Slashdot Log In
How Kernel Hackers Boosted the Speed of Desktop Linux
Posted by
timothy
on Thu Oct 02, 2008 03:17 PM
from the mild-claims dept.
from the mild-claims dept.
chromatic writes "Kernel hackers Arjan van de Ven and Auke Kok showed off Linux booting in five seconds at last month's Linux Plumbers Conference. Arjan and other hackers have already improved the Linux user experience by reducing power consumption and latency. O'Reilly News interviewed him about his work on improving the Linux experience with PowerTOP, LatencyTOP, and Five-Second Boot."
Related Stories
[+]
Reducing Boot Time On a General Linux Distro 354 comments
Linzer writes "In this blog entry, Fred Crozat (head of Mandriva's engineering team in France) explains in great detail how his team has been detecting and getting rid of bottlenecks in the boot process, from the early stages to loading the desktop environment, thus decreasing overall boot time. An informative tour of the nuts and bolts of the boot process and how they can be tinkered with: initrd, initscripts, udev, modprobe calls. The basic tool they use for performance analysis is bootchart, which produces a map of process information and resource utilization during boot. The final trick: preloading desktop environment files while waiting for the user to type her password."
[+]
Hardware: PC Makers Try To Pinch Seconds From Their Boot Times 399 comments
Some computers are never turned off, or at least rarely see any state less active than "standby," but others (for power savings or other reasons) need rebooting — daily, or even more often. The New York Times is running a short article which says that it's not just a few makers like Asus who are trying to take away some of the pain of waiting for computers, especially laptops, to boot up. While it's always been a minor annoyance to wait while a computer slowly grinds itself to readiness, "the agitation seems more intense than in the pre-Internet days," and manufacturers are actively trying to cut that wait down to a more bearable length. How bearable? A "very good system is one that boots in under 15 seconds," according to a Microsoft blog cited, and an HP source names an 18-month goal of 20-30 seconds.
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
Should lead to possibly great advertisements (Score:5, Insightful)
Re:Should lead to possibly great advertisements (Score:5, Funny)
Kudos to you sir, for reducing the time it takes to type congratulations by instead using grats!
Parent
Re:Should lead to possibly great advertisements (Score:4, Informative)
Nope. Kudos is Greek and singular, just like pathos :)
Parent
Re:Should lead to possibly great advertisements (Score:4, Funny)
Parent
Re:Should lead to possibly great advertisements (Score:5, Funny)
Parent
Re:Should lead to possibly great advertisements (Score:4, Funny)
Parent
Re:Should lead to possibly great advertisements (Score:5, Funny)
Yeah, Safe Mode.
Parent
Get out of my lawn! (Score:4, Funny)
CP/M was probably OK but my Zilog-based PC had floppies only so it sucked too.
MS-DOS 3.0 was up and running in 1-2 seconds (assuming you had a hard drive and empty config.sys and autexec.bat).
Then MS rewrote DOS in that punky and slow new language "C" and since then everything went down. The next thing you see is that HIGHMEM.SYS driver taking your precious memory out of 640KB for the promise of semi-useless XMS memory for overlays. Oh well...
Now my kernel sits during boot on 4GB RAM looking for un-present USB devices and waiting for eth0 to figure out DHCP.
I guess Bill Gates was right and 640KB is right amount for everybody so OS would not get confused with all those amounts of bits laying around.
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
32-bit operating systems on x86 can get around this with PAE. Windows XP and Vista don't have support for it, though.
http://www.appleinsider.com/articles/08/09/03/road_to_mac_os_x_snow_leopard_64_bits_santa_rosa_and_the_great_pc_swindle.html [appleinsider.com]
The half gig "lost" is because x86 is a memory-mapped architecture. The graphics card memory and all your peripherals are mapped over the top part of your memory address space.
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
You obviously fail it.
The 32 bit limitation is 4 gigs of RAM (total).
Your video card's memory is mapped into the same address space as your system ram, but from the top of the address space downward, rather than system ram (mapped from bottom of address space upward).
The video ram has priority, thus if you have 3G of ram and a 1G video card you are using 100% of your address space. any more system memory installed will be masked by the video ram and thus is "lost".
-nB
Parent
Re:"Pretty Quick"? Drug of Choice? (Score:4, Informative)
There is no such thing as end-user-OS-boot-time. It depends a lot on device drivers and system background utilities. For example, some piece of hardware AND some release version of its driver maybe causing your trouble... especially if that hardware is removed and the driver probes a lot of time just to be sure. Same for the AV software doing weird things in order to "secure" the system *before* user interaction... At least in the hardware side, this apply for Linux too.
Parent
Re:Should lead to possibly great advertisements (Score:5, Insightful)
The only issue is that they had to cut some corners to make this work. Axing sendmail? Ok, I understand that (I think was arguing that 10 years ago -- still don't wonder why that's on by default in the desktop distributions). But "The 'done booting' time did not include bringing up the network"? Um, ok... no. With the proliferation of devices solely used to read information from networks (Netbooks, those "quick-loading" Linux apps some laptop manufacturers are including so people can check their email, etc.) accessing the network is one of the main purposes for turning on the machine in the first place. It would royally piss people off to have a quick loading screen, log in and then see "Hold up, still starting up the network". (Just as frustrating as starting a Windows or Mac, getting to the desktop and still waiting while services and programs are loaded).
Come to think of it, what people really need to do is take a good look at modern OSes and determine EXACTLY what still needs to be there and what's cruft. Some of the daemons/services we're launching made sense 15-20 years ago. Does the fax daemon really need to start on my Mac? Does the Group Policy Client need to be started on my Vista box when I'm not on a domain? There's lots of stuff that at one point probably made sense to someone but now is just extraneous.
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
actually we do bring up the network daemon (either connmand or network manager) as soon as we can, but we don't stop the entire startup process.
on my test system here it runs dhclient about when X starts up and the network card receives a reply with a few seconds from that, long before I can start firefox :)
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
A link to the article is here. [ibm.com]
Bear in mind, this is circa 2003. 8)
Parent
Re:Should lead to possibly great advertisements (Score:5, Insightful)
Sendmail's main purpose in the typical Linux desktop configuration (say, Fedora) is delivering logwatch output to root. [Logwatch attempts to distill the important stuff from system log files.]
But sendmail can be started lazily (in the background) so as not to slow the boot. Or sendmail can be replaced with a lighter weight smtp daemon. Truly though, logwatch-by-email should die for non-enterprise desktops. It's so 1980s it just hurts.
IMHO logwatch should be replaced by some kind of graphical notification widget which requires authentication to actually view the details, since they can be sensitive. As it is, I haven't read my logwatch emails in months, but if SMART is complaining about an immanent disk failure I'd *really* like to know.
Parent
Re:Should lead to possibly great advertisements (Score:4, Interesting)
Yeah, but you don't need a full ESMTP server for that - a wrapper for the local delivery agent that speaks classic SMTP (but ignores most of it) should be sufficient. In fact, if you're only using it to deliver to root, you've a choice of a tiny bit of text formatting, putting into a huge block of text and whapping it onto the end of root's mailbox, or doing a tiny bit of other text formatting and use the local mail delivery agent to do all of the work.
If you've only one login account (the rest are for daemons or accessed via sudo), then the login code is excessively heavy. There's effectively only one user and effectively only one password. Those need to be in a password/shadow file for compatibility with other apps, but for machines that are essentially single-user, where the data is essentially fixed-length, you don't need search algorithms, routines to scan for the correct column, etc. You store two fixed-length blocks of data and then do a string compare and a byte compare. No files to open, no multi-layer authentication modules, etc. For a straight single-user desktop, you don't need such weight for a console login. You do for servers and other remote activity, but not for the console.
XDM/GDM/KDM could be rigged to work under GGI or XGGI. They don't need the full X system. You can complete booting that whilst the user logs in.
Parent
Re:Should lead to possibly great advertisements (Score:5, Insightful)
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
but if this can be applied generically to most distributions then it should present an excellent opportunity for advertisement.
Not going to happen. If you read the article, you'll see that they compiled all drivers directly into the kernel, so it is essentially an embedded device now. Also consider the fact that they are using a SSD, which is going to decrease boot times regardless of any boot-process improvements.
So basically, you could never apply these speed increases to a generic distro.
Parent
Re:Should lead to possibly great advertisements (Score:5, Insightful)
... but an OEM can. (If they were so inclined.)
Parent
Re:Should lead to possibly great advertisements (Score:5, Interesting)
says who?
We talked with both the fedora and ubuntu developers at the LPC and even they agreed that a LOT more drivers should be compiled into the kernel instead of being modules (c'mon, ext3 as a module? really?).
99% of what we did to make this work in 5 seconds applies straight for generic laptops and even most people's desktop sytems.
The speedups _still_ are relevant with generic spinning media too. Maybe those are not as fast as SSD's, but the principle is still the same (IOW, for instance reading data in the order that you need it, is better than reading it in the order that it is scattered across the hard disk)
speeding up the kernel to boot in 1 second is TOTALLY applicable to generic distros (not only that, it's relatively easy and we basically already did that).
speeding up X startup to be 1.5s is TOTALLY applicable to generic distros.
Parent
Re:Should lead to possibly great advertisements (Score:5, Interesting)
Why not?
How about the first time the system boots, it profiles what drivers are installed. It then recompiles the kernel to include those drivers.
On subsequent reboots, it uses the recompiled kernel and then, once the system is up and running, check to see if something that is compiled in is no longer needed, and see if something has been added that should be compiled in.
I'm sure it's not *that* easy, but maybe an idea for the future?
Parent
Re:Should lead to possibly great advertisements (Score:5, Interesting)
So basically, you could never apply these speed increases to a generic distro.
Oh come now! Never say never!
You could:
- Boot with modular kernel.
- Probe devices and get a list of loaded modules.
- Recompile kernel with said modules built-in.
- Boot with that kernel from now on.
It's relatively scriptable - in fact, I think there's a "probe loaded modules and generate new .config" script already about the place. If the user is unwilling to wait for a kernel recompile during install, just stick with the modular kernel and incrementally compile during idle time.
It's trivial. I'd code it up myself, but I'm a little busy at the moment, you understand.
Parent
Re:Should lead to possibly great advertisements (Score:5, Informative)
we don't even need to recompile the kernel, we can just link the modules in as needed. Something most older UNIX systems actually did.
Parent
Re:Should lead to possibly great advertisements (Score:5, Funny)
Ah, the old, "I have discovered a truly marvelous method to improve Linux which the margins of my free time are too narrow to allow me to code."
Still keep writing.
Parent
Re:Should lead to possibly great advertisements (Score:5, Interesting)
Once the distro installer has finished it would attempt to boot the system to the graphical login. If the login screen came up it would save the state of the machine to a fast loading RAM image that GRUB could directly inject to RAM.
Reading ~100MB of system should take seconds on any machine, and the code area taken up by the GRUB routine could be overwritten with a memory offset command embedded in the first few bytes of the image.
Once the image is in RAM the execution starts up again immediately waiting for your login details.
Of course, hardware would have to be hashed to make sure that the image was still compatible with the machine and that the disk hadn't been moved to a different one. Upgrading the hardware or the kernel, software updates et cetera would require the image to be resaved, but those are easily achieved.
Taking into account the size of the image, I guess that someone could code the installer to compile the kernel with the modules the system uses built in. Maybe as a function of the exit procedure. "Optimise load time - warning! This will take quite some time"
Basically I guess what I'm saying is something like a hibernate file, but one that is rarely changed and only contains the system, not the applications running in a session.
Parent
Does it matter? (Score:4, Insightful)
I see a lot of comments on the LWN article of people talking about starting services after the user sees the desktop as cheating. However, I ask, does this really all matter. I'm not sure how everyone else uses their computer but I only need to boot my Linux machine about once every 30-60 days. I don't need to dual boot like I did back in say 2002 and comparitively, the amount of time it takes for Linux and X to start up are practically irrelivent. I can imagine laptop users may feel much differently about this, but I thought that was the point of being able to suspend/hibernate.
One thing that worries me is that a focus on ensuring a quick boot at the expense of a potentially less stable system is not a good thing. Fortunately however quick booting is not something that Linux requires, its something that distributions can decide to do or not, which is one of the strengths of the open source/Linux way.
Re:Does it matter? (Score:4, Informative)
Yes.
Parent
Another one word answer. (Score:3)
Why?
Re:Does it matter? (Score:4, Funny)
I'm not sure how everyone else uses their computer but I only need to boot my Linux machine about once every 30-60 days.
Some people like to power down their computers to stop them from wasting energy at night.
How they sleep without the sweet, sweet sound of fans running though leaves me dumbstruck.
Parent
Re: (Score:3, Informative)
I'm one of the less lucky ones that only have half of hibernation working, the "resume" part fails.
Re:Does it matter? (Score:5, Funny)
But at least someone, somewhere can boot linux in 5 seconds.
Parent
Re:Does it matter? (Score:5, Insightful)
Being able to boot in NN seconds isn't so impressive when you look at the incompatibilities it creates.
On my networks, the servers connect to the DHCP server and get not only an IP back, but also the name of NIS servers, who in turn returns (among other things) autofs maps which are used to mount the home directories as well as providing login authentication. The xdm login window returns a list of currently available X servers.
In other words, there are reasons why things run in the order they run, and any deviation will cause things to stop working.
Improving things are fine, but not when it's at the expense of current and well-known functionality.
Parent
Re:Does it matter? (Score:5, Insightful)
On my networks, the servers connect to the DHCP server and get not only an IP back, but also the name of NIS servers, who in turn returns (among other things) autofs maps which are used to mount the home directories as well as providing login authentication.
What you describe is similar to what Windows calls "domain authentication". Not every computer logs on to a domain, especially in the home or home office environment where a fast boot is paramount.
The xdm login window returns a list of currently available X servers.
Then have it refresh the list whenever a network interface comes up.
Parent
Re:Does it matter? (Score:4, Insightful)
so you are saying that you would rather stare at a hung boot in text mode instead of having the possibility of working in offline mode in X?
that does not make sense at all :)
for network-client setups like you describe, we should still start X immediately and if the network fails or is slow, at least provide some interaction with the system (work offline, nudge network with login attempt etc).
Parent
Re:Does it matter? (Score:4, Informative)
Linux has always tried to be an excellent choice for a very broad range of uses. Just because you do not need fast boots does not nobody else does.
Parent
TFA (Score:5, Interesting)
And no cheating. "Done booting means CPU and disk idle," Arjan said. No fair putting up the desktop while still starting services behind the scenes. (An audience member pointed out that Microsoft does this.) The "done booting" time did not include bringing up the network, but did include starting NetworkManager.
It seems to me that the five seconds could concievably be brought down to virtually zero with cheating! My work PC slows down so much sometimes from antivirus, inventory controls, etc that it takes longer than that to add a record or open a table in an Access database. With a keyboard buffer you could stick a fake desktop and login in, and have the real desktop and login take over before the user finished typing in his password.
Re: (Score:3, Funny)
Re:TFA (Score:5, Interesting)
Yes, please. More common users would benefit from this greatly. I would love a prompt for user and password while it is booting, because most users do not leave their computer all day, especially laptop and handheld/UMPC users.
With desktop computers, I really think that the time it takes to boot into Ubuntu, THEN going into a Gnome prompt, THEN loading the services and desktop is a silly idea.
Parent
On an old Pentium III laptop... (Score:5, Funny)
My stepfather still has an old Pentium III laptop with Windows 95 running on it. Booting the laptop to read an E-mail takes around 20 minutes. His advice to anyone who wants to use it, "switch on the PC, do something else like have a bath, do the lawn, read the newspaper, or have a coffee, and the PC will be ready to use before you know it".
In case Oreilly goes down again... (Score:4, Informative)
O'Reilly News recently interviewed Arjan van de Ven [fenrus.org] about his efforts to improve Linux performance and reduce power consumption. Arjan works for Intel in the Open Source Technology Center [intel.com]. This interview is approximately 30 minutes.
One of the projects you're probably most known for in the past couple of years is the PowerTOP [lesswatts.org] utility, which I found very fascinating. Looking at some of the gains you've made over the past 18 months, it seems like Linux-based devices are saving a lot more power than they used to. What do you consider the big successes in the past year and a half?
To be honest we fixed effectively the entire Linux desktop space. It's not--PowerTOP is more--it's not just what we fixed with PowerTOP is not individual pieces. We fixed everything. For me that was a success.
Is that everything in terms of not just desktop but servers as well?
Yeah; we fixed not just Evolution. We fixed Firefox; the thing with Firefox was that it wasn't one thing that was broken. Everything had problems and we had to fix all of it. So for me the success was how quickly everything got fixed; it was just amazing.
In this context you consider fixed--everything is no longer broken in the same way or--?
Everything is no longer keeping the CPU out of idle basically.
Do you have a reference machine? I guess I'm asking what's your benchmark for this, a particular software configuration stack or particular type of machine, or are you willing to say it's pretty much every Linux based machine out there?
I'm looking at several machines--my own laptop but to be honest, what runs on my own laptop is what I care about most. At least that's where I got more battery life, this is where I see the changes. I tend to run a quite rich environment on my laptop but I also look at service. We look at all kinds of machines and we see the same trend everywhere in that all the various pieces of it--never polling or keeping the CPU up. They all got fixed.
In fixing this, is there a component of education, for example, saying "Instead of doing a busy wait on a select loop or continually polling you should set a kernel timer and wait for that to call you"?
That's part of it but the biggest thing is that you had no visibility. Just two days ago at IDF I spoke with a developer of the GNOME desktop and he said, yeah; when I saw it happen I fixed it in 10 minutes, but you don't know it's there until you see it from PowerTOP. Adding the visibility turns out to be enough for people to start fixing it. They know how to fix--how to not poll most of the time.
You can't fix something you can't measure.
If you don't see that it happens you don't know it happens and you can't fix it.
Are you getting the same sort of results from other projects you run into?
GNOME was there but it's almost everybody goes oh yeah; we should have not done that; either they fix it themselves or some--a lot of people give them the fix and in general it's like oh yeah; we shouldn't have done that. Unless you see what's happening you don't know what to fix, so the biggest thing that PowerTOP did was add visibility. We can see under the hood what's going on and then we can fix it. And quite often the fix is very simple.
It sounds then, maybe I should be able to say that just about everybody is happy to see this. Is that the case?
Yes; people--all the developers I've worked with--and that's quite a few--they all go oh yeah. Thank you for the fix; we should have no problems in the first place. We didn't know this; it's fixed now. In the beginning I did most of the fixing when PowerTOP was very new and now days the people do it themselves. The developers learn
Let's head some comments off (Score:5, Interesting)
This is effectively related to an earlier Slashdot story about the changes Mandriva are making to speed up boot on their distro [slashdot.org]
In an attempt to head off the inevitable here's a link straight to the existing
Interesting but how useful, really? thread (Yes! No! I have a Mac! I use suspend! I use hibernate! Suspend is broken for me! Hibernate is broken for me! Hibernate takes too long with 500Mbytes! Why do Linux people always say change your habits? Etc.)
What I really want to know is what can be done about usb-storage and pciehp (PCI Express hotplug). I have an EeePC 900 using a kernel with Arjan's fastboot patches [lkml.org] and with USB entirely disabled and pciehp turned off the kernel mounts the root filesystem in just over one second. With USB on and pciehp in use it's over 5 seconds....
Finally here's a link to Arjan's slides from the presentation about 5 second boot in PowerPoint format [fenrus.org] and a YouTube video of the 5 second boot on an EeePC 901 [youtube.com].
Re:Let's head some comments off (Score:5, Informative)
We've sent a patch to Greg KH making USB initialization go in parallel which reduces usb initialization from [N * 0.1] seconds (where N is the number of usb ports in your system to [0.1]. This patch is currently in linux-next afaik.
I'm wondering why you would even have PCIe HP turned on on an asus 900 :)
Parent
Good for "appliance boxes" (Score:5, Interesting)
I'd love for my MythTV box to boot faster. Since it's not silent (though the TV fans are louder, the TV isn't always on either), I leave it turned off, and the long boot time makes it less appliance-like.
Important details I'm not seeing (Score:5, Interesting)
Re:Important details I'm not seeing (Score:5, Informative)
The 5-second demo video is of an asus eee pc 901, however the same image boots on an acer aspire 1 also in exactly 5 seconds.
These systems have 512mb of memory, a slow consumer-grade SSD (we used a 4gb partition) and a 1.6GHz atom processor (single core, HT enabled).
this is a _slow_ system compared to any desktop-grade system currently on the shelves.
Parent
Re:Important details I'm not seeing (Score:4, Informative)
and we used the 'stock' SSDs in these systems. from the bootcharts you can see that our SSDs top out at 25-30mb/sec throughput on read, far below what most regular hard discs or server-grade ssd's can do.
Parent
Compilers Suck (Score:4, Interesting)
The real lesson from all this load time buisness is that our compilers still really really suck. I mean the truth is that when you boot your computer there is only a tiny bit of logic that really needs to go on because only a small amount of stuff changes between any two boots (and less between a boot and a power off).
A truly well desgined system wouldn't care about arbitrary boundaries between this program and that one, it would hunt down optimization opportunities everywhere and automatically reduce boot up to an extremely lean and quick procedure without adopting the harms of merely loading an old image.
I mean to take one example a substantial amount of time during start up is probably spent searching for and then parsing configuration files. So long as their is no cross cutting OS level JIT compiler that can deal with both system IO code and application code there isn't much we can do about this without massive investment of effort. However, in principle there is no reason that the system couldn't simply read the preparsed data from a cache and jump directly to the real substantive logic that needs to be done on boot (checking out network conditions, looking for changed hardware, dealing with changed configs)
Re:Yeah but... (Score:5, Interesting)
ROM was a wonderful thing. Simply flip the switch and the software is already loaded into memory. There was about a second or two of initialization (on a ~1MHz 8-bit processor!) and you were ready to go. It's still possible to create such fast boot times using ROM. Especially with re-flashable ROM. These sorts of boot times are seen in systems like Game Consoles.
Unfortunately, desktop OSes are so complex that using re-flashable ROM adds a great deal of complexity and cost to the design. Thus you aren't likely to see any systems keep their OS in Flash. Compounding the problem is that modern OSes are rarely designed to boot from a ROM configuration and would require substantial changes to boot properly.
Parent
Re:Boots in 5 seconds... (Score:5, Informative)
it's just a normal linux distro. for the demo we used an auto-login but you can have it start to a password screen as well, so it's just as secure as your base distro is....
Parent
Kernel modesetting is slow arriving (Score:4, Informative)
Kernel modesetting is kinda here for Intel and ATI graphics cards [phoronix.com] (in Fedora) but it's still stabilizing.
So the answer to your first question is: soon you will be able to set the final video mode in the kernel. As for your second question, doing it in X is not the best solution (as doing it in the kernel means less flickering when X starts, the ability to support graphical kernel panics and nicer virtual terminal switching).
Parent