Windows vs. Linux

Message boards : Number crunching : Windows vs. Linux
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 5 · Next

AuthorMessage
DeMus
Avatar

Send message
Joined: 5 Jan 08
Posts: 238
Credit: 1,765,862
RAC: 0
Netherlands
Message 723822 - Posted: 9 Mar 2008, 18:07:24 UTC

As was mentioned in the thread: "Switching from one PC to another" I found out that an installation of Linux Fedora Core 8-64bit runs 2/3 as fast than Windows XP-64bit on the same machine.
I have made a dual boot and installed on both OS'es Boinc software. When I look at the computers info I see for the Linux version:
1565 Measured floating point speed and
4426 integer speed.
The Windows version gives these results:
2260 Measured floating point speed and
7298 integer speed.

Are there people out there who have the same, or just the opposite, results? Especially Linux fans always shout that Linux is faster than Windows, and how much I hate to admit it, I do have to say that on my machine Linux simply can not beat Windows.

I started this new thread because the subject of the first does not cover this contents. I wrote it for other purposes.

DeMus.
ID: 723822 · Report as offensive
Profile Fivestar Crashtest
Volunteer tester
Avatar

Send message
Joined: 10 Dec 99
Posts: 226
Credit: 5,377,978
RAC: 0
United States
Message 723827 - Posted: 9 Mar 2008, 18:21:10 UTC - in response to Message 723822.  


Are there people out there who have the same, or just the opposite, results? Especially Linux fans always shout that Linux is faster than Windows, and how much I hate to admit it, I do have to say that on my machine Linux simply can not beat Windows.



DeMus.


It is my experience that Linux has not been as fast as Windows on my computers for Seti@home. Windows XP > Linux 64 bit > Linux 32 bit.
ID: 723827 · Report as offensive
Robert Smith
Avatar

Send message
Joined: 15 Jan 01
Posts: 266
Credit: 65,761
RAC: 0
United Kingdom
Message 723882 - Posted: 9 Mar 2008, 20:06:05 UTC

ID: 723882 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 12194
Credit: 125,038,160
RAC: 38,970
United Kingdom
Message 723884 - Posted: 9 Mar 2008, 20:17:53 UTC - in response to Message 723882.  

http://setiathome.berkeley.edu/forum_thread.php?id=45737

Copying from DeMus's original thread:
Interesting observation. Not often we have such a direct comparison between stock Windows 5.27 application and stock Linux 5.28 application on the same hardware, and with a nice even supply of consistent ARs too.

~10,000+ seconds for Windows: ~15,000+ seconds for Linux. Someone might want to look into that one.

So this isn't a benchmark question, Robert: it's actual computation.
ID: 723884 · Report as offensive
Profile RandyC
Avatar

Send message
Joined: 20 Oct 99
Posts: 714
Credit: 1,704,345
RAC: 0
United States
Message 723907 - Posted: 9 Mar 2008, 21:50:26 UTC - in response to Message 723822.  

As was mentioned in the thread: "Switching from one PC to another" I found out that an installation of Linux Fedora Core 8-64bit runs 2/3 as fast than Windows XP-64bit on the same machine.
I have made a dual boot and installed on both OS'es Boinc software. When I look at the computers info I see for the Linux version:
1565 Measured floating point speed and
4426 integer speed.
The Windows version gives these results:
2260 Measured floating point speed and
7298 integer speed.

Are there people out there who have the same, or just the opposite, results? Especially Linux fans always shout that Linux is faster than Windows, and how much I hate to admit it, I do have to say that on my machine Linux simply can not beat Windows.

I started this new thread because the subject of the first does not cover this contents. I wrote it for other purposes.

DeMus.


Do you have a process named 'Beagle' running? Over at Einstein they found it was chewing up to 25% of cpu cycles. See thread http://einstein.phys.uwm.edu/forum_thread.php?id=6527
ID: 723907 · Report as offensive
Profile Mumps [MM]
Volunteer tester
Avatar

Send message
Joined: 11 Feb 08
Posts: 4454
Credit: 63,702,996
RAC: 46,316
United States
Message 723958 - Posted: 9 Mar 2008, 23:24:04 UTC - in response to Message 723907.  

Do you have a process named 'Beagle' running? Over at Einstein they found it was chewing up to 25% of cpu cycles. See thread http://einstein.phys.uwm.edu/forum_thread.php?id=6527


Actually a very good question Randy. And a quick peek at the "Average CPU efficiency" on the computer summary page would give a hint if there was anything consuming a lot of cycles. Beagle, or anything else for that matter. On a system thet pretty much was running just BOINC, that number should be very close to 1.000. Anything less means something else has been using cycles, and would explain the lower performance. Especially if the Windows CPU Efficiency is showing a higher percentage.
ID: 723958 · Report as offensive
Profile Pilot
Avatar

Send message
Joined: 18 May 99
Posts: 534
Credit: 5,475,482
RAC: 0
Message 724033 - Posted: 10 Mar 2008, 3:13:24 UTC - in response to Message 723958.  
Last modified: 10 Mar 2008, 3:14:58 UTC

Do you have a process named 'Beagle' running? Over at Einstein they found it was chewing up to 25% of cpu cycles. See thread http://einstein.phys.uwm.edu/forum_thread.php?id=6527


Actually a very good question Randy. And a quick peek at the "Average CPU efficiency" on the computer summary page would give a hint if there was anything consuming a lot of cycles. Beagle, or anything else for that matter. On a system thet pretty much was running just BOINC, that number should be very close to 1.000. Anything less means something else has been using cycles, and would explain the lower performance. Especially if the Windows CPU Efficiency is showing a higher percentage.

Let me do a quick logic check here.
If Unix/Linux platform shows less performance than Windows on the same hardware, then it can be assumed or suspected that some other program must be causing the problem.
At the same time, if the results are reversed, then the obvious answer is it's because of bloatware in Windows.
Hmmmm sounds like some bias is surfacing here?
Just for a sanity check, how much difference is there between credit claimed by 5.27 and 5.28 applications?
When we finally figure it all out, all the rules will change and we can start all over again.
ID: 724033 · Report as offensive
DeMus
Avatar

Send message
Joined: 5 Jan 08
Posts: 238
Credit: 1,765,862
RAC: 0
Netherlands
Message 724050 - Posted: 10 Mar 2008, 4:24:06 UTC - in response to Message 724033.  

Do you have a process named 'Beagle' running? Over at Einstein they found it was chewing up to 25% of cpu cycles. See thread http://einstein.phys.uwm.edu/forum_thread.php?id=6527


Actually a very good question Randy. And a quick peek at the "Average CPU efficiency" on the computer summary page would give a hint if there was anything consuming a lot of cycles. Beagle, or anything else for that matter. On a system thet pretty much was running just BOINC, that number should be very close to 1.000. Anything less means something else has been using cycles, and would explain the lower performance. Especially if the Windows CPU Efficiency is showing a higher percentage.

Let me do a quick logic check here.
If Unix/Linux platform shows less performance than Windows on the same hardware, then it can be assumed or suspected that some other program must be causing the problem.
At the same time, if the results are reversed, then the obvious answer is it's because of bloatware in Windows.
Hmmmm sounds like some bias is surfacing here?
Just for a sanity check, how much difference is there between credit claimed by 5.27 and 5.28 applications?


Hi,
Beagle is not running. Seti uses 4 times 24% of the CPU (Quad core) so I don't think it is caused by another process. I have been monitoring the CPU usage for a longer period and the percentages don't change much. The same happens in the Windows version although I must say there I see that each process uses 25% of the CPU, with one falling to 24% if another process kicks in.


______
DeMus


ID: 724050 · Report as offensive
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1806
Credit: 254,311,688
RAC: 10,904
Australia
Message 724069 - Posted: 10 Mar 2008, 6:34:55 UTC
Last modified: 10 Mar 2008, 6:41:19 UTC

Hey Demus
What is the difference in the Task Duration Correction Factor between the 2 OS's on that machine ?

All my Linux Boxes a TDCF between 0.20 & 0.5 indicating that they are crunching a lot quicker than the benchmarks estimate (and the slower ones are busy otherwise) My Windows machines vary from 1.2 to 0.3 with the average around 0.7. All the "serious" SETI machines have a CPU effiency of better than 99.5%

If you sit down and work it out by dividing the Benchmark by the TDCF (?), the 2 OS's probably come out about even

It is known that the Linux bench marks are flawed as stated in this thread

http://setiathome.berkeley.edu/forum_thread.php?id=45737


I have found that on a given machine the Linux benchmarks are only about 50% of the Windows benchmarks but the Linux unit processing times are slightly faster.

One of my machines with an Athlon 2200XP CPU doubled it's RAC when I changed from Windows to PCLinuxOS. Although on an Athlon 3200-64, Windows is about 1.5 credits per hour ahead of Linux, but this machine also got new memory and a BIOS upgrade when Windows was installed.

Regards
Brodo
ID: 724069 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 12194
Credit: 125,038,160
RAC: 38,970
United Kingdom
Message 724103 - Posted: 10 Mar 2008, 9:38:08 UTC

It's not DCF, benchmarks, or another process stealing CPU cycles: none of those would show up as a diffierence in the crunching time of real WUs reported to Berkeley.

All I can think of is some sort of energy saving feature running the CPU slower in Linux.
ID: 724103 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9424
Credit: 7,359,571
RAC: 225
United Kingdom
Message 724118 - Posted: 10 Mar 2008, 10:33:20 UTC
Last modified: 10 Mar 2008, 10:35:40 UTC

Which OS is 'faster' is not that simple a question...

What you see in whatever speed results, is it due to:

    * The way the application is compiled
    * Other applications/processes stealing CPU time
    * Other 'bloatware' stealing time and RAM space
    * 'Powersaving' features
    * Any other differences
    * How you are measuring whatever it is you're measuring



Just one small example. In years gone by with s@h-classic, the best speeds achieved were with:

s@h (windows compile) running on WINE on Linux (fastest)
s@h (linux compile) running on Linux
s@h (windows compile) running on Windows (slowest)


There are even reports that for multicore CPUs, you can optimise WU throughput by selecting what mix of WUs from various projects are run in parallel...

Look again? (Especially at the various Lunatics compiles :-) )

Happy crunchin,
Martin


See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 724118 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9424
Credit: 7,359,571
RAC: 225
United Kingdom
Message 724194 - Posted: 10 Mar 2008, 15:02:02 UTC - in response to Message 724118.  
Last modified: 10 Mar 2008, 15:03:05 UTC

Ooops! Correction on the speeds list!!

Which OS is 'faster' is not that simple a question...

What you see in whatever speed results, is it due to:

    * The way the application is compiled
    * Other applications/processes stealing CPU time
    * Other 'bloatware' stealing time and RAM space
    * 'Powersaving' features
    * Any other differences
    * How you are measuring whatever it is you're measuring



Just one small example. In years gone by with s@h-classic, the best speeds achieved were with:


s@h (windows compile) running on WINE on Linux (fastest)
s@h (windows compile) running on Windows
s@h (linux compile) running on Linux (slowest)

There are even reports that for multicore CPUs, you can optimise WU throughput by selecting what mix of WUs from various projects are run in parallel...

Look again? (Especially at the various Lunatics compiles :-) )


Happy crunchin,
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 724194 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 12194
Credit: 125,038,160
RAC: 38,970
United Kingdom
Message 724199 - Posted: 10 Mar 2008, 15:20:28 UTC - in response to Message 724194.  

Just one small example. In years gone by with s@h-classic, the best speeds achieved were with:

s@h (windows compile) running on WINE on Linux (fastest)
s@h (windows compile) running on Windows
s@h (linux compile) running on Linux (slowest)

Martin,

Can you put any actual figures to that?

Did it match what we're seeing here?

Linux: over 15,000 seconds for 72 credits
Windows: 10,000 seconds for 72 credits

We're assured it's the same hardware (and the CPU steppings match): is the Linux stock app really that awful?
ID: 724199 · Report as offensive
DeMus
Avatar

Send message
Joined: 5 Jan 08
Posts: 238
Credit: 1,765,862
RAC: 0
Netherlands
Message 724227 - Posted: 10 Mar 2008, 17:06:19 UTC - in response to Message 724069.  

Hey Demus
What is the difference in the Task Duration Correction Factor between the 2 OS's on that machine ?

All my Linux Boxes a TDCF between 0.20 & 0.5 indicating that they are crunching a lot quicker than the benchmarks estimate (and the slower ones are busy otherwise) My Windows machines vary from 1.2 to 0.3 with the average around 0.7. All the "serious" SETI machines have a CPU effiency of better than 99.5%

If you sit down and work it out by dividing the Benchmark by the TDCF (?), the 2 OS's probably come out about even

It is known that the Linux bench marks are flawed as stated in this thread

http://setiathome.berkeley.edu/forum_thread.php?id=45737


I have found that on a given machine the Linux benchmarks are only about 50% of the Windows benchmarks but the Linux unit processing times are slightly faster.

One of my machines with an Athlon 2200XP CPU doubled it's RAC when I changed from Windows to PCLinuxOS. Although on an Athlon 3200-64, Windows is about 1.5 credits per hour ahead of Linux, but this machine also got new memory and a BIOS upgrade when Windows was installed.

Regards
Brodo


Hi Brodo,

The TDCF for the Linux machine is 0.283949, while for the Windows machine it is 0.295484. Not that much difference, is it?

More numbers to think about:
Linux version:
1565 Measured floating point speed and 4426 integer speed.
The Windows version gives these results:
2260 Measured floating point speed and 7298 integer speed.

Linux:
While BOINC running, % of time work is allowed 99.9929 %
Windows:
While BOINC running, % of time work is allowed 99.9472 %

Linux:
Average CPU efficiency 0.986624
Windows:
Average CPU efficiency 0.990829

Did I forget anything? My computers are visible so take a look for yourself if you need more info.


______
DeMus


ID: 724227 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9424
Credit: 7,359,571
RAC: 225
United Kingdom
Message 724288 - Posted: 10 Mar 2008, 20:37:51 UTC - in response to Message 724199.  
Last modified: 10 Mar 2008, 20:39:55 UTC

Just one small example. In years gone by with s@h-classic, the best speeds achieved were with:

s@h (windows compile) running on WINE on Linux (fastest)
s@h (windows compile) running on Windows
s@h (linux compile) running on Linux (slowest)

Martin,

Can you put any actual figures to that?

This is back in the day of 1 credit per WU, setiSpy and setiQ, and the s@h client not changing for a very long time and all running on much slower hardware than now.

Try searching for comments on the old forums and for setiSpy.

That example was just given as an example that the story is a lot more than just "what OS".


Did it match what we're seeing here?

Linux: over 15,000 seconds for 72 credits
Windows: 10,000 seconds for 72 credits

Sorry, there's no direct comparison to be made. You now have a very different client. doing a greater depth of analysis, on much more recent hardware.

We're assured it's the same hardware (and the CPU steppings match): is the Linux stock app really that awful?

The difference back then was in how the code was compiled.

There is still a big difference to be had in how the code is compiled that is far more significant than which OS you are using. Hence the various Lunatics and KWSN and others' efforts, sometimes with spectacular (fast) results.

A real nasty "gotcha" in my opinion is also for what appears to me to be deliberate sabotage for AMD CPUs for certain important circumstances when you use the Intel compiler for your code. Look up the many words written about "Naughty Intel". That 'fiendish feature' (my description only) has caused many thousands of wasted hours of needless compute time for some of my machines...


It's all in the code, if you can see it.

Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 724288 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 12194
Credit: 125,038,160
RAC: 38,970
United Kingdom
Message 724310 - Posted: 10 Mar 2008, 22:12:24 UTC - in response to Message 724288.  

Did it match what we're seeing here?

Linux: over 15,000 seconds for 72 credits
Windows: 10,000 seconds for 72 credits

Sorry, there's no direct comparison to be made. You now have a very different client. doing a greater depth of analysis, on much more recent hardware.

OK

Forget the ancient history. This is the here-and-now.

And forget fancy (paid for) compilers, with fancy (anticompetitive) vendor strings.

This is straight, one-size-fits-all, open-source, Berkeley-issued stock applications, running on the same hardware. The simplest, clearest, most direct, fairest and best comparison I can think of. From the pages I linked earlier:

Windows:
CPU time [b][size=15][color=red]10453.421875[/color][/size][/b]
stderr out <core_client_version>5.10.30</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_enhanced 5.27 DevC++/MinGW

WU true angle range is :  0.396037

Claimed credit 72.5075475775156 

Linux:
CPU time [b][size=15][color=red]15575.432173[/color][/size][/b]
stderr out <core_client_version>5.10.28</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_enhanced 5.28 Revision: 26 g++ (GCC) 4.1.2 (Ubuntu 4.1.2-0ubuntu4)
libboinc: BOINC 6.1.0

WU true angle range is :  0.396111

Claimed credit 72.5059502526259

Now, can we address the issue, or do you want me to stamp

LINUX is slower than a tortoise

below your post, every time you point out how much cheaper it is than Windows?
ID: 724310 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9424
Credit: 7,359,571
RAC: 225
United Kingdom
Message 724386 - Posted: 11 Mar 2008, 0:01:08 UTC - in response to Message 724310.  
Last modified: 11 Mar 2008, 0:01:47 UTC

Now, can we address the issue, or do you want me to stamp

LINUX is slower than a tortoise

below your post, every time you point out how much cheaper it is than Windows?

I've never emphasized about how Linux is "cheaper" than Windows, in whatever way.

I go with what is more fit for purpose or 'better' in whatever way.

So, why do you think you got the results you have for "Linux takes 1.5 times longer than Windows"?


Happy crunchin',
Martin
See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 724386 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 12194
Credit: 125,038,160
RAC: 38,970
United Kingdom
Message 724389 - Posted: 11 Mar 2008, 0:10:53 UTC - in response to Message 724386.  

Now, can we address the issue, or do you want me to stamp

LINUX is slower than a tortoise

below your post, every time you point out how much cheaper it is than Windows?

I've never emphasized about how Linux is "cheaper" than Windows, in whatever way.

I go with what is more fit for purpose or 'better' in whatever way.

So, why do you think you got the results you have for "Linux takes 1.5 times longer than Windows"?


Happy crunchin',
Martin

I haven't a clue. I've never run Linux in my life. That's why I'm asking someone who I had, until now, assumed knew something about the issue he posts so frequently about.

And no, they're not my results. They're results from the one and only, dual-boot, Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [Family 6 Model 15 Stepping 11], owned by DeMus, the originator of this thread - as you would have seen if you'd read any of the preceding posts or followed any of the links I've so painstakingly prepared.
ID: 724389 · Report as offensive
Gerry Green

Send message
Joined: 15 May 99
Posts: 13
Credit: 3,034,952
RAC: 0
United States
Message 724393 - Posted: 11 Mar 2008, 0:21:55 UTC - in response to Message 724310.  

Did it match what we're seeing here?

Linux: over 15,000 seconds for 72 credits
Windows: 10,000 seconds for 72 credits

Sorry, there's no direct comparison to be made. You now have a very different client. doing a greater depth of analysis, on much more recent hardware.

OK

Forget the ancient history. This is the here-and-now.

And forget fancy (paid for) compilers, with fancy (anticompetitive) vendor strings.

This is straight, one-size-fits-all, open-source, Berkeley-issued stock applications, running on the same hardware. The simplest, clearest, most direct, fairest and best comparison I can think of. From the pages I linked earlier:

Windows:
CPU time [b][size=15][color=red]10453.421875[/color][/size][/b]
stderr out <core_client_version>5.10.30</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_enhanced 5.27 DevC++/MinGW

WU true angle range is :  0.396037

Claimed credit 72.5075475775156 

Linux:
CPU time [b][size=15][color=red]15575.432173[/color][/size][/b]
stderr out <core_client_version>5.10.28</core_client_version>
<![CDATA[
<stderr_txt>
setiathome_enhanced 5.28 Revision: 26 g++ (GCC) 4.1.2 (Ubuntu 4.1.2-0ubuntu4)
libboinc: BOINC 6.1.0

WU true angle range is :  0.396111

Claimed credit 72.5059502526259

Now, can we address the issue, or do you want me to stamp

LINUX is slower than a tortoise

below your post, every time you point out how much cheaper it is than Windows?


I dual boot 64 bit openSUSE 10.3 and 64 bit Vista Ultimate on a quad core. My processor benchmarks are very close to the original post (Linux 4787/1741 and Windoze 7111/2270). 72 units of credit takes about 11k seconds on windoze but 16.5k on linux.

The task correction factors aren't too dissimilar (.29 window v.s. .34 linux).

Neither side of the machine does very much except BOINC. Task Manager on windows / top on linux both indicate that BOINC tasks are getting 100% of all 4 cores. I do see beagled show up on linux but it grabs very little cpu.

This maybe anecdotal but the processor fan speed runs up quite a bit under windoze while BOINC is active. BOINC alone never causes the fan to kick into high speed under linux (I know because I almost always use the machine under linux and tonight under windoze the extra noise is driving me nuts...). On the linux side a long running query against the Oracle database will cause the fan to speed up just like it does under vista. Seems to imply there's "more to give" that BOINC alone isn't using under linux?

(Oh and both sides have oracle installed and running -- I don't use it all that much on either side and I doubt it has any impact on this...)
ID: 724393 · Report as offensive
Profile ML1
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 9424
Credit: 7,359,571
RAC: 225
United Kingdom
Message 724395 - Posted: 11 Mar 2008, 0:22:24 UTC - in response to Message 724389.  

I haven't a clue. I've never run Linux in my life. That's why I'm asking someone who I had, until now, assumed knew something about the issue he posts so frequently about.

And no, they're not my results. They're results from the one and only, dual-boot, Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [Family 6 Model 15 Stepping 11], owned by DeMus, the originator of this thread...

Sorry, no time to trace back through all the detail...

So I've to do your research and thinking for you?...

Linux does have a reputation of giving a lower overhead than Windows, and of surviving much lower spec hardware.

Just like Windows, Linux can be slugged to a standstill by other effects. Two good ones for that with (K)Ubuntu is if 'powersaving' or 'Cool n Quiet' might be enabled or if file indexing (Beagle or whatever) is enabled.


Now, try a dual boot and compare for yourself?

If you're a Windows fan, then Kubuntu is likely the best bet. However, when comparing remember that it isn't Windows...

Good luck,
Martin

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 724395 · Report as offensive
1 · 2 · 3 · 4 . . . 5 · Next

Message boards : Number crunching : Windows vs. Linux


 
©2018 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.