Re: Re: Linux/Windows Universal Benchmark
On Thu, Apr 15, 2004 at 12:13:59AM +0100, Andy Morris wrote:
> Just been browsing and came across this dicussion, and thought i might be
> able to give some input.
>
> I'm a software grad (well this summer) and use both linux(gentoo 2.6.xxx
> cant remember specifically) and xp pro sp1 as home os's (games, dev etc)
> and when building a fairly small application i felt that it was taking too
> long to build my code using linux (approx 70 secs). So I logged into my
> windows and it did it in under 40 secs, quite a difference.
>
You are also testing compilers here and gcc isn't a very fast
compiler. The resulting code is good though. If you are comparing it to
msvc you also need to make sure that its not using precompiled headers
which would affect the compilation speed.
> However, these two machines were of different spec
>
> xp: 2500xp(1.833mhz) barton 512cache,1gig ram, 7200 sata hdd (high spec)
> linux:1900xp(1.6mhz) thoroughbred 256cache, 1gig ram, 5400 ide hdd (fairly
> low spec)
>
I would check your setup since for me linux is usually much faster then
xp. I did a test once with matlab doing some heavy computation that
took several minutes. Running matlab on a laptop running linux with a
1.3G amd athelon cpu with 256M ram and 5400 rpm ide disk, finished at
the same time as a 2G pIII desktop with 2G ram and 7200 rpm disk
running XP pro (it didn't hit the disk though so I don't think the
amount of ram was a factor).
> I would accept a relatively small diff between the two machines (but a
> difference since one is more powerful) and since there were approx 150
> source files the hard drive reads could have caused a difference.
>
> So what I had was a biased test result.
>
> My first thoughts would be to write a tiny application that would not use
> the harddrive and compare again, so i made a simple program to count from
> 0 to a parameter x number of times, test data was to count 0-999999999 100
> times. XP box did it in 3mins 10 secs, linux 5 mins 4 secs ( i did this
> numerous times and results were always v similar). Again big difference, so
> it definetely wasn't the harddrive's access time etc. So the only way I
> was going to find out was to install Linux on my xp box and dual boot it
> (which I did).
>
> So XP and Gentoo Linux on the same box, and xp still beats it hands down
> (it only improved by around 30 secs), which really dissapointed me. It
> makes me want to use my xp box for more stuff than previously, which i had
> been trying not since i like using kde over the Windows desktop
>
> Now one thing I should have mentioned earlier is that the app was written
> in Java (compiled using Sun's jdk1.4.2 (not gentoo's blackdown since it's
> too buggy)) and we are therefore also testing the platform implementation
> of the VM (if you don't know about java basically when code runs there is a
> middleman between the code running and the OS (the Java Virtual Machine).
> It is possible that the jdk does not work as well on Linux but this is what
> I use and so do millions of others, and therefore it's can be a v good
> benchmark. I will shortly if i get some time repeat the tests in C++ to
> remove this factor (it interfaces directly with the OS since it's compiled
> into native binaries) and if any1 does care for the result then let me know.
>
You are testing java implementations and optimizations. If you want to
have a better comparison you could try running sun's jdk on both
machines, although I don't know how similar the implementation for
windows and linux is.
> Now if you think my benchmark is totally unsafe/inaccurate as a means for
> benchmarking then feel free to state so, but compiling and running java
> code is something I do frequently, and Linux can't seem to do it as quickly
> as XP.
>
> Just to sum up my findings, from my testing XP seems to run simple
> sequential cpu tasks not much less than twice the speed of Linux.
>
> My afterthoughts about the two OS's is that if I ran multiple threads of my
> test (say 500) that my XP box would flake out and linux would handle it
> much better, perhaps when i go away for the wkend next i'll do that but
> right now iv got far too much work to do on them (final yr proj).
>
> Hope this helped any1,
>
> -andy
>
> _________________________________________________________________
> It's fast, it's easy and it's free. Get MSN Messenger today!
> http://www.msn.co.uk/messenger
>
>
> --
> To UNSUBSCRIBE, email to debian-user-REQUEST@lists.debian.org
> with a subject of "unsubscribe". Trouble? Contact
> listmaster@lists.debian.org
>
>
> +++++++++++++++++++++++++++++++++++++++++++
> This Mail Was Scanned By Mail-seCure System
> at the Tel-Aviv University CC.
>
Reply to: