM) Matthias: liheng: okay, we can also do that (make a new project), of course.
(5:18:34 PM) lihen1: :)
(5:19:14 PM) Matthias: I only think, it would be sufficient for now, to work with the existing "tools" project, but I am not against making a new one.
(5:19:44 PM) lihen1: Yes, sure
(5:19:51 PM) Matthias: liheng: do have an idea how and when Peter will do something ?
(5:20:05 PM) lihen1: It's important that work is going on :)
(5:20:27 PM) Matthias: yes, indeed. We should not wait for Peter :-)
(5:20:55 PM) Matthias: [he has so many things to do]
(5:21:09 PM) lihen1: Okay, we will work on understanding of benchmark tooling first:)
(5:21:36 PM) Matthias: well, yes that would be helpful I think, thanks.
(5:22:46 PM) lihen1: that's all for me
(5:23:08 PM) lihen1: Dieter L:thank you for your attending this meeting
(5:23:11 PM) Dieter L: I will take the action item to follow up with Peter on an Performance project at OOo
(5:23:17 PM) Matthias: That's all for me also.
(5:23:36 PM) Dieter L: I think we had a good meeting - thanks!
(5:23:50 PM) lihen1: My pleasure
(5:23:53 PM) Matthias: dieter: yes, that would be good (follow up with Peter)
(5:24:07 PM) Dieter L: Bye!
(5:24:12 PM) xiuzhi: bye
(5:24:16 PM) lihen1: Bye
(5:24:17 PM) Malte: Thanks for the meeting! :)

Performance/Meetings/2008/11/13

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

Date: 2008/11/13
Time: 16:56– 17:28
Meeting No.:

Agenda:

(4:57:13 PM) yugq: Hi, xiuzhi. It seems they have forgetton the IRC today.
(4:59:38 PM) xiuzhi: yugq:really?
(5:00:17 PM) xiuzhi: yugh: the start time is 5:00 ,isn't it?
(5:00:27 PM) yugq: Liheng said we start at 4:30.
(5:01:58 PM) xiuzhi: yugq: next week is a good time, IMHO
(5:02:12 PM) yugq: OK.
(5:02:28 PM) xiuzhi: yugq: Maybe they are tired or too busy this week
(5:02:46 PM) yugq: Oh. Yes.
(5:02:56 PM) yugq: :-D
(5:03:56 PM) xiuzhi: yugq:because it was a long trip, and nomorraly, the first week is very busy after the OOoCon
(5:05:10 PM) yugq: Yes. We will wait until 5:30.
(5:05:15 PM) yugq: :)
(5:05:47 PM) yugq: Maybe they are confused about the new time.
(5:06:02 PM) mhu: Hi all, sorry for being late...
(5:06:03 PM) yugq: Hi, mhu.
(5:06:09 PM) kuangl: hello mhu
(5:06:23 PM) lihen1: hi,mhu
(5:06:25 PM) mhu: I just arrived at my office...
(5:06:52 PM) lihen1: :)
(5:07:09 PM) yugq: mhu: It's good to see you at IRC again. Thank you very much for the little gift:-D
(5:07:28 PM) mhu: ...36 hours back in Hamburg...need to read >600 emails ... but already found your agenda for today...
(5:07:56 PM) mhu: yugq: you are welcome, it was nice to meet you in Beijing.
(5:08:51 PM) yugq: mhu: Nice to meet you, too.
(5:09:14 PM) mhu: ...inviting Dieter and Malte to our meeting...yes I wanted to do that...maybe we start next week ? ...
(5:10:11 PM) lihen1: Thank you,:)
(5:11:08 PM) xiuzhi: mhu: I just talked with yugq that you must be very tired this week
(5:11:13 PM) mhu: okay, I will then see that they can attend.
(5:12:33 PM) mhu: xiuzhi: oh, well, a little bit tired due to the timezone shift; I get tired early and wake up early, part of my body is still some hours shifted in Beijing direction :-)
(5:13:01 PM) lihen1: By the way, we can start our next IRC at 16:30 in Beijing, because we get off work at 17:30 week, and most people must take the bus:)
(5:13:41 PM) lihen1: can't you?
(5:13:49 PM) mhu: Liheng: okay, that is fine with me; it means 30 min earlier than today, right ?
(5:13:52 PM) xiuzhi: mhu: :)
(5:13:59 PM) lihen1: yes
(5:14:08 PM) mhu: xiuzhi: :-)
(5:14:26 PM) mhu: Liheng: okay, I mark my calendar appropriately.
(5:15:11 PM) lihen1: today, we can just talk only about "Which do developers want to pick up in reports? There is a simple template I maked in attachment " :)
(5:15:46 PM) mhu: okay, let me look at the attachment for a few seconds ...
(5:15:56 PM) lihen1: ok
(5:17:38 PM) mhu: well, the template looks fine so far, but for me to judge, I think we need to play a little with some datasets to see its usefullness.
(5:18:09 PM) mhu: that is, with an empty spreadsheet, I cant say much...
(5:18:39 PM) mhu: ...when there are data in it, I might better see what is missing (or too much).
(5:18:52 PM) mhu: but it is a good start.
(5:20:58 PM) lihen1: yes, if there are not essential data to add to the table, we will do a first test case with this template
(5:21:45 PM) lihen1: Would you like to start it with Save/Load case?
(5:21:46 PM) mhu: okay; I don't think anything essential is missing. It looks good so far.
(5:22:04 PM) mhu: We only have like 8 min left ?
(5:22:32 PM) lihen1: For xiuzhi it is, but not for me
(5:22:33 PM) lihen1: :)
(5:22:41 PM) mhu: okay :-)
(5:23:01 PM) xiuzhi: lihen1: :)
(5:23:17 PM) mhu: anyway, I think that would a good topic to discuss with Dieter and Malte, and have xiuzhi there as well.
(5:24:01 PM) mhu: (and I am not really prepared today also)
(5:24:02 PM) lihen1: Yes, we can discuss them at next time
(5:24:37 PM) mhu: okay
(5:25:14 PM) mhu: so, then I ask Dieter and Malte to join us next week, and we shift the time to 30 min earlier.
(5:25:33 PM) xiuzhi: mhu: good solution
(5:25:35 PM) lihen1: I think so, yugq cab also have enough time to finish the roadmap of Load/Save:)
(5:25:36 PM) mhu: ...and we discuss the Load / Save documents improvements
(5:26:03 PM) mhu: yes, a good preparation is always good
(5:26:30 PM) lihen1: Okay, that all today for me
(5:27:10 PM) mhu: yes, okay for me.
(5:27:18 PM) xiuzhi: nothing from me
(5:27:40 PM) yugq: bye
(5:27:44 PM) lihen1: :), bye
(5:27:50 PM) kuangl: bye
(5:27:50 PM) mhu: Thanks for all you did for us in Beijing; it has been a very good experience for me (and my wife), thanks again
(5:28:07 PM) lihen1: You are welcome!
(5:28:07 PM) mhu: bye, see you next week.
(5:28:08 PM) xiuzhi: mhu:we are pleasure
(5:28:21 PM) xiuzhi: mhu:8(bye)
(5:28:24 PM) lihen1: 8
(5:28:26 PM) lihen1: ;)
(5:28:27 PM) yugq: :)
(5:28:34 PM) mhu: bye :-)

Go back