Apache OpenOffice (AOO) Bugzilla – Issue 1585
16 second startup time is way too long.
Last modified: 2003-01-17 13:35:32 UTC
Soffice takes over 16 seconds to start up. This is just too long and will totally discourage me from using it. This is from an NFS mounted, 100bT connected home directory. AbiWord takes less than 3 seconds.
1. 605 is way too old. This almost kept me from taking this issue serious ;-) 2. Most of the users USE the application and don't spend their time to start it and shut it down. How long does it take to boot your system? 3. The vi takes less than one second... The more features, the bigger the application (=>more and bigger files). Nevertheless, performance is an issue that is constantly worked on (Believe it or not: Things happened between a 605 and now, see 1. and 8.). 4. Does your system have RAM and a CPU or just the fastest connection on this planet? - In other words: Please give further system details. 5. Did you give it a second chance (there is a lot of registering going on at the first startup, therefore the second one is faster)? 6 I doubt that the other applications go any faster than the word processor => changed from "word processor" to "framework" 7. A colleague of mine has the current version up in 5 seconds on his Linux machine at home. (See 8.) 8. Please verify your findings in with a 638 and comment. Thank you.
If it helps, I'd like to offer another data point. I just installed 638 on my laptop (installed on local disk, RedHat 7.0, 2.4.9 kernel, XFree86 4.1.0, 128MB RAM, 266MHz mobile Pentium.) I started "soffice" then shut it down then restarted it. It took 56 seconds to startup the second time. So, I agree that startup performance is a problem, at least in my case.
Please compare the current startup time with StarOffice 5.2. You will agree that we made great progress. There are 46733 different ways (maybe slightly less) to speed up a Linux system or to slow it down. This is not the forum how to speed up a Linux machine. Only one hint: Get familiar with "hdparm", DMA is deactivated by default on some systems.Hard disk cache is something to be aware off as well because HD performance is the key to performance concerning the startup. Neither is this a democratic process "still too slow on mine, do something because my macine is not outdated yet!" As I said before, we are constantly checking the options to get a better performance as well as checking meaningful features to get in the product. Time changes and a 266 MHz machine is simply slow. If you have a slow machine, don't use Linux unless you are aware of the things you need to do to get a performance you can live with.
Your comments are just plain rude and insulting! 638c now takes 36 seconds to start up! And yes I do know about hdparm! No it isn't a democracy, but it is a marketplace. My machines are not 266 MHz 386's. They are 500MHz 2way SMP machines which are probably representative of the upper end of machines in use today. Since I am not the only one complaining of slow start up times perhaps you should accept this as your problem and not react by pretending it must be the reporters fault! How many other problem reports are you ignoring because of your bad attitude?
Hi All, I'd like to add to this discussion, We too are experiencing very slow startup times under certain circumstances. We want to go into pilot for openoffice (638C) and there are three things between today and the start of our pilot: startup time, a working help system, and a working spellchecker. I'll elaborate on the first here, since it is relevant. Using OO on my PentiumIII 700Mhz, 192MB RAM laptop (and other machines strewn throughout our lab) can take anywhere between 35 / 67 seconds. Yes, first time after reboot is slower then the rest, but still..... Working in OO is usually fine, but switching OO to the background, and then back to the foreground has me sometimes waiting in excess of 2 minutes. Granted, my laptop setup of Linux can be tuned better (i _do_ have DMA, etc enabled, no funny stuff running in the background...) , but on the other hand, switching Mozilla, Opera, etc does not take that long, even in proportion to resource usage of the app in question. However, the pilot will not be run on my laptop :-) but from a server - this machine is a Pentium Xeon 800 Mhz, with 512 MB RAM (HP Netserver, SuSE Linux 7.2). We connect either straight X (from Windows clients) or X over SSH (Linux boxes). As with my laptop (and other machines) no specific tuning was done on this machine. We have startup times of around 6 seconds - not bad! I suspect the differences are in disk system, my laptop's disk is a plain IDE (and almost full), while the server's are very fast, RAID 5, SuperUltraWideMegaLightningTurboChargedOverClockedVerySexy SCSI. This suspicion is also based on the fact that the first user loads at 6 seconds, and this remains more or less constant (so far we have tested opening 10 sessions) - i.e. the startup time seems less dependent on resources as it is on diskspeed. Right now, we are working on devising a serious stress test. -Stefan, If you are the guy handling startup, we would be delighted to provide you detailed test plans and results, and you might want to suggest some possible test scenario's - we are happy to work with you on this. One of the idea's we are playing with is loading from a dedicated RAMDisk, but we don't know if this makes sense, and how to install to RAMDisk in a durable way (i.e. survive reboots, etc.) While this may sound a bit drastic, we eventually hope to serve to about 10.000 users, so perfomance is an issue. stefan, if you are interested in the testing bits, please drop me a line at mdekkers@unforgettable.com Hope this helped out a bit, Cheers, Martijn As to Stefan's points: 2.) Our usabillity research shows that a _LOT_ of users spend their time shutting down and starting again, rather then leaving OO running in the background. Education will help here, but user education sometimes seems an oxymoron.... However, you cannot compare stystem boot with application boot - As a user, I am starting an app, and expect response. My expectations are different when I boot my system. 638 is an _amazing_ improvement over 5.2 - without any doubt. Myslef, I counted 43672 ways to speed up a Linux system - please let me know what your other tricks are ;-)
Since I am only QA and not a developer, I will not keep on discussing this. There is newsgroups for discussions and issuezilla for defects. "Too slow in general" is an issue I can not specify, especially when several people with various findings get involved into one issue. As I said, one of my colleagues gets it up within a few seconds, therefore "...in general" can not be as true for every system. My conclusion: Not in general, but depending on the user's environment. I never said that the user's environment is equal to a user's fault. Due to the fact that I don't get paid for platform research but for product testing, I can't contribute more on this issue. Personal remark: My different sense of humor or my way to give things a "plastic description" (yes, this includes a little exaggeration sometimes) to make things clearer is something else than being rude or insulting. It's just that some people think that business (especially computing) and being funny can not mix. You're not alone, John and you are by far not the first to feel insulted when I try to find my way between communicating facts, getting things done with reasonable effort and not losing my sense of humor. SBA->MBA: Please take care of this item that seems to make the product unusable ;-) [<-See, there is a smiley!] Reassigned to Mathias.
Hi Guys, have you ever thought of technical reasons (besides hardware issues), why OO starts so slow in comparison to AbiWord, as you said. Have a look in OOs program directory on the size of its libs. Yes, we do provide our own toolkit, frameworks etc. Now go to /etc/ld.so.conf and have a look what kind of libs already get cached at system start. What prevents you from configuring OO there too, if start-up performance is really important for you? Don't forget to run ldconfig after that to load the libs into the cache. I'm pretty sure that you'll gain a performance boost. Start-up time for about 3-5 seconds on a PIII 700MHz with 256MB RAM is what I'd expect. Have fun.
Did this work for anybody? Please comment.
In general the startup time on Linux is not worse than the startup time on windows, if OO is started from a local disk. Starting it via NFS is slower, and different configurations may cause different results. We are working on that issue permanently, a startup is NEVER fast enough, but I'm not sure that Issuezilla is the right place to keep track of that work, because this will be an issue that will never get fixed. We have a dedicated team that monitors the startup performance and looks for ways to improve it. The startup times of every build are measured and compared. You can be sure that we are interested in optimizing the startup performance, this is a permanent issue. The startup performance you find in any build is the best we could achieve so far. On a PIII 500 with 256MB RAM (OO installed local, not NFS mounted) a first startup time after boot of 16s is OK for me, we will not gain more speed in the very near future. "OK" means not "I think this is enough", it only means "I know it's too slow, but I also know that this is not a simple bug". This is indeed an architectural problem that will be worked on continously, but it is nothing that can be treated as a bugfix. Please don't think we have a low opinion of your estimation, but I'd prefer to close this issue and treat it as an ongoing process, that can be discussed much better in our mailing lists. IMHO discussing several optimization strategies (we already have some ideas!) in the public will be more helpful. Comments welcome!
Because I didn't get any comments, I'll set the state to "Works for me". This means, the current startup performance is the best we can achieve currently. The optimization of this performance will stay a major goal in the framework team, but is an ongoing process.
*** Issue 2259 has been marked as a duplicate of this issue. ***
Just wanted to comment further on this point. I've used the Linux and Windows versions of OO here at our school for several months now. I agree that the load times do seem excessive. Is there a simple explanation for this in comparison to say, Word XP, a program that loads in 3-5 seconds. Not being picky understand, just curious. Does M$ have some sort of lock on speed because of unknown O.S. calls?
Hi Chuck, I think the behaviour you are reffering to (of WordXP) is because Word (since version 4) pre-loads most of the required .dll's at boot time, the argument being that a longer boot time of your PC is acceptable 9users are expecting an OS to take long to boot. An interesting way to check this (if you have the opportunity) is to time the boot from when the BIOS appears until your disk stops thrashing on a clean system. Then install Office, and do the same. Add the difference of the boot time (there _will_ be a significant difference) to the time it takes to start office (don't time the first office start - office will do some "housekeeping" - start office, reboot, and start office again) and compare that to the time it takes to start OOo. Martijn
*** Issue 3151 has been marked as a duplicate of this issue. ***
I close this task for the named reasons. Be sure that we don't forget to optimize our startup time. ;-)