Message boards :
Number crunching :
Linux vs Windows
Message board moderation
Author | Message |
---|---|
Spectrum Send message Joined: 14 Jun 99 Posts: 468 Credit: 53,129,336 RAC: 0 |
Hi everyone. I was wondering if anyone has done a comparison between a system running Windows to one running Linux (same Computer dual boot) any comments, ideas or experience would be appreciated.! |
Metod, S56RKO Send message Joined: 27 Sep 02 Posts: 309 Credit: 113,221,277 RAC: 9 |
> I was wondering if anyone has done a comparison between a system running > Windows to one running Linux (same Computer dual boot) any comments, ideas or > experience would be appreciated.! Actually I just did it today. It's pain in the a#$%( to measure time for crunching one WU properly under Windows (if you run app standalone, graphics window pops up. You can minimize it, but I'm not sure wether graphics consumes any CPU or not). I also noticed that the speed of official seti crunching app depends greatly upon CPU in question. Said that, here's the result: windows cruncher is approximately 30% slower than linux cruncher on a HP xw4200 with P4 HT @3.2GHz and 1GB of RAM. For more results (different machines, different linux binaries) see my page. Metod ... |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
> It's pain in the a#$%( to measure time for crunching one WU properly under Windows (if you run app standalone, graphics window pops up. You can minimize it, but I'm not sure wether graphics consumes any CPU or not). You mean the screen saver with graphics? Yes, it does slow down crunching very much. Don't know how much, but in Classic I saw about 50% diference. So, I never use a screen saver. > Said that, here's the result: windows cruncher is approximately 30% slower than linux cruncher on a HP xw4200 with P4 HT @3.2GHz and 1GB of RAM. Interesting! What if you disable the screen saver in Windows? Is it the same 30% then? |
Spectrum Send message Joined: 14 Jun 99 Posts: 468 Credit: 53,129,336 RAC: 0 |
> > I was wondering if anyone has done a comparison between a system running > > Windows to one running Linux (same Computer dual boot) any comments, > ideas or > > experience would be appreciated.! > > Actually I just did it today. > > It's pain in the a#$%( to measure time for crunching one WU properly under > Windows (if you run app standalone, graphics window pops up. You can minimize > it, but I'm not sure wether graphics consumes any CPU or not). I also noticed > that the speed of official seti crunching app depends greatly upon CPU in > question. > > Said that, here's the result: windows cruncher is approximately 30% slower > than linux cruncher on a <a> href="http://www.hp.com/workstations/pws/xw4200/index.html">HP xw4200[/url] with > P4 HT @3.2GHz and 1GB of RAM. > > For more results (different machines, different linux binaries) see <a> href="http://mkx.feridot.com/boinc/">my page[/url]. > Thanks for the input, I only really look at the average times so if I can see an increase of around OF 20 TO 30% by running Linux I think I will run my machine on Linux overnite when it is idle and at other times when I am watching movies and such until I get a lot more familiar with the Linux platform, my son is a bit of a wiz at it so I am Trying in vain to catch up! (I only just got him to change over to Boinc and he is the team Creator) Best regards and Greets from West OZ Lou Manzi |
Metod, S56RKO Send message Joined: 27 Sep 02 Posts: 309 Credit: 113,221,277 RAC: 9 |
> You mean the screen saver with graphics? No. What I did was to run SETI/boinc science app in standalone mode. This can be done as follows: - create a new directory. - copy setiathome-?.??-windows-intel.exe and setiathome-?.??-windows-intel.pdb from your boinc/projects/setiathome.berkeley.edu directory - get reference_work_unit.sah and rename it to work_unit.sah - open a command window (start->run cmd) and run setiathome-?.??-windows-intel.exe It will open a graphics window and I have been said that if you minimize it immediately, no graphics processing will take place. When the setiathome app finishes, you get the time it took in init_data.xml. I compared this number to the one from linux obtained the same way (more or less). Metod ... |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
It's a little surpricing that Linux is more efficient on BOINC, because many people here in the forum seems to think that Windows is better optimized for BOINC. See here for instance: Optimum os to run Seti on |
Metod, S56RKO Send message Joined: 27 Sep 02 Posts: 309 Credit: 113,221,277 RAC: 9 |
> It's a little surpricing that Linux is more efficient on BOINC, because many > people here in the forum seems to think that Windows is better optimized for > BOINC. Indeed. I was surprised too. But might be that it's due to large L2 cache (1M) of this processor. I've compiled a set of optimized seti binaries for linux which show something like 45% of speedup on P4 comparing to official linux binary. On this machine, speedup was miserable 15%, so it seems that official linux seti binary really does its best on this configuration. My test is by no means definitve. And I don't intend to run tests on other machines (well, one perhaps) as the rest of my boxes are not dual-boot. A side note: I also tried to run windows seti cruncher with wine in linux and it performed similarly as natively. This might open more room for comparision of windows vs. linux binaries. [edit] I did briefly follow the discussion you're mentioning. The problem with it is that it's based on feelings and assumptions (like mean time for processing some tens of WUs which depends very much on WUs themselves). My case is a bit more scientiffic as the exactly the same amount of work had to be done. [/edit] [edit 2] From what I wrote above I could conclude that normally windows and linux binaries should perform similarly on the same hardware. However, you have a choice of optimized binaries for linux for different processors (Ned built some for AMD CPUs, I built some for Intel CPUs) so the odds for beating Windows in time per WU with linux on any hardware are good. [/edit 2] Metod ... |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
|
Metod, S56RKO Send message Joined: 27 Sep 02 Posts: 309 Credit: 113,221,277 RAC: 9 |
> I see. Well, I have only two licenced Windows XP Pro, so if I were to set up > a SETI farm, I'd have no other choice but to use Linux. Otherwise I'd be > ruined just by purchasing operating systems... Indeed. Most of my machines are not dual-boot as they are Linux only. Most of my machines are Windows as I'm forced to do things which I can't do on linux in a company-compatible way :( Metod ... |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
I saw your page with binaries and timings, and I may have to go there to get optimized Linux binaries if I need one later. My current setup is AMD 2500+ (Barton), Windows XP Pro SP2 and 1 GB PC2700 (333 MHz) RAM. It performs decently well and WU's land in about 12-13000 secs average. So I'm about 200 BOINC credits average per day. I use BOINC v4.19 and CLI v4.09. |
N/A Send message Joined: 18 May 01 Posts: 3718 Credit: 93,649 RAC: 0 |
I "lost" about 100 MIPS float and integer when I went from OS X.3.8 to YDL 4. I haven't crunched actively under Linux, so I can't produce any corroborating evidence. |
Raymon Purgason Send message Joined: 24 Oct 03 Posts: 27 Credit: 14,251,492 RAC: 0 |
Hello everone- I just built a new P-4 3.0 hpt, 1 gig mem, I have dual loaded win xp home and red hat 9 ver 2.4.20-6smp, linux ver 4.27 boinc and win ver 4.19. The linux kernel and boinc are just as they came , same with the win ver. For 10 work units linux had a mean of 3.8 hr and the win xp had a mean of 3.2 hr. I ame looking forward to getting an optimized ver of linux ver 4.27 if one becomes available. |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
From my experience, the native Windows client is 10-20% faster than the native linux client. But I only have experience of AthlonXP processors and it might vary for different processors/cache levels. However, there are now optimized seti clients available for linux (it's far easier to compile your own custom optimised client for linux) that are faster than the windows client on the same hardware. I would say if you're running a dual boot system with the native clients, leave it running in windows for best performance unless you know your own optimized linux client is faster. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
at1839 Send message Joined: 8 Feb 01 Posts: 29 Credit: 5,899,847 RAC: 0 |
Hello all. I think we can try two different kind of comparison. We can compare STANDARD crunchers on the same hardware. Usually this will sort up Windows as winner. We can compare, over the same hardware, the BEST cruncher can be build for. I must admit that will again have Windows as winner. Take note an optimized cruncher can be as fast to cut in half times. Mine are both. This is my story, I do have both Windows & Linux optimized clients up. A friend built up Windows one and I built Linux lol, based on fftw3 and icc. They're both really aggressive but Windows is faster, grrrrrr :) Paolo, iw6dak (at arrl.net) > > It's a little surpricing that Linux is more efficient on BOINC, because > many > > people here in the forum seems to think that Windows is better optimized > for > > BOINC. > |
terrorhertz Send message Joined: 26 Mar 00 Posts: 401 Credit: 31,534 RAC: 0 |
I have tried a comparison with 2 machines. one with windows and one with optimized linux on Boinc and I'm sad to say that the windows soooo far has beat it. BUT Boinc is new just as SETI@Home classic was new and in the end we beat it!!!!!! I have all the confidence in the world that we will again. @ MS people: what are or can you do to speed up the processes of your computers operating system? wait on Bill? He has a lot of VERY well payed programmers but they have goals others than SETI. There are a lot of programers for windows out there but as this thread asks... can they stand up to Linux programmers? I don't think so. Human will will win! |
Chris Bosshard Send message Joined: 5 Jun 99 Posts: 86 Credit: 3,474,583 RAC: 0 |
Before we drift too far in Tux Vs. Bill discussions, I would like to throw in my experiences: Generally the original Linux Seti client is around 10% slower than the original Windows version. Most of the machines I have access to are AMD and a few Intel machines. The Intel based machines are also slower on Linux than on Windows, when looking at the average of real results. The picture changes when you start to compile your own clients on Linux. You will have the chance to get similar results or even a bit better than the original Windows client. Fact is that your off the shelf Linux system comes with all the tools to compile your own clients. Windows does not... Chris Bosshard Visit my homepage astroinfo SETI page |
Razorirr Send message Joined: 3 Apr 99 Posts: 92 Credit: 7,414 RAC: 0 |
some other crazy people like me too then. i want to get a copy of sparc/solaris on here too but i would need a memory upgrade for the laptop. 18 gigs between three OS'es would be sorta small. ~boinc its not the credit or satisfaction its the screensavers~ S@h Berkeley's Staff Friends Club ©members |
Prognatus Send message Joined: 6 Jul 99 Posts: 1600 Credit: 391,546 RAC: 0 |
|
at1839 Send message Joined: 8 Feb 01 Posts: 29 Credit: 5,899,847 RAC: 0 |
Bjorn, > at1839, > > > This is my story, I do have both Windows & Linux optimized clients > up. > > Did you compile the Windows client yourself > the Windows client was built by two Italian guys for team members of Hardware Upgrade http://setiweb.ssl.berkeley.edu/team_display.php?teamid=30265 Sorry, but it's reserved to team members :-) I compiled the Linux one and you can find more info here: https://sourceforge.net/forum/forum.php?thread_id=1256083&forum_id=422289 Both clients are based on fftw3 libraries thanks to the original work of Eric Eheien. Paolo. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.