How fast is Linux really?

chaosgeisterchen · 2007-03-07 18:54:52

Good evening,

in a rather biased discussion, a user repeatedly stated, that the Kernel architecture of FreeBSD (and hence Mac OS X) and Linux have deficits and are therefore not able to get the most out of the hardware resources. He proclaimed, that the Mac kernel is only able to do the work at a significantly lower performance as Windows XP does and Linux is about 20 to 30% faster.

This made me think. In fact, a lot of applications run a lot faster under Windows, compared to Linux. Especially, if you got Windows XP and cut it down to minimal requirements _and maintain it properly, it seems to be lightning fast, at least a lot faster than Linux seems to me. I use rather new age hardware (notebook with CentrinoDuo 1,83 and desktop with Core2Duo E6300) but Linux was not really fast on either. It is my fault or is it the truth, that the Windows kernel is able to do processing faster?

A problem with the threading under FreeBSD was mentioned and this link mentioned as a reference.

I would like to get to know your opinions on this topic.

Thanks in advance.

cg

skymt · 2007-03-07 19:22:40

One small correction: Mac OS X doesn't use the FreeBSD kernel. Instead, it uses XNU, a variant of Mach (a microkernel notorious for poor performance). XNU uses some components from the BSD kernel, but many of the bits that affect performance (threads, processes, IPC, memory management) are from Mach.

pelle.k · 2007-03-07 19:52:21

I would say the "linux $desktop" wasn't built from the ground up, to be a desktop, ie GUI OS. Linux is the kernel btw, and i assume if you compare a bare nt/vista kernel against a bare linux kernel, they would probably match each other in terms of speed. This might not be the case when comparing GUI apps responsiveness, since the nt kernel + xp on top, was in fact designed to be a solid and unified desktop OS, but a "linux $desktop" is a glue together of quite a bit of components (all with different approaches and not always intented for a desktop OS system).

Lately, there has been some standardization in the world of "desktop $linux" though, and components that form the GUI desktop become more unified and built for this specific purpose. freedesktop.org is one such approach. We're getting there

chaosgeisterchen · 2007-03-07 21:59:14

@skymt: From what I knew, Darwin emphasizes on a BSD-style kernel. I formulated it in a wrong way.

@pelle.k: What about the statement, that the X Server is actually the worst part of today's GNU/Linux environment, because it's too slow and has a flawed architecture? I am rather concerned that X has to merge with the Kernel in order to achieve a better overall performance - for instance to load X not as a layer above but instead as a kernel module, to have the GUI directly integrated into the Kernel.

What about this apporach? It is absolutely utopic or anything like this?

Dieter@be · 2007-03-07 22:31:23

I heard from a friend (who is a c/c++/objective C programmer) and uses a mac, that macosX's kernel isn't very good at memory management (when to swap in and out pages etc)

skale · 2007-03-08 03:02:40

Well, a "bare" Windows XP install is very fast, but also very useless. As soon as you start to install drivers/applications (even just network and office) it bogs down heavily.

An Arch install comparable to the XP bare install, which would have X/Icewm/some file manager/network would be a bit slower.

The Windows XP that could actually do anything, as in have office/wmp/photoshop/printer drivers/etc would boot in upwards of three minutes. Not to mention the fifteen second wait first time you try to run Photoshop, or the crawling when you try to run more than three applications at once.

Fully functional standard desktop Arch, with sshd/network/dbus/hal/cupsd/hplip/acpid/alsa/etc would boot in 30 seconds, and reach the login in just three more. Linux seems to multitask much better; to see how much it could take, I started running lots of things. I ran openoffice writer, openoffice powerpoint, firefox, gaim, gnome-network-manager, amarok, and could play thirty-one youtube videos at a time, before the videos started to lag. And that was lag, not freeze. And, this was in Ubuntu with gnome full-bloat and everything, all the toolbar thingies (like CPU Monitor) on fast refresh. XP would have crashed long before.

I am speaking from experience only, so don't take anything I say too seriously. XP could very well be 10x faster, and I am just biased out the wazoo. But I doubt it.

Can't say about Macs, I haven't used them much.

Anonymo · 2007-03-08 04:23:14

From my experience, server 2003 seems to work better than XP in the memory management and the services that are run. I installed a modified version which you can make with nlite and it works and boots faster than XP. And the compatibility is there. I think it also depends on what wm you run on linux. if you just run a wm and not a whole de, then it is much better in responsiveness. I used to run KDE, but now only can stand fluxbox. I think a trimed down 2003 can work like minimal archlinux with fluxbox. You can also run blackbox in windows which also cuts down on resources if you don't add too many plugins.

Last edited by Anonymo (2007-03-08 04:24:06)

iBertus · 2007-03-08 05:43:45

Like most things, different kernels are going to perform differently depending on the task. It's like comparing apples to oranges. You really cannot draw a proper comparison.

That said, with a proper configuration linux is almost as fast as Windows XP on this machine. With the x86_64 varient of Arch Linux the advantage is more towards the linux side.

Last edited by iBertus (2007-03-08 05:45:25)

chaosgeisterchen · 2007-03-08 06:20:02

Just to throw something in...

@skale: The WXP you talk about is hopelessly unoptimized, a fully set up xp machine can also boot up to the ready desktop in about one minute, if all these memory consuming background tasks without proper reason of existing are killed as soon as possible. But most people do not care about setting their machine up properly. Mine - for instance - gets a lot of problems if I try to strip it down. It's because of some specific drivers which make quite a hassle if you are - only slightly - trying to deactivate their functionality.

@Anonymo: W2K3 is also newer than XP and therefore better I would assume. It's built for servers and therefore is rather slim and reliable, things which XP can certainly lack sometimes.

@iBertus: XP never offered proper x86_64-support, that's true. But the performance surplus of a x86_64 distribution over a i686 distribution is next to nothing from what I experienced so far.

To generalize it, the problem seems to be KDE, which I run. But KDE ist slicker than the Windows desktop is, concerning the amount of memory consumed. So: how come? Is it programmed that badly? I would rather tend to continue to see X at fault.

sokuban · 2007-03-08 07:27:48

I don't know why but I have problems with KDE being too slow too. It was too slow for my old computer to handle. Even though I have a new computer now I ended up getting used to using xfce.

I do admit that a lot of times windows programs can feel faster than Linux programs, but I think that windows programs crash more than Linux programs do. Also Windows isn't good with old computers. My head practically explodes whenever I have to use the school library's computers. You know you have to wait a few seconds to open up IE, and while you wait for the page to load. Then when you type things in the address bar they show up a few seconds later and all that. Though this is only for old library commputers.

Whereas on the other hand, Linux with my 8 year old computer doesn't have much slowness. But I used debian with xfce. So I guess I could be cheating.

Also I think Vista obviously loses to Linux.

chaosgeisterchen · 2007-03-08 07:47:38

sokuban wrote:

Also I think Vista obviously loses to Linux.

Vista eats up a horrid amount of resources, so Linux wins the comparison hands down as long as Vista does not get more lightweight. XP is still the main competetant.

Last edited by chaosgeisterchen (2007-03-08 07:49:09)

iphitus · 2007-03-08 12:46:38

chaosgeisterchen wrote:

Good evening,
in a rather biased discussion, a user repeatedly stated, that the Kernel architecture of FreeBSD (and hence Mac OS X)

Your user needs to check his sources. OSX only uses parts of FreeBSD, and this includes the BSD userspace too. So deriving OSX as faulted because of issues in FreeBSD is a flawed argument.

and Linux have deficits and are therefore not able to get the most out of the hardware resources.

Every operating system has it's deficiencies. There's no such thing as a perfect general purpose operating system, and no such thing as a perfect specialised operating system.

He proclaimed, that the Mac kernel is only able to do the work at a significantly lower performance as Windows XP does and Linux is about 20 to 30% faster.

Proclaimed he did. These statistics should be ignored without some sort of reference. And it depends on the task, different systems are better for different tasks and workloads.

This made me think. In fact, a lot of applications run a lot faster under Windows, compared to Linux. Especially, if you got Windows XP and cut it down to minimal requirements _and maintain it properly, it seems to be lightning fast, at least a lot faster than Linux seems to me. I use rather new age hardware (notebook with CentrinoDuo 1,83 and desktop with Core2Duo E6300) but Linux was not really fast on either. It is my fault or is it the truth, that the Windows kernel is able to do processing faster?

hardly a scientific comparison. you don't state what makes up 'fast'. Is the linux system cut down to minimal requirements and maintained correctly?

A problem with the threading under FreeBSD was mentioned and this link mentioned as a reference.
I would like to get to know your opinions on this topic.

The person didn't read the article, and has no idea what they're talking about.

FTA:

This means that applications use slower user-level threads like in FreeBSD and not fast kernel threads like in Linux. It seems that FreeBSD 5.x has somewhat solved the performance problems that were typical for user-level threads, but we are not sure if Mac OS X has been able to take advantage of this.
In order to maintain binary compatibility, Apple might not have been able to implement some of the performance improvements found in the newer BSD kernels.

This is discussing OSX in particular, not FreeBSD. FreeBSD's latest production release is 6.2, not 5, and 7.x is due not far away (and does have further threading improvements iirc).

The article is completely irrelevant to a desktop user too. It talks about forking large amounts of threads in a short period of time, which is not a common desktop load. On top of that, the article is a year and a half old at least, meaning that Apple could well have fixed their complaints in never releases. Heck, the Apples in that article are all still PPC's, Apple are an intel shop now.

Operating system comparisons are pointless at best. They're very subjective, and depend on the load and usage being compared. Different horses for different courses.

James

Last edited by iphitus (2007-03-08 12:48:48)

test1000 · 2007-03-08 15:56:45

for me, there are two applications which you will experience slowness with the most:

1. the filemanager
2. the webbrowser

1. Rox is the only icon-based filemanager i've tried which is as fast on initial startup as windows explorer. HOWEVER you will want to switch that default theme as fast as you can if you care anything about aesthetics

thunar comes in as a close second and is only a bit slower than explorer at initial startup time.

If you take into consideration how fast the filemanagers are when preloaded or already launched, then i wouldn't say there's too much difference. konqueror perhaps is faster than nautilus.

2. I recall opera was faster before i downloaded the latest desktop release but i'm not sure. Apart from that firefox is very very slow for me in comparison to swiftfox, which i find as fast as the windows version of firefox. Even so, swiftfox still starts up a bit slower on initial startup than firefox does in windows. Opera moreso.

I don't experience any difference with mplayer on either platform. Mousepad is neglible slower than notepad. Gvim seems to have a different startup time everytime i start it if you preload this stuff it gets very fast. kde applications are very responsive when preloaded especially i find.

So I guess in conclusion; it's easier to find 'instant-programs' for windows than it is for linux, but rox is an example of that it is possible in linux as irfanview is an example that it is possible in windows. And whatever the bottleneck(if there is one) it can only get better from here on as xorg has picked up development a lot, and the underlaying critical libraries (when you think about speed) like qt and gtk will also be faster

Of course it could also be that developers just need to pay more attention to inital-startup-time but i wouln't know

anyway, if we get this down, AND standards, we will reach our goal of conquering the world! MOHAHAHAHA

phrakture · 2007-03-08 16:08:30

chaosgeisterchen wrote:

In fact, a lot of applications run a lot faster under Windows, compared to Linux.

The difference here has to do with the way graphics are drawn. Windows has their windowing routines mostly _in the kernel_. Linux does not.

Try a command line application (hell, install cygwin and use bash on windows. You quickly see how sluggish it is)

pelle.k · 2007-03-08 18:19:24

To generalize it, the problem seems to be KDE

_My_ KDE installation is certainly _not_ less responsive than my XP installation. I also find _my_ KDE slightly snappier than gnome, but on par with xfce.
This is not meant as flaimbait though...
Maybe it's because KDE has some sort of advantage over other DE:s/Windows, running on a dual core cpu? (plus, I use an nvidia card + RenderAccel etc...)

Also, If i remeber correctly, parts of QT are openGL accelerated. GTK is not, so that might explain why _some_ "linux" apps migt feel more sluggish than their XP counterparts.

chaosgeisterchen · 2007-03-08 19:32:26

phrakture wrote:

chaosgeisterchen wrote:
In fact, a lot of applications run a lot faster under Windows, compared to Linux.
The difference here has to do with the way graphics are drawn. Windows has their windowing routines mostly _in the kernel_. Linux does not.

This is something I miss in Linux. As it has the approach to become a operating system for each and every desktop on the world, they would do themselves good to integrate the drawing directly in the Kernel.

Hopefully, Kernel 2.8 will link X and the Kernel at a level which lets the bottleneck disappear. This would give Linux an enormous advantage over Windows.

test1000 · 2007-03-08 19:40:16

well rox is still fast even without that maybe such a "link" is not needed.

funkyou · 2007-03-08 19:44:30

From my experience, the only thing where linux is slower than windows are the loading times of apps... But it depends. On my machine, opera for example starts faster under linux than under windows...

And really, i dont hope that X gets integrated into the kernel or what, because that would also be the source of more errors and bugs... I like it the way it is now, clean and separated... Kernel --> X --> DE/WM... Take a look at windows: A badly written application or driver can kill the whole system, and thats mainly because they have integrated so much stuff into the kernel...

I think the biggest problem with speed is, that there is no proper driver support from the hardware manufacturers on linux and therefore its not as optimized as it could be...

chaosgeisterchen · 2007-03-08 19:47:15

If one of the kernel module dies, is the kernel also dead? I just do not know already, but it seems unrealistic to me. I imagine X as a kernel module in order to achieve better performance without killing the system when being broken.

But I am not a kernel developer and hence have a rather small idea of how it acutally works. Just guessing things.

skale · 2007-03-09 04:24:53

I actually really like the ROX default theme-- easy to read and easy on the eyes, and that's all I really care about computer-wise. If I wanted something aesthetic, with lots of color and texture and contrast, and personalization, I'd be gardening. Does lined paper have to be customized? Surely those lines could be prettier. You see what I mean. The computer should be like that paper. You just shouldn't have to think about it.

I turned off pretty much all the horsecrap in services.msc and msconfig, and I have to say, it's faster than any linux I have seen. Of course, it can't do anything, but hey, I only use it for SimCity 4 anyway. Of course, if I install the network driver, I have to install an antivirus, and then watch that speed go up in a puff of smoke.

I always keep thinking that X is needlessly complicated, but cannot think of anything better. I am not a programmer, and do not really know, but I suppose the developers cannot make anything simpler, without sacrificing loads of functionalities that most users have grown to depend on.

Actually, you all are neglecting the security aspect. Microsoft's recommendations for security holes routinely involve "multiple layers of security and antivirus software". That alone kills whatever speed you gain. Just think about it. "multiple layers of security software" a solution? I am trying not to burst out laughing.

iphitus · 2007-03-09 12:44:13

chaosgeisterchen wrote:

If one of the kernel module dies, is the kernel also dead? I just do not know already, but it seems unrealistic to me. I imagine X as a kernel module in order to achieve better performance without killing the system when being broken.
But I am not a kernel developer and hence have a rather small idea of how it acutally works. Just guessing things.

Won't ever happen, and is just entirely impractical.

Things are better in userspace than the kernel. For security, simplicity, and practicality reasons, X won't be implemented in the kernel.

As for X dealing with the kernel to acheive better performance, that's direct rendering, where X is given near complete control over the graphics card. Useful for 3D yes, but tends to result in a snappier 2D desktop too.

James

chaosgeisterchen · 2007-03-09 15:22:44

iphitus wrote:

chaosgeisterchen wrote:
If one of the kernel module dies, is the kernel also dead? I just do not know already, but it seems unrealistic to me. I imagine X as a kernel module in order to achieve better performance without killing the system when being broken.
But I am not a kernel developer and hence have a rather small idea of how it acutally works. Just guessing things.
Won't ever happen, and is just entirely impractical.
Things are better in userspace than the kernel. For security, simplicity, and practicality reasons, X won't be implemented in the kernel.
As for X dealing with the kernel to acheive better performance, that's direct rendering, where X is given near complete control over the graphics card. Useful for 3D yes, but tends to result in a snappier 2D desktop too.
James

Thanks for the information. I cannot yet call myself enlightened but I already thought that X and the Kernel have some special purpose, each on their own, otherwise they would have long ago been merged.

gunnix · 2007-03-21 19:47:45

Linux is faster then windows on every computer I tried. Windows hangs a lot when you try to do more then one thing at a time... never experienced this in linux. Mind you I do use icewm and other lightweight apps... I get irritated by ubuntu's sluggyness on my bro's computer... Yet I did run all light apps on windows too... and there's no great multitasking..

arch linuxs boots faster, is faster to upgrade n install apps. It's faster to defragment. Faster in anything... except crashing.

Damnshock · 2007-03-21 22:27:14

There are many things that "explains" why Windows seems faster than Linux

1- Precached apps (try to preload an instance of konqueror and see how quick it starts)
2- Filesystem: ntfs and fat are horrible but as they *don't* do a proper journaling they're faster than linux filesystems ( in fact, fat doesn't support journaling
3- Simpler window manager: try openbox, enlightenment or even fluxbox and you'll notice a significant improvement in speed
4- Windows has a x-windows system integrated into the kernel: if it fails, the computer freezes.

Damnshock

skittlehawk · 2007-04-01 21:03:51

Hmm.. I think it really depends on how you use both operating systems. I used to have an overly tweaked and optimized windows xp and it loaded really fast, even faster than most of the linux distributions that I have tried and it was comparable to my gentoo and arch installs. That was after I disabled many of the system services that I did not need and after tweaking the registry. My gentoo install, on the other hand, logged in faster and shutdown faster than my windows xp ever did (although I didn't shut down much to begin with). So if anyone wants to get the most out of their system, I suggest they better have a good amount of knowledge about their system and use a source based distribution with the proper compiler flags for optimization.

Then again, this is just my experience and thoughts, so don't take it too seriously.

D L

Arch Linux

#1 2007-03-07 18:54:52

How fast is Linux really?

#2 2007-03-07 19:22:40

Re: How fast is Linux really?

#3 2007-03-07 19:52:21

Re: How fast is Linux really?

#4 2007-03-07 21:59:14

Re: How fast is Linux really?

#5 2007-03-07 22:31:23

Re: How fast is Linux really?

#6 2007-03-08 03:02:40

Re: How fast is Linux really?

#7 2007-03-08 04:23:14

Re: How fast is Linux really?

#8 2007-03-08 05:43:45

Re: How fast is Linux really?

#9 2007-03-08 06:20:02

Re: How fast is Linux really?

#10 2007-03-08 07:27:48

Re: How fast is Linux really?

#11 2007-03-08 07:47:38

Re: How fast is Linux really?

#12 2007-03-08 12:46:38

Re: How fast is Linux really?

#13 2007-03-08 15:56:45

Re: How fast is Linux really?

#14 2007-03-08 16:08:30

Re: How fast is Linux really?

#15 2007-03-08 18:19:24

Re: How fast is Linux really?

#16 2007-03-08 19:32:26

Re: How fast is Linux really?

#17 2007-03-08 19:40:16

Re: How fast is Linux really?

#18 2007-03-08 19:44:30

Re: How fast is Linux really?

#19 2007-03-08 19:47:15

Re: How fast is Linux really?

#20 2007-03-09 04:24:53

Re: How fast is Linux really?

#21 2007-03-09 12:44:13

Re: How fast is Linux really?

#22 2007-03-09 15:22:44

Re: How fast is Linux really?

#23 2007-03-21 19:47:45

Re: How fast is Linux really?

#24 2007-03-21 22:27:14

Re: How fast is Linux really?

#25 2007-04-01 21:03:51

Re: How fast is Linux really?

Board footer