Comparison between Win XP and Linux Distros? (Benchmark)

[civad]

A friend of mine asked me recently:
"Is there a way to do a head-to-head comparison between Win XP (or any other version of Windows) and Linux? (Any distro)"

Is there a benchmarking program available that could do so?

 

[n0cmonkey]

They're fairly different, so Im not sure there would be much point. Who do you want to win? What do you want to benchmark? I see this as being almost as useless as comparing WinXP and Mac OS X benchmarks...

 

[Bremen]

What exactly would you compare? Seriously. They are two very different operating systems. While low level stuff like file creation could be relevant, you're hampered by 5 possible choices for the file system in linux. It is very much like comparing apples to oranges. Besides which very few people even tax the hardware of low end systems anymore (I write this on a PIII 750 that just browses the net). High end stuff (big iron) cares more about uptime/service than raw numbers.

Sure you could run some benchmarks and find Linux servers 1.5 million webpages a second while windows does 1.4 a second (I made the numbers up, no idea whats realistic) but who has a pipe that can handle that anyway??? Add in the inevitable claims you didn't set one up correctly (there are always additional features to disable you don't need). Its much more hassle than its worth (which is very little). Just look at any mobo comparison on Toms Hardware or Anandtech, they're all so close I don't give a darn... I get Asus cuz they have a rep for quality, what do I care if the Abit board did .3 fps better?

Ok, sorry, that was a bit more of a rant than I intended.

 

[Nothinman]

Yea, you have to pick what you want to benchmark first.

 

[zephyrprime]

In terms of low level system stuff, all versions of windows are generally slower in just about evey category than any variant of *nix. I remember for class I had to program a memory allocator and compare it's speed to the memory allocation function in Unix. Since I have a windows xp machine at home, I also compared my code to windows. I was shocked to see that Windows XP allocated memory 200x slower than Solaris. My comparison isn't apples to apples though because the hardware platforms were totally different (the Sun I was using was an older model and my pc is a athlon 1700).

In terms of applications though, windows may add some overhead but performance would be more a function of the application itself and the hardware being used.

 

[CTho9305]

Originally posted by: zephyrprime
In terms of low level system stuff, all versions of windows are generally slower in just about evey category than any variant of *nix. I remember for class I had to program a memory allocator and compare it's speed to the memory allocation function in Unix. Since I have a windows xp machine at home, I also compared my code to windows. I was shocked to see that Windows XP allocated memory 200x slower than Solaris. My comparison isn't apples to apples though because the hardware platforms were totally different (the Sun I was using was an older model and my pc is a athlon 1700).

In terms of applications though, windows may add some overhead but performance would be more a function of the application itself and the hardware being used.

Read this. XP/2k do context switches faster as long as you have more than two theads (say, init and bash). Low level slower? . Solaris -> windows is NOT usefil info because (as you said), the hardware is different. just because the sun was old and the Athlon was new, you can't expect the athlon to win. Sparcs are COMPLETELY different and do some things really well, even when slower.

 

[Nothinman]

XP/2k do context switches faster as long as you have more than two theads (say, init and bash). Low level slower?

For that test to mean anything it has to be done under moderate load, and we all know how well Windows handles load =)

Also it would have been good to see if the new O(1) scheduler improves on context switch speed at all.

 

[Vehementi]

You could compare MFLOPs/MIPs on the same system.

 

[MGMorden]

One thing for certain: if you're looking at the GUI aspect of the system Windows is faster (by a ways). This is pretty much because all X11 apps (even local ones) are drawing to the screen through the TCP/IP stack. On a local system that's just going around your elbow to get to your thumb (works great for remote administration though). Granted that's not part of Linux, but you mentioned comparing distros, not Linux itself. On a lower system level Linux can do just as good if not better than Windows.

 

[MGMorden]

Originally posted by: Vehementi
You could compare MFLOPs/MIPs on the same system.

That's not really going to vary depending on the OS. The processor can churn out the same amount of work but what really matters is how efficiently the OS is dividing up that work and such (MIPS might be the same, but if Linux can do something in 20 instructions and it takes Windows 31, then over time Linux is going to do something that depends on that faster).

 

[Nothinman]

This is pretty much because all X11 apps (even local ones) are drawing to the screen through the TCP/IP stack.

Actually local X apps talk to the X server via a local unix socket, not TCP/IP. Infact most Linux distros disable the X server's TCP listening port by default because it's an unnecessary security risk.

 

[CTho9305]

Originally posted by: Nothinman

XP/2k do context switches faster as long as you have more than two theads (say, init and bash). Low level slower?

For that test to mean anything it has to be done under moderate load, and we all know how well Windows handles load =)

Also it would have been good to see if the new O(1) scheduler improves on context switch speed at all.

I was just trolling because he made an un-backed generalization. I highly doubt he knows anything about kernel differences (not that I know much myself )

 

[Nothinman]

I was just trolling because he made an un-backed generalization

Even so I was still surprised that Windows task switches are faster than Linux, especially considering they have to basically do the same thing.

 

[CTho9305]

Originally posted by: Nothinman

I was just trolling because he made an un-backed generalization

Even so I was still surprised that Windows task switches are faster than Linux, especially considering they have to basically do the same thing.

Yea, that is pretty interesting. A related question is, which task switches more frequently?

 

[Nothinman]

Yea, that is pretty interesting. A related question is, which task switches more frequently?

I would have to say Windows, looking at perfmon right now shows my Win2K laptop doing ~1000 context swtiches per second, and it's basically idle.

 

[CTho9305]

Originally posted by: Nothinman

Yea, that is pretty interesting. A related question is, which task switches more frequently?

I would have to say Windows, looking at perfmon right now shows my Win2K laptop doing ~1000 context swtiches per second, and it's basically idle.

how do you check that in linux? and how would I check this in windows?

remember, linux has a lot fewer cool fade and scroll effects, so turn em off in windows for a fair test

 

[Nothinman]

how do you check that in linux? and how would I check this in windows?

Perfmon in Windows and procinfo or vmstat in Linux.

remember, linux has a lot fewer cool fade and scroll effects, so turn em off in windows for a fair test

I even have a local Group Policy that turns off all my desktop icons =)