Re: [Cooker] Boot-up speed
On Thu, 2003-10-30 at 12:02, Vedran Ljubovic wrote: > --- Buchan Milne <[EMAIL PROTECTED]> wrote: > > Yes, just accept the username and password, and show > > some moving > > graphics until you have authenticated the user. > > > > Ever noticed how long it takes to tell you your > > password is wrong on > > first start? > > Couldn't we just hack the bootsplash so that it shows > fields for login and password :) Now THAT sounds cool to me. LX -- °°° Linux Mandrake 9.1 Kernel 2.4.21-0.13mdk *Catch Star Trek Enterprise, Wednesdays on UPN*
Re: [Cooker] Boot-up speed
--- FACORAT Fabrice <[EMAIL PROTECTED]> wrote: > put them at the end after dm start Wouldn't work, because dm service just calls prefdm with a fork, so it's effectively executing X and KDE and everything else in parallel. As I've said, this makes X start veeery slowly. When I remove the fork, the service just never ends, so whatever is after it never gets started. --- Han Boetes <[EMAIL PROTECTED]> wrote: > Tweaking the bootstuff isn't hard. It's very > rewarding for people who want to > learn something about linux and having 30 seconds > extra boottime is not a > mayor problem since linux is not a reboot OS. --- guran <[EMAIL PROTECTED]> wrote: > Otherwise I think that Linux or Unix has other > walues than a fast start. > regards Well, I don't know about you two but I would like to see Mandrake become better in all aspects, even if some of them are not directly useful to myself. Wrt reboot issue, I keep my PC in the same room where I sleep, and I can't fall asleep when it works because my athlon coolers make too much noise. __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
I lost where this particular thread came in, especially since I posted a message titled "Boot Linux Faster". So at the risk of being repetitive here it is again, since the article seems to address some of the concerns being raised here: Date: 24/08/2003 3:06 pm Hello, I don't know if any one else saw this article but it seemed very good to me. http://www-106.ibm.com/developerworks/linux/library/l-boot.html A chap at IBM Hursley, in the UK, has come up with a way to make Linux boot faster using make. I have not got around to trying it out. However, given the speed that XP now starts up compared to most Linux setups I would think anything to get the boot times down must be good. I responded to this e-mail asking why everything needs to be started at boot time. For example it takes a huge amount of time to get CUPS up and running. Why not defer this until the first print job is required ? It appears that extension of his idea could be to work out which services really need to be started at boot time and which might be deferred. Anyway, food for thought, Owen I was suggesting that CUPS and the like should be started when the first attempt to print, or whatever, is made. I don't know if that is actually possible but it seems to me to be somewhat more elegant than just waiting for 60 seconds. What happens if a print request IS made during that 60s, does it just vanish into the ether ? A follow up post on this thread suggested that XP was spoofing by clever use of changing eye candy. Although this does play a small role XP IS a lot faster at getting to the desktop. Since Linux is making a play for the desktop it cannot be under estimated how important this speed, perceived or otherwise, is with the average punter. I generally find that Window$ users are generally not a forgiving lot, especially when their mental picture / world order is under threat !!! The idea of using any kind of suspend does not address the key issue, neither does the argument that Linux is not designed as a reboot OS. Linux is going to get rebooted and all the more often as it starts to make inroads onto the desktop. Linux is still legal, despite the best efforts of large vested interests and monopolies and their pet US government ! We need to give users a better user experience than M$, or anyone else, and that includes bootup and shutdown. I find it unacceptable to wait 4 minutes, even on my PIII 500. Thanks again, Owen On Friday 31 Oct 2003 3:33 am, Jos wrote: > On Thursday 30 Oct 2003 23:06, guran wrote: > > Is it possible to use swsusp as a regular "halt-susp" > > to get around the need of everyone who thinks that the next second is > > important. > > Otherwise I think that Linux or Unix has other walues than a fast start. > > regards > > guran > > Well... about the fact swsusp is even existing in Mandrake 9.2 I got my own > not so very positive ideas. Of course I tested it, and as expected it > didn't work well. Swsusp is far from mature on linux -and will never be in > 2.4 kernels-. Swsusp is a nice way to prevent a laptop from loosing it's > data when the battery runs out, not a way to reboot a system day in day > out. > > Sorry, but I've had it with the guys who say "Mandrake works fine with me, > so anyone stating otherwise should shut up", and unfortunately there are > far too many of them on this list... > > Jos
Re: [Cooker] Boot-up speed
On Thursday 30 Oct 2003 23:06, guran wrote: > Is it possible to use swsusp as a regular "halt-susp" > to get around the need of everyone who thinks that the next second is > important. > Otherwise I think that Linux or Unix has other walues than a fast start. > regards > guran Well... about the fact swsusp is even existing in Mandrake 9.2 I got my own not so very positive ideas. Of course I tested it, and as expected it didn't work well. Swsusp is far from mature on linux -and will never be in 2.4 kernels-. Swsusp is a nice way to prevent a laptop from loosing it's data when the battery runs out, not a way to reboot a system day in day out. Sorry, but I've had it with the guys who say "Mandrake works fine with me, so anyone stating otherwise should shut up", and unfortunately there are far too many of them on this list... Jos
Re: [Cooker] Boot-up speed
Adam Williamson wrote: > On Thu, 2003-10-30 at 22:29, Han Boetes wrote: > > Adam Williamson wrote: > > > On Thu, 2003-10-30 at 19:41, Han Boetes wrote: > > > > > > > mayor problem since linux is not a reboot OS. > > > > ^^^ > > please quote carefully. This hardly reflects what I said. > > Uh? It's exactly what you said. You argued that bootup time is > irrelevant "since linux is not a reboot OS". I point out that there are > perfectly normal real-world situations in which end users do in fact > boot Linux frequently. Where's my flaw? > > > > Well. It's certainly designed to be capable of running more or less > > > perpetually, that's true. That doesn't mean that's how people use it, > > > though. I boot my system at a minimum once a day, because I power it off > > > every night in order to save power. I expect most non-geek home users do > > > the same with their systems. > > > > You just repeat an already used argument. > > Whether or not it's already used is irrelevant; it's not one I've seen > *refuted*, which is all that matters. Right... # Han -- http://www.xs4all.nl/~hanb/software http://www.xs4all.nl/~hanb/documents/quotingguide.html
Re: [Cooker] Boot-up speed
On Thu, 2003-10-30 at 22:29, Han Boetes wrote: > Adam Williamson wrote: > > On Thu, 2003-10-30 at 19:41, Han Boetes wrote: > > > > > mayor problem since linux is not a reboot OS. > > ^^^ > please quote carefully. This hardly reflects what I said. Uh? It's exactly what you said. You argued that bootup time is irrelevant "since linux is not a reboot OS". I point out that there are perfectly normal real-world situations in which end users do in fact boot Linux frequently. Where's my flaw? > > Well. It's certainly designed to be capable of running more or less > > perpetually, that's true. That doesn't mean that's how people use it, > > though. I boot my system at a minimum once a day, because I power it off > > every night in order to save power. I expect most non-geek home users do > > the same with their systems. > > You just repeat an already used argument. Whether or not it's already used is irrelevant; it's not one I've seen *refuted*, which is all that matters. -- adamw
Re: [Cooker] Boot-up speed
Is it possible to use swsusp as a regular "halt-susp" to get around the need of everyone who thinks that the next second is important. Otherwise I think that Linux or Unix has other walues than a fast start. regards guran -- Mandrake Linux 9.2 kernel-2.4.22.21mdk-1-1mdk Only in a society that has 'a priori' defined what is the truth can the result from the evolution of life be defined false.
Re: [Cooker] Boot-up speed
Adam Williamson wrote: > On Thu, 2003-10-30 at 19:41, Han Boetes wrote: > > > mayor problem since linux is not a reboot OS. ^^^ please quote carefully. This hardly reflects what I said. > Well. It's certainly designed to be capable of running more or less > perpetually, that's true. That doesn't mean that's how people use it, > though. I boot my system at a minimum once a day, because I power it off > every night in order to save power. I expect most non-geek home users do > the same with their systems. You just repeat an already used argument. # Han -- http://www.xs4all.nl/~hanb/software http://www.xs4all.nl/~hanb/documents/quotingguide.html
Re: [Cooker] Boot-up speed
On Thu, 2003-10-30 at 19:41, Han Boetes wrote: > mayor problem since linux is not a reboot OS. Well. It's certainly designed to be capable of running more or less perpetually, that's true. That doesn't mean that's how people use it, though. I boot my system at a minimum once a day, because I power it off every night in order to save power. I expect most non-geek home users do the same with their systems. -- adamw
Re: [Cooker] Boot-up speed
Vedran Ljubovic wrote: > - I also agree that using bash slows down things. Time it. You're making conclusions based on assumptions. > But I don't think that writing rc script itself in C will help much. rc script > is only a small portion of overhead, I think that the bigger problem are > scripts in init.d that are 100% bash, such as devfsd script that makes a bunch > of symbolic links and stuff. C isn't human readable. Tweaking the bootstuff isn't hard. It's very rewarding for people who want to learn something about linux and having 30 seconds extra boottime is not a mayor problem since linux is not a reboot OS. What _is_ hard is to make a proper init for _everyone_. # Han -- http://www.xs4all.nl/~hanb/software http://www.xs4all.nl/~hanb/documents/quotingguide.html
Re: [Cooker] Boot-up speed
Le jeu 30/10/2003 à 17:01, Vedran Ljubovic a écrit : > --- Jos Hulzink <[EMAIL PROTECTED]> wrote: > > > > Also shared printers can be set up after the desktop > > is there, though > > indeed, for real servers you might want to change > > priority. My problem is > > how to do implement your suggestions (which I agree > > with) in a neat way? > > Also consider that some people use a dm to login, > while others use auto-login feature and go directly to > desktop... it's not easy, there are really two > realistic options: > - start services when needed (e.g. start cups only > when i try to print something) but it would require > hacking kde wrong point, or valid only for desktop PC 1°/ in workstation or when you share the printer, others computers should be able to see and print to it 2°/ ok, you hack kde and gnome. and WM, fluxbox etc ... or if i use CLI ? now you need to hack directly lpr > or > - use an arbitrary delay before starting a service and > renice it so that user experience is not harmed. > I think this second solution is not that bad, what do > you think? put them at the end after dm start -- La tolerance est la charite de l'intelligence. -- Jules Lemaitre
Re: [Cooker] Boot-up speed
BLINDAUER Emmanuel wrote: Le Mercredi 29 Octobre 2003 00:41, Han Boetes a écrit : S01xfs S02dm S03iptables S09network Yes but dm needs networking at the least. This solution may work for you but mandrake has to have sane defaults for everyone. Which is the general problem with suggestions. Most people only take their own preferences and situation in account. Starting X and the correspondant dm take some time, perhaps 2 ou 3 second, to be operational to input a login name. I think that this time can be used to start network and related stuff. other process like postfix or cups can be started at later time, because the logging action take some time Since it seems that networking is one of the bottlenecks: Having networking up is generally just to have the loopback interface running. Without knowing more, does dm's/X need more than that really? I'd then propose to set up net interfaces in parallell: loopback - guess it starts 'instantly', fine for most dependencies eth0 - dhcp/zcip might take some time: comes up when it comes up eth1 - static? fine & dandy! ... ethN The only problem I see is services that are bound to specific ip:s. How does Apache react when it want's to bind to a specific ip that doesn't exist? On the other hand, starting interfaces in the order of loopback, static interfaces and last dhcp/zcip and returning a [OK] after static is done would solve that problem. Yes, this is maybe a solution for my needs, but it seems like it could catch many situations in the wild. Maybe initscripts..rpm should fork a initscripts-hiavail for servers... Or better: check to see if we are in runlevel 5 or not, since a server normally don't run X (or?). Just my swedish krona... /m
Re: [Cooker] Boot-up speed
--- Buchan Milne <[EMAIL PROTECTED]> wrote: > Yes, just accept the username and password, and show > some moving > graphics until you have authenticated the user. > > Ever noticed how long it takes to tell you your > password is wrong on > first start? Couldn't we just hack the bootsplash so that it shows fields for login and password :) __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
Hello, I agree with most of your points. --- Jos Hulzink <[EMAIL PROTECTED]> wrote: > On Thu, 30 Oct 2003, Vedran Ljubovic wrote: > > - I don't think parallelizing helps on computers > with > > a single CPU that doesn't support hyperthreading. > Sure > > Not true I think... Think about initializing network > cards with a dhcp > server. This isn't a busy wait. It is one of the > few examples that really > matter, but ok. We need benchmarks to see what's > true. You could isolate all such cases and fork & them. I agree that it would be much nicer to have a signle file that sets all dependencies in a clean style. But it a) takes time to implement, and more importantly bug-proof! (but if you already did it, then kudos to you!) b) is nonstandard - as in administrators coming from other unix/linux systems would be completely lost. > > * people not sharing printers and scanners can > afford > > to start cups and scannerdrake after the desktop > is up > > (10 seconds) > > Also shared printers can be set up after the desktop > is there, though > indeed, for real servers you might want to change > priority. My problem is > how to do implement your suggestions (which I agree > with) in a neat way? Also consider that some people use a dm to login, while others use auto-login feature and go directly to desktop... it's not easy, there are really two realistic options: - start services when needed (e.g. start cups only when i try to print something) but it would require hacking kde or - use an arbitrary delay before starting a service and renice it so that user experience is not harmed. I think this second solution is not that bad, what do you think? > If your network card is useless, you can't open > email :) So it's not that > easy... What is the limit of non-working hardware ? Ok, so we agree that network is immediately necessary for desktop, while printing and scanning are not. What else? Through such discussion we could reach a list of services that can be delayed until after the desktop is up. > away. I used a > binary for /etc/rc.d/rc for that script isn't > interesting, and > scripts don't allow multithreading. Another solution > hacks init, > drops /etc/rc.d/rc and calls everything directly. > Question is: what is > right ? do we (Mandrake) care about dropping > standards in favour of a few > seconds ? I've read a very convincing post on Slashdot about this. Basically, bash is loaded immediately after kernel is booted and subsequent calls do not cause any overhead. Bash is also very fast. I'd like to see exact benchmarks of using c vs. bash in any of these scripts, given that all other circumstances are the same. Also, if you are going to use posix threads in C than good for you, but otherwise old-style threads are nothing but processes and multi-processing can be achieved in shell with fork operator. > > One another thing: consider the services that > already > > work in background: scannerdrake, usb detection, > xfs, > > apmd, cups, crond, xinetd, kheader, devfsd. > > Parallelizing tasks is as easy as adding a & to > > appropriate script. > > And finding the dependencies, which is a nice puzzle > sometimes :) > > Jos > __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
Le Mercredi 29 Octobre 2003 00:41, Han Boetes a écrit : > > S01xfs > > S02dm > > S03iptables > > S09network > > Yes but dm needs networking at the least. This solution may work for > you but mandrake has to have sane defaults for everyone. Which is the > general problem with suggestions. Most people only take their own > preferences and situation in account. Starting X and the correspondant dm take some time, perhaps 2 ou 3 second, to be operational to input a login name. I think that this time can be used to start network and related stuff. other process like postfix or cups can be started at later time, because the logging action take some time I've tried my own solution at home, who was booting a ltsp terminal. between the bootup of the computer (pressing the power button) and getting a login screen: 60s: booting on a disquette some bytes are read -> netboot -> DL of the kernel, booting the kernel, setting network, launching X, READY!
Re: [Cooker] Boot-up speed
Buchan Milne wrote: Only load services when they're needed (or after gui is up) I think that's what WinXP does. It's not that simple. For me to be able to log in (on my desktop in a LAN), I need at least NFS (ie portmap, nfslock) and autofs up and running. For a disconnected LDAP setup (see the article on mandrakesecure.net), I need LDAP up and running too (otherwise is is not possible to login, or show user lists if using user lists on the dm). I'm guessing here, but when booting Mandrake on a reasonably fast machine (2.4GHz), the major delays I see are the depmod and activities which are looking for things which aren't there, such as NFS mounts from other machines which aren't up, or LAN adapters which aren't in use (e.g. a laptop NIC which is only used in-home or in-office). When these things ARE there, startup proceeds quickly. The problem may be the timeouts involved in looking for things you're not going to find. This is *definitely* a problem in shutdown. If, for example, my desktop mounts an NFS directory from my laptop, and I then shut the laptop down, shutting down the desktop will retry umount for the mount 3 times while shutting down NFS, and another 3 times during the final "unmounting file systems" phase. Each retry takes something like 90-120 seconds. Granted, on the particular issue of NFS, it's probably possible to tweak things to make this better, but that doesn't hold in every case. My laptop, when used in-house, often has its on-board NIC plugged into my internal LAN. If it doesn't (when on the road), there is a delay of at least a minute trying to bring up eth0. Part of the solution may simply be to revisit these timeouts, and adopt the philosophy that in today's networks, if you don't find it in a few seconds, it's probably not there, and you should either abandon the attempt or else background it. Another part of the solution would be a way to let particular components "block" on the availability of others, so that one component's initialization can proceed asynchronously, and a second component only blocks from the time it decides it requires the service until the time it is ready. A partial solution would be to create a few parallel initialization streams each of which serially runs components which depend upon one another.
Re: [Cooker] Boot-up speed
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Jos Hulzink wrote: > On Thu, 30 Oct 2003, Michael Lothian wrote: > > >>Would it be possible to have a desktop pc option at install where all >>prioritories are for a desktop user >> >>Make the gui no 1 prioritory at boot time >> >>Only load services when they're needed (or after gui is up) >> >>I think that's what WinXP does. > > > Indeed, but this is impossible due to the fact Xfree acts like a > monolythic block of concrete that relies on networking and stuff. As long > as linux gets no decent separation between accellerated drivers and an X > that supports this, (see also KGI, XGGI, fbdev) this is not an option. > > Mind that even in XP you can't log in untill your networking is up, if you > are logging on to a domain. But it sure looks 1000 x faster and more > smooth. Yes, just accept the username and password, and show some moving graphics until you have authenticated the user. Ever noticed how long it takes to tell you your password is wrong on first start? Regards, Buchan - -- |--Another happy Mandrake Club member--| Buchan MilneMechanical Engineer, Network Manager Cellphone * Work+27 82 472 2231 * +27 21 8828820x202 Stellenbosch Automotive Engineering http://www.cae.co.za GPG Key http://ranger.dnsalias.com/bgmilne.asc 1024D/60D204A7 2919 E232 5610 A038 87B1 72D6 AC92 BA50 60D2 04A7 -BEGIN PGP SIGNATURE- Version: GnuPG v1.2.3 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQE/oRCtrJK6UGDSBKcRAtpgAJ0c8l24eHCb/QUigsDcH9AH12YfuACgyi2d sNSR8QqrO58Rag+AVgX9fBw= =lufS -END PGP SIGNATURE-
Re: [Cooker] Boot-up speed
On Thu, 30 Oct 2003, Michael Lothian wrote: > Would it be possible to have a desktop pc option at install where all > prioritories are for a desktop user > > Make the gui no 1 prioritory at boot time > > Only load services when they're needed (or after gui is up) > > I think that's what WinXP does. Indeed, but this is impossible due to the fact Xfree acts like a monolythic block of concrete that relies on networking and stuff. As long as linux gets no decent separation between accellerated drivers and an X that supports this, (see also KGI, XGGI, fbdev) this is not an option. Mind that even in XP you can't log in untill your networking is up, if you are logging on to a domain. But it sure looks 1000 x faster and more smooth. Jos
Re: [Cooker] Boot-up speed
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Michael Lothian wrote: > Would it be possible to have a desktop pc option at install where all > prioritories are for a desktop user > > Make the gui no 1 prioritory at boot time > > Only load services when they're needed (or after gui is up) > > I think that's what WinXP does. It's not that simple. For me to be able to log in (on my desktop in a LAN), I need at least NFS (ie portmap, nfslock) and autofs up and running. For a disconnected LDAP setup (see the article on mandrakesecure.net), I need LDAP up and running too (otherwise is is not possible to login, or show user lists if using user lists on the dm). So, maybe a "Windows XP home" type scenario and a "Windows XP Pro in Windows domain"-type scenario. Regards, Buchan - -- |--Another happy Mandrake Club member--| Buchan MilneMechanical Engineer, Network Manager Cellphone * Work+27 82 472 2231 * +27 21 8828820x202 Stellenbosch Automotive Engineering http://www.cae.co.za GPG Key http://ranger.dnsalias.com/bgmilne.asc 1024D/60D204A7 2919 E232 5610 A038 87B1 72D6 AC92 BA50 60D2 04A7 -BEGIN PGP SIGNATURE- Version: GnuPG v1.2.3 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQE/oQizrJK6UGDSBKcRAqbZAJ0SQriYn+c9iq5d4fJM/pD39r6pbwCfWw/G wXF8Ov+0fK2uq0Foq9tLCBM= =eiTP -END PGP SIGNATURE-
Re: [Cooker] Boot-up speed
Le jeu 30/10/2003 à 12:01, Jos Hulzink a écrit : > On Thu, 30 Oct 2003, Vedran Ljubovic wrote: > > Here are some thinkings: > > > > - Harddrake should *not* be disabled by default. If > > one changes some piece of hardware (they'll probably > > do that while their system is off :) strange things > > may happen to their system. You may not change your > > hardware often, but I for example have my case on > > Removing harddrake might be sensible for laptops and > > computers that are known to be black-boxes (servers, > > non tech-savvy users; I'd like to make that call > > myself). > > My point is that someone who changes hardware knows what he is doing, and > pressing "H" at boot time to run harddrake is only a small step after > hacking your hardware. you're wrong. 1°/ linux users since 5 years, cooker user since 3 years, sysadmin and i don't even know that pressing H was launching harddrake ... Where this feature is writed and easily accessible to someone who don't even try to find it ? 2°/ people changing hardware don't know what they do, especially on linux. Normally if you change your hardware ( put USB/Firewire devices, or add a CDROM/CDRWM ), you expect it should work automatically ( how many people praise XP when it was able to detect and configure alone their hardware when XP had the correct drive ? of course when something was wrong ... ). Even for graphic card/monitor we all waiting for the day when we will be able to switch the graphic card and that linux/mdk will detect this at boot and simply automatically configure it, etc ... > 95% of the users will never change hardware, or it > must be an usb stick, which should not heb handled by harddrake, for > harddrake only detects at boot time. (My system comes with /dev/sda1 not > found while mounting filesystems for I happened to have an USB stick in my > machine during the install of 9.2. Have been too lazy to fix it yet.) or a USB printer, or a new USB keyboard/mouse to replace your PS/2 keyboard/mouse, a USB HD, an USB camera, ... > > and other such examples. That's 18 seconds or more > > than 20% of overall boot-up time. I don't think > > parallelization can bring that much benefit. > > Also consider splitting harddrake in part that is > > needed to run X (detecting video card, monitor and > > mouse) and the rest that runs after the desktop is up. > > It can even show a nice GUI for new hardware :) What > > else can be delayed after the GUI is up? > > Harddrake isn't _THAT_ slow, splitting it up would cause more overhead. > > /me wonders if XFree can be stripped. 99 % of the users will never use the > remote connect features of X. I'm not an X expert, are the remote connect > features and such loaded when X starts ? the network support in XFree doesn't slow XFree startup. culprits here are the toolkits, the De ( gnome/KDE ) and ldd. the network support may slow XFree during usage ( i.e not everything is optimized for direct access, but that's a wrong debate ) --- Faites-vous des amis prompts a vous censurer. -- Nicolas Boileau, Satires
Re: [Cooker] Boot-up speed
Would it be possible to have a desktop pc option at install where all prioritories are for a desktop user Make the gui no 1 prioritory at boot time Only load services when they're needed (or after gui is up) I think that's what WinXP does. Mike
Re: [Cooker] Boot-up speed
On Thu, 30 Oct 2003, Vedran Ljubovic wrote: > --- Jos <[EMAIL PROTECTED]> wrote: > > http://www.stack.nl/~josh/Mandrake > > Hello Jos! > > I think you have an excellent page there. > > Here are some thinkings: > > - Harddrake should *not* be disabled by default. If > one changes some piece of hardware (they'll probably > do that while their system is off :) strange things > may happen to their system. You may not change your > hardware often, but I for example have my case on > Removing harddrake might be sensible for laptops and > computers that are known to be black-boxes (servers, > non tech-savvy users; I'd like to make that call > myself). My point is that someone who changes hardware knows what he is doing, and pressing "H" at boot time to run harddrake is only a small step after hacking your hardware. 95% of the users will never change hardware, or it must be an usb stick, which should not heb handled by harddrake, for harddrake only detects at boot time. (My system comes with /dev/sda1 not found while mounting filesystems for I happened to have an USB stick in my machine during the install of 9.2. Have been too lazy to fix it yet.) > - I don't think parallelizing helps on computers with > a single CPU that doesn't support hyperthreading. Sure Not true I think... Think about initializing network cards with a dhcp server. This isn't a busy wait. It is one of the few examples that really matter, but ok. We need benchmarks to see what's true. > we both need hard figures to back up our claims, but I > believe the effort can be spent much better by > rethinking some services and making their selection > smarter! Think about it: > * laptops don't need harddrake (4 seconds) > * people without a network card don't need > NFS/SMB/TMDNS (3-4 seconds) > * people not sharing printers and scanners can afford > to start cups and scannerdrake after the desktop is up > (10 seconds) Also shared printers can be set up after the desktop is there, though indeed, for real servers you might want to change priority. My problem is how to do implement your suggestions (which I agree with) in a neat way? > and other such examples. That's 18 seconds or more > than 20% of overall boot-up time. I don't think > parallelization can bring that much benefit. > Also consider splitting harddrake in part that is > needed to run X (detecting video card, monitor and > mouse) and the rest that runs after the desktop is up. > It can even show a nice GUI for new hardware :) What > else can be delayed after the GUI is up? Harddrake isn't _THAT_ slow, splitting it up would cause more overhead. /me wonders if XFree can be stripped. 99 % of the users will never use the remote connect features of X. I'm not an X expert, are the remote connect features and such loaded when X starts ? Oh and of course the reason X / KDE / Gnome boots so slow: the ELF file format... > To people bashing windows for this: please think about > it, it's a feature. I want to start doing productive > things as soon as possible. This is a desktop machine > we're talking about. I don't care if 90% of hardware > isn't working yet if I can startup my > editor/browser/e-mail program (sometimes I want to > hack rc so openoffice is started before everything > else, so i don't have to wait 20 sec for it ;) If your network card is useless, you can't open email :) So it's not that easy... What is the limit of non-working hardware ? > - I agree completely on depmod. > > - I also agree that using bash slows down things. But > I don't think that writing rc script itself in C will > help much. rc script is only a small portion of > overhead, I think that the bigger problem are scripts > in init.d that are 100% bash, such as devfsd script > that makes a bunch of symbolic links and stuff. That's not the issue. The issue is that a binary isn't hackable anymore. Using binaries will cause some unix freaks to run away. I used a binary for /etc/rc.d/rc for that script isn't interesting, and scripts don't allow multithreading. Another solution hacks init, drops /etc/rc.d/rc and calls everything directly. Question is: what is right ? do we (Mandrake) care about dropping standards in favour of a few seconds ? > One another thing: consider the services that already > work in background: scannerdrake, usb detection, xfs, > apmd, cups, crond, xinetd, kheader, devfsd. > Parallelizing tasks is as easy as adding a & to > appropriate script. And finding the dependencies, which is a nice puzzle sometimes :) Jos
Re: [Cooker] Boot-up speed
--- Jos <[EMAIL PROTECTED]> wrote: > http://www.stack.nl/~josh/Mandrake Hello Jos! I think you have an excellent page there. Here are some thinkings: - Harddrake should *not* be disabled by default. If one changes some piece of hardware (they'll probably do that while their system is off :) strange things may happen to their system. You may not change your hardware often, but I for example have my case on table sitting open so I could quickly change sth if I wanted to. Just last year I upgraded my memory three times, added a hard disk, a burner, and many other tiny changes I can't remember now. This is how many people use their PCs, including people that consider themselves competent enough to mess inside their computer case but shell scares the hell out of them. At work we change monitors, keyboards and mice all the time (because all are really crappy and everyone tries to get a better one) and all Linux computers tend to freak out when the monitor is changed. Removing harddrake might be sensible for laptops and computers that are known to be black-boxes (servers, non tech-savvy users; I'd like to make that call myself). - I don't think parallelizing helps on computers with a single CPU that doesn't support hyperthreading. Sure we both need hard figures to back up our claims, but I believe the effort can be spent much better by rethinking some services and making their selection smarter! Think about it: * laptops don't need harddrake (4 seconds) * people without a network card don't need NFS/SMB/TMDNS (3-4 seconds) * people not sharing printers and scanners can afford to start cups and scannerdrake after the desktop is up (10 seconds) and other such examples. That's 18 seconds or more than 20% of overall boot-up time. I don't think parallelization can bring that much benefit. Also consider splitting harddrake in part that is needed to run X (detecting video card, monitor and mouse) and the rest that runs after the desktop is up. It can even show a nice GUI for new hardware :) What else can be delayed after the GUI is up? To people bashing windows for this: please think about it, it's a feature. I want to start doing productive things as soon as possible. This is a desktop machine we're talking about. I don't care if 90% of hardware isn't working yet if I can startup my editor/browser/e-mail program (sometimes I want to hack rc so openoffice is started before everything else, so i don't have to wait 20 sec for it ;) - I agree completely on depmod. - I also agree that using bash slows down things. But I don't think that writing rc script itself in C will help much. rc script is only a small portion of overhead, I think that the bigger problem are scripts in init.d that are 100% bash, such as devfsd script that makes a bunch of symbolic links and stuff. One another thing: consider the services that already work in background: scannerdrake, usb detection, xfs, apmd, cups, crond, xinetd, kheader, devfsd. Parallelizing tasks is as easy as adding a & to appropriate script. __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
On Wednesday 29 Oct 2003 10:48, Frank Griffin wrote: > Jos wrote: > >I did some testing myself, working on something similar, though I still > > don't know what people prefer: that I hack init, or replace /etc/rc.d/rc > > by a binary... > > Is the problem that the necessary initializations take too long, or that > they take too long when run serially under a single thread ? 1) We do too many things. One example is: running depmod every boot is nonsense: if modules are installed by rpm, rpm can call depmod. If the user installs a kernel by hand, make modules_install calls depmod. If the user hacks the modules by hand he is on his own. Depmod -A is supposed to be fast, but it still requires scanning a directory structure which hasn't been cached yet. Hacking together a kernel.h at each boot is nice for the few guys who change kernel every five minutes, not for the mainstream systems. My Mandrake starts NFS support by default. IMHO I should see no NFS services started unless configured by the user. Now they are started with empty tables. 2) We do them serial, which consumes extra time. Parallel is faster, the few reports of slower boot times are due to incomplete service dependency tables, causing services to timeout waiting for other services. 3) Ok, many things done don't take much time. but 20x not much time is still much time. 4) We don't give the user the feeling the system is quick. The famous "Windows XP is fast" remarks hold, for Windows XP makes the user think it is fast.This is almost unresolvable unless we can drop XFree86, which is utter bloatware. For init 3 (console) I managed to show the console within 3 seconds after init entered runlevel 3. WOW that looked fast. Sure, I can't do much for my network wasn't up yet, but the psychological effect was huge. Jos Windows XP is a trademark of the company which will probably sue me for comparing it to Linux.
Re: [Cooker] Boot-up speed
Jos wrote: I did some testing myself, working on something similar, though I still don't know what people prefer: that I hack init, or replace /etc/rc.d/rc by a binary... Is the problem that the necessary initializations take too long, or that they take too long when run serially under a single thread ?
Re: [Cooker] Boot-up speed
On Wednesday 29 Oct 2003 07:08, Chris Picton wrote: > I used the program 'serel' from http://www.fastboot.org for a while > during the 9.1 -> 9.2 cooker phase, and had no problems with it. Bootup > was faster. The program has not been updated in a while, but seems to > be fairly complete as it stands. I had to manually add a few service > dependancies (like harddrake and dm), as it is currently tuned for > redhat systems. I did some testing myself, working on something similar, though I still don't know what people prefer: that I hack init, or replace /etc/rc.d/rc by a binary... http://www.stack.nl/~josh/Mandrake Jos
Re: [Cooker] Boot-up speed
--- John Allen <[EMAIL PROTECTED]> wrote: > Well I was on dial up (DSL now), but I have a lan, > and the server shares the > inet connection, and I need NFS enabled by default. > > Perhaps the installer/control panel should ask > whether you want NFS > (server/client), or not at all. Yes, but you had both a modem AND a network card. My idea was to disable lan services (nfs, samba, tmdns) if no network card is detected. __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
On Tuesday 28 October 2003 10:52 pm, Vedran Ljubovic wrote: > --- [EMAIL PROTECTED] wrote: > > > don't even own a network card. Therefore I > > > > probably > > > > > don't need sshd at all. > > > > Isn't xinetd startup of ssh disabled by default, in > > favour of starting as > > a daemon? > > Hm you're right. It seems that xinetd is used for fam. > > > > - While I'm at it, I also don't need netfs, > > > > nfslock > > > > > (thus portmap) > > > > If you want to use NFS you need them all. If you > > don't use NFS, 'urpme > > nfs-utils-client' or whatever it is. AFAIK you need > > portmap for fam to > > work? > > Ok, I'll remove nfs. No use for it over dialup. Btw do > you think that it would cause too much trouble if nfs > is not enabled by default, atleast for us dialup > users? > Well I was on dial up (DSL now), but I have a lan, and the server shares the inet connection, and I need NFS enabled by default. Perhaps the installer/control panel should ask whether you want NFS (server/client), or not at all. -- John Allen, mailto:[EMAIL PROTECTED] MandrakeClub Silver Member. http://allentech.homelinux.org/
Re: [Cooker] Boot-up speed
On Wed, 2003-10-29 at 01:41, Han Boetes wrote: > Frederic Conrotte wrote: > > Since I switched to Linux with MDK8.2, I've always noticed the XFS > > daemon (X Font Server) and the DM daemon (Display Manager) are not > > started among the first services to be started. DM need XFS started > > before it can be launched. > > > > DM is the services that launch KDE and therefore the sooner it is > > started the sooner you will get KDE up. > > > > For example, for MDK9.2, I moved XFS and DM to S01xfs and S02dm in > > /etc/rc.d/rc5.d/: > > > > S01xfs > > S02dm > > S03iptables > > S09network I used the program 'serel' from http://www.fastboot.org for a while during the 9.1 -> 9.2 cooker phase, and had no problems with it. Bootup was faster. The program has not been updated in a while, but seems to be fairly complete as it stands. I had to manually add a few service dependancies (like harddrake and dm), as it is currently tuned for redhat systems. Possibly somebody could try packaging it for cooker, as it does seem to be a useful piece of software. Chris -- -+-- Chris Picton | PGP Key ID: 9D28A988 (wwwkeys.pgp.net) Solutions Developer | PGP Key Fingerprint: Tangent Systems | 2B46 29EA D530 79EC D9EA 3ED0 229D 6DD6 9D28 A988 [EMAIL PROTECTED] | http://www.tangent.co.za/keys/chrisp.asc -+-- signature.asc Description: This is a digitally signed message part
Re: [Cooker] Boot-up speed
Hello, I read that there is a new feature in the kernel2.6 called futexes. It's seems to allow fast context switching... I hope it helps you. (I really don't know) Fabrice. Selon Han Boetes <[EMAIL PROTECTED]>: > Frederic Conrotte wrote: > > Since I switched to Linux with MDK8.2, I've always noticed the XFS > > daemon (X Font Server) and the DM daemon (Display Manager) are not > > started among the first services to be started. DM need XFS started > > before it can be launched. > > > > DM is the services that launch KDE and therefore the sooner it is > > started the sooner you will get KDE up. > > > > For example, for MDK9.2, I moved XFS and DM to S01xfs and S02dm in > > /etc/rc.d/rc5.d/: > > > > S01xfs > > S02dm > > S03iptables > > S09network > > Yes but dm needs networking at the least. This solution may work for > you but mandrake has to have sane defaults for everyone. Which is the > general problem with suggestions. Most people only take their own > preferences and situation in account. > > > > # Han > -- > http://www.xs4all.nl/~hanb/software > http://www.xs4all.nl/~hanb/documents/quotingguide.html > >
Re: [Cooker] Boot-up speed
Frederic Conrotte wrote: > Since I switched to Linux with MDK8.2, I've always noticed the XFS > daemon (X Font Server) and the DM daemon (Display Manager) are not > started among the first services to be started. DM need XFS started > before it can be launched. > > DM is the services that launch KDE and therefore the sooner it is > started the sooner you will get KDE up. > > For example, for MDK9.2, I moved XFS and DM to S01xfs and S02dm in > /etc/rc.d/rc5.d/: > > S01xfs > S02dm > S03iptables > S09network Yes but dm needs networking at the least. This solution may work for you but mandrake has to have sane defaults for everyone. Which is the general problem with suggestions. Most people only take their own preferences and situation in account. # Han -- http://www.xs4all.nl/~hanb/software http://www.xs4all.nl/~hanb/documents/quotingguide.html
Re: [Cooker] Boot-up speed
--- [EMAIL PROTECTED] wrote: > > don't even own a network card. Therefore I > probably > > don't need sshd at all. > > Isn't xinetd startup of ssh disabled by default, in > favour of starting as > a daemon? Hm you're right. It seems that xinetd is used for fam. > > - While I'm at it, I also don't need netfs, > nfslock > > (thus portmap) > > If you want to use NFS you need them all. If you > don't use NFS, 'urpme > nfs-utils-client' or whatever it is. AFAIK you need > portmap for fam to > work? Ok, I'll remove nfs. No use for it over dialup. Btw do you think that it would cause too much trouble if nfs is not enabled by default, atleast for us dialup users? > > or tmdns. Isn't it possible to disable > > these services if no NIC is detected during > > installation? Also since this is a desktop pc i > don't > > need apmd. > > And you don't want to be able to use the sleep/wake > buttons on your > keyboard ;-). I don't have them ;) On one keyboard i had the misfortune to work with, i wanted to pry them out cause i kept hitting shutdown instead of page down. But you're right, there are uses for apmd on desktop. Then the description needs to state it more clearly, cause many would disable it. > The > > "loading session" part in kde can be accellerated > by > > disabling KOrganizer (if someone wants to use it, > > he/she can find it in the menus) > > I don't think korganiser has a huge impact, and it's > point is to remind > you, not to tell you the things you want to be > reminded of when you > remember to run it, and it can easily be disabled. Well it's the only thing in the session? It has nothing to remind me of until i enter some tasks. And when i do it will automatically add itself to systray, so when i log off the session is saved and next time i boot it will be brought up. > Yay, then Mandrake samba print servers will now only > be usable 2 minutes > after having fully booted ... since we HUP samba a > minute after starting > it to give CUPS time to pick up all the printers (I > haven't removed this > since samba-2.2.8a can't update the printers > automatically as it is > supposed to - it worked in 2.2.7a though ...). obviuosly boot needs to be tweaked considering the planned use for the machine and other features (such as having sleep button on your keyboard ;) Which is exactly what I'm doing here. I'm tweaking my home machine and so i asked for some advice. > BTW, Windows (even Windows 2000) shuts down very > quickly (ie usually < 5 > seconds on my laptop), and I think the margin is > greater on shutdown ... I guess one can wait 15-20 sec for anything, but not much more. Thanks for your feedback! I appreciate it. __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
Just my 2 cents... since I switched to Linux with MDK8.2, I've always noticed the XFS daemon (X Font Server) and the DM daemon (Display Manager) are not started among the first services to be started. DM need XFS started before it can be launched. DM is the services that launch KDE and therefore the sooner it is started the sooner you will get KDE up. For example, for MDK9.2, I moved XFS and DM to S01xfs and S02dm in /etc/rc.d/rc5.d/: [EMAIL PROTECTED] rc5.d]$ ll total 4 lrwxrwxrwx1 root root 20 oct 21 01:40 K15postgresql -> ../init.d/postgresql* lrwxrwxrwx1 root root 16 oct 21 01:40 K55routed -> ../init.d/routed* lrwxrwxrwx1 root root 13 oct 21 01:32 S01xfs -> ../init.d/xfs* lrwxrwxrwx1 root root 12 oct 21 01:31 S02dm -> ../init.d/dm* lrwxrwxrwx1 root root 18 oct 21 01:34 S03iptables -> ../init.d/iptables* lrwxrwxrwx1 root root 17 oct 21 01:42 S09network -> ../init.d/network* lrwxrwxrwx1 root root 18 oct 21 01:46 S11internet -> ../init.d/internet* lrwxrwxrwx1 root root 17 oct 21 01:42 S11portmap -> ../init.d/portmap* lrwxrwxrwx1 root root 16 oct 21 01:42 S12syslog -> ../init.d/syslog* lrwxrwxrwx1 root root 17 oct 21 01:31 S13partmon -> ../init.d/partmon* lrwxrwxrwx1 root root 17 oct 21 01:34 S14nfslock -> ../init.d/nfslock* -rwxr--r--1 root root 721 oct 21 19:02 S15fw* lrwxrwxrwx1 root root 14 oct 21 01:31 S17alsa -> ../init.d/alsa* lrwxrwxrwx1 root root 15 oct 21 01:42 S18sound -> ../init.d/sound* . . . . it saves me 5 seconds of total boot time maybe it should be setup like that for MDK10.0 ? regards, effco Le mar 28/10/2003 à 21:26, [EMAIL PROTECTED] a écrit : > > Hi, > > > > > - Sysv is already mostly parallelized. It takes 26 > > seconds which is better than I expected. Actually, as > > I understand, on this single-cpu machine the overhead > > of context switching is so high that xinetd and cups > > would probably start faster if they were not > > parallelized with X. > > - Idea: sshd is the only service that uses xinetd. > > Standalone sshd would probably boot faster? This > > computer connects to the Internet using a modem, I > > don't even own a network card. Therefore I probably > > don't need sshd at all. > > Isn't xinetd startup of ssh disabled by default, in favour of starting as > a daemon? > > > - While I'm at it, I also don't need netfs, nfslock > > (thus portmap) > > If you want to use NFS you need them all. If you don't use NFS, 'urpme > nfs-utils-client' or whatever it is. AFAIK you need portmap for fam to > work? > > > or tmdns. Isn't it possible to disable > > these services if no NIC is detected during > > installation? Also since this is a desktop pc i don't > > need apmd. > > And you don't want to be able to use the sleep/wake buttons on your > keyboard ;-). > > > Can't this also be detected? > > - I just can't understand why kde takes so long to > > start!??!? Unfortunately, gnome isn't much better. The > > "loading session" part in kde can be accellerated by > > disabling KOrganizer (if someone wants to use it, > > he/she can find it in the menus) > > I don't think korganiser has a huge impact, and it's point is to remind > you, not to tell you the things you want to be reminded of when you > remember to run it, and it can easily be disabled. > > > - Idea: postpone cups and scannerdrake startup for 60 > > seconds to give desktop a chance to start and avoid > > unnecessary context switching. > > Yay, then Mandrake samba print servers will now only be usable 2 minutes > after having fully booted ... since we HUP samba a minute after starting > it to give CUPS time to pick up all the printers (I haven't removed this > since samba-2.2.8a can't update the printers automatically as it is > supposed to - it worked in 2.2.7a though ...). > > > It's very unlikely that > > someone would like to print or scan something in the > > first 60 seconds? > > Maybe on a desktop, and I guess it's not too bad on a server unless > there's an MCSE driving it ... > > > The same trick could be used for > > crond and atd I believe. > > > > What do you think? Should I apply these optimizations > > to my Pc? will anything burn up (except the lg cd-rom) > > ;) > > I wonder if the 23s to start KDE can be reduced significantly by pre-linking? > > BTW, Windows (even Windows 2000) shuts down very quickly (ie usually < 5 > seconds on my laptop), and I think the margin is greater on shutdown ... > > Regards, > Buchan > > > > > >
Re: [Cooker] Boot-up speed
> Hi, > > - Sysv is already mostly parallelized. It takes 26 > seconds which is better than I expected. Actually, as > I understand, on this single-cpu machine the overhead > of context switching is so high that xinetd and cups > would probably start faster if they were not > parallelized with X. > - Idea: sshd is the only service that uses xinetd. > Standalone sshd would probably boot faster? This > computer connects to the Internet using a modem, I > don't even own a network card. Therefore I probably > don't need sshd at all. Isn't xinetd startup of ssh disabled by default, in favour of starting as a daemon? > - While I'm at it, I also don't need netfs, nfslock > (thus portmap) If you want to use NFS you need them all. If you don't use NFS, 'urpme nfs-utils-client' or whatever it is. AFAIK you need portmap for fam to work? > or tmdns. Isn't it possible to disable > these services if no NIC is detected during > installation? Also since this is a desktop pc i don't > need apmd. And you don't want to be able to use the sleep/wake buttons on your keyboard ;-). > Can't this also be detected? > - I just can't understand why kde takes so long to > start!??!? Unfortunately, gnome isn't much better. The > "loading session" part in kde can be accellerated by > disabling KOrganizer (if someone wants to use it, > he/she can find it in the menus) I don't think korganiser has a huge impact, and it's point is to remind you, not to tell you the things you want to be reminded of when you remember to run it, and it can easily be disabled. > - Idea: postpone cups and scannerdrake startup for 60 > seconds to give desktop a chance to start and avoid > unnecessary context switching. Yay, then Mandrake samba print servers will now only be usable 2 minutes after having fully booted ... since we HUP samba a minute after starting it to give CUPS time to pick up all the printers (I haven't removed this since samba-2.2.8a can't update the printers automatically as it is supposed to - it worked in 2.2.7a though ...). > It's very unlikely that > someone would like to print or scan something in the > first 60 seconds? Maybe on a desktop, and I guess it's not too bad on a server unless there's an MCSE driving it ... > The same trick could be used for > crond and atd I believe. > > What do you think? Should I apply these optimizations > to my Pc? will anything burn up (except the lg cd-rom) > ;) I wonder if the 23s to start KDE can be reduced significantly by pre-linking? BTW, Windows (even Windows 2000) shuts down very quickly (ie usually < 5 seconds on my laptop), and I think the margin is greater on shutdown ... Regards, Buchan
Re: [Cooker] Boot-up speed
--- Michael Scherer <[EMAIL PROTECTED]> wrote: > > - Do we need to run depmod if no new modules are > > installed (many ways to check)? > > i think it use -A switch and so is runned only if > it should. Well if there are new modules it takes almost 30 seconds, without them it's 4 seconds consistently. I did a few runs and each time it isn't any faster. Go figure. > > - I just can't understand why kde takes so long to > > start!??!? > > This has to do with the disk. a faster disk help moe > than a faster > processor. > Maybe using -Os could help, but recompiling to test > is maybe too much. At work it starts in less then 10 seconds, but there i don't use auto-login. This shows that context switching can really canibalize performance. Ah well... let's just wait for the 2.6.0 kernel. > > - Idea: postpone cups and scannerdrake startup for > 60 > > seconds to give desktop a chance to start and > avoid > > unnecessary context switching. It's very unlikely > that > > someone would like to print or scan something in > the > > first 60 seconds? The same trick could be used for > > crond and atd I believe. > > So, the first 60 secondes would be as unusable as on > windows ? omg! whole 60 seconds! i can't print this file for full 60 seconds, what will i do!??!?!? well i can barely find a file, open it and click on print (using a gui) in less then 60 seconds. I can do it faster in console, but that is runlevel 3 so the above must not necessarily apply __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/
Re: [Cooker] Boot-up speed
On Tuesday 28 October 2003 20:14, Vedran Ljubovic wrote: > Hi, > > > Recently there were some articles on Slashdot > proposing alternative boot procedures. Most of them > advocate parallelizing of SysV-init tasks. Motivated > by these I've decided to make a detailed analysis of > boot procedure and see how can I make it faster. > > I have an oldish PC that can run most modern software > fast enough. This is a vanilla Mdk 9.2 install and I > use the auto-login feature with KDE. Complete time > from pressing enter in lilo to a usable desktop is > exactly 81 seconds. On this same computer Win98 is > usable in around 35 seconds (I did some tweaking > though). > > Here is a detailed account of the boot procedure. I > made it using logs and a stopwatch. It should clearly > point the likely targets for optimization. > > [ skip ] > - Do we need to run depmod if no new modules are > installed (many ways to check)? i think it use -A switch and so is runned only if it should. > - I just can't understand why kde takes so long to > start!??!? This has to do with the disk. a faster disk help moe than a faster processor. Maybe using -Os could help, but recompiling to test is maybe too much. > - Idea: postpone cups and scannerdrake startup for 60 > seconds to give desktop a chance to start and avoid > unnecessary context switching. It's very unlikely that > someone would like to print or scan something in the > first 60 seconds? The same trick could be used for > crond and atd I believe. So, the first 60 secondes would be as unusable as on windows ? -- Michaël Scherer
[Cooker] Boot-up speed
Hi, Recently there were some articles on Slashdot proposing alternative boot procedures. Most of them advocate parallelizing of SysV-init tasks. Motivated by these I've decided to make a detailed analysis of boot procedure and see how can I make it faster. I have an oldish PC that can run most modern software fast enough. This is a vanilla Mdk 9.2 install and I use the auto-login feature with KDE. Complete time from pressing enter in lilo to a usable desktop is exactly 81 seconds. On this same computer Win98 is usable in around 35 seconds (I did some tweaking though). Here is a detailed account of the boot procedure. I made it using logs and a stopwatch. It should clearly point the likely targets for optimization. kernel loading into memory ~2s screen is blank ~2s kernel boots, mounts initrd, performs various configuration tasks ~5s rc.sysinit kicks in, some config tasks ~1s running devfsd 3s unmounting initrd, applying sysctl.conf 1s setting up date & time, loading keymap, setting hostname~1s starting USB service** *9s fsck1s root is remounted rw, run swapon<1s execute depmod -A 4s a bunch of config tasks are completed 4s run mandrake_everytime service 3s switching to runlevel 5 <1s harddrake service looks for new hardware and saves configuration 4s services: network, portmap 2s services: syslog, partmon, nfslock, sound, random 3s xfs *1s netfs <1s apmd*2s dm service goes to background, screen goes blank1s services: atd, tmdns, rawdevices; kdm_config is being read1s xinetd; screen still blank 4s cups; blue screen with a cursor is shown8s services: keytable, numlock, internet 1s crond 1s kheader <1s devfsd actions 1s service "local" sets the console screens<1s KDE is starting up (splash screen)*** 23s KDE session is restored 5s * proces is just spawned to the background; these figures should not be included when summarizing ** I have a USB scanner so it is detected and the appropriate config tool (scannerdrake) is started *** at ~11th second scannerdrake is run for the second time total: 81 seconds Some conclusions: - Overall time to boot a minimal usable console is 27 seconds which is IMHO too much. The biggest problem is kernel. Unfortunately there seems to be no tool that can show a log of kernel activities with times (please correct me), but it seems that most of the time was spent initializing my two hard disks (more than a second each). A recompile should solve remaining problems. - Why is devfsd taking so long? - Do we need to run depmod if no new modules are installed (many ways to check)? - Sysv is already mostly parallelized. It takes 26 seconds which is better than I expected. Actually, as I understand, on this single-cpu machine the overhead of context switching is so high that xinetd and cups would probably start faster if they were not parallelized with X. - Idea: sshd is the only service that uses xinetd. Standalone sshd would probably boot faster? This computer connects to the Internet using a modem, I don't even own a network card. Therefore I probably don't need sshd at all. - While I'm at it, I also don't need netfs, nfslock (thus portmap) or tmdns. Isn't it possible to disable these services if no NIC is detected during installation? Also since this is a desktop pc i don't need apmd. Can't this also be detected? - I just can't understand why kde takes so long to start!??!? Unfortunately, gnome isn't much better. The "loading session" part in kde can be accellerated by disabling KOrganizer (if someone wants to use it, he/she can find it in the menus) - Idea: postpone cups and scannerdrake startup for 60 seconds to give desktop a chance to start and avoid unnecessary context switching. It's very unlikely that someone would like to print or scan something in the first 60 seconds? The same trick could be used for crond and atd I believe. What do you think? Should I apply these optimizations to my Pc? will anything burn up (except the lg cd-rom) ;) __ Do you Yahoo!? Exclusive Video Premiere - Britney Spears http://launch.yahoo.com/promos/britneyspears/