Linux vs. Win initial results.

[Belegost]

A few weeks ago I asked if anyone knew whether it was faster to process SETI WUs with Linux or Windows. Some people suggested that using the Win CLI client in Linux through WINE might produce the fastest results. So I went ahead and tested this.

Hardware used:
K6-2 450@333 (the mobo is old and doesn't support higher)
Gigabyte GA-586TX
32MB of PC66 SDRAM
5GB WD HD UDMA33
Everything else should make no difference.

The results can be found here.

The first four units up to Jan 01 were processed in Mandrake 8.1 using the latest version of WINE. The OS was loaded to command line only and only necessary modules were loaded. To save on memory I used the TTY display drivers instead of X11 in the WINE configuration.

The remaining units were processed in Win98SE, it's a fairly clean install with no extra programs running at startup.

I think the results on my system are interesting. Throwing out the highest and lowest times as statistically useless, Windows maintains an average of 45 minutes faster times than Linux with WINE.

Tomorrow I plan on getting the linux native client and running that for a few WUs and will bring back the results to compare.

 

[GulDukat]

Interesting! I always figured that the Linux/Wine combo would win out for something like this because of the (supposedly) lower overhead imposed by a "customized" Linux system.

Thanks, Belegost!

Dukat

 

[Shuxclams]

Belegost, in all fairness, please dsiable all the networking daemons in the rc.x dir and set the client with

[belegost@ pcname/SETI]$setiathome -proxy orangekid.teamanandtech.com:5001 &

The & sends it to the background, leaving the command line free.

I look forward to your results.






SHUX

 

[Poof]

Some people suggested that using the Win CLI client in Linux through WINE might produce the fastest results. So I went ahead and tested this.

I don't "suggest", I test.

Here.
And in here.
And also in here.
Here too.

And in many other threads over at Ars.

The specs of the dual-boot (98SE/Red Hat 6.2) test machine were:

PIII 600 coppermine (overclocked to 800 w/133 FSB)
GlobalWIN VGS08 HS/Fan & Arctic Silver thermal paste
Abit BE6-II rev. 2 mobo
Mushkin PC100 ECC CAS3 RAM (@ CAS 3-3-3)
Voodoo3 2000 AGP
Quantum Bigfoot 10GB IDE HD
D-Link 530TX NIC
Enlight 7237 case w/250W PS
Windows 98SE (1st partition)
Red Hat 6.2 w/kernel 2.4.0-test10 recompiled for i686 (2nd partition)
WINE ver. 20001002

This machine is now OC'd to 972.

Take note of this - you are running your tests on a K62 machine. None of the clients are optimized for that machine. In fact, SETI run on that basically sucks. I have a K6-2/333 (tri-boot 98/NT/SuSE 7.1), K6-2/450 (NetBSD 1.4.1), and K6-2/500 (dual-boot 98/Mandrake 6.5). Believe me, they suck. However, your results might be interesting for archival purposes as a comparison of running SETI on that platform.

 

[RaySun2Be]

I would also suggest using the "test" WU from Ars Technica for each of the benchmarks for each OS. Unless you use the same WU or WUs for each test, there are a number of variable factors angle range, quassians, etc. that can affect the results.

 

[Poof]

Agree with Ray. In order to fairly test, you must use the same benchmark WU (which Ars already has setup).

Also note what parameters you ran WINE with and whether you point to a native windows partition for system files or one of the "fake windows" files that can be downloaded. My test machines had native partitions that I have WINE use.

I run WINE as (having shortened the CLI to seti.exe):

wine --winver nt40 -- seti.exe -proxy (ip of my Setiqueue:5517)

The above uses the NT40 file system, which will give different times than if you use the "win95" or "win98" options (these being WINE's default).

 

[Sukhoi]

Hey Poof, do I need NT 4.0 installed to use that switch, or can I just copy the i386 folder to the HDD?