same simulation win/linux different results.... why?

[OJAG]

Dear All,

I checked this http://www.neuron.yale.edu/phpBB/viewto ... inux#p5758
but I already modify what possible and I am still getting the same problem: My simulation produces different results running on windows or linux. The only factor I could suspect of as culprit is that I am using random numbers to set the amplitude of some current inputs onto the cells and another to create my connectivity, of course also with use_mcell_ran4(1),

could any of you tell me if the random generator can produce that behavior, from windows to linux, changing results?
I ran the same simulation in versions 5.8 5.9 7.0 and 7.1 on windows xp (32bit),7(64bit) (precompiled version), producing always the same result but on ubuntu (X_86_64bit) it fully changed..

Thanks in advance for ANY advice

OJAG

[ted]

You have the code, so you have the solution at hand. Why don't you insert print statements into your program so that every time your program picks a random number, it prints the value of that number? Then you can answer your own question, and tell everybody else what you found.

[OJAG]

Thanks for the tip Ted,

My results were as follows

winXP NEURON7.0 //// UBUNTU X86_64 /// win_7_64bit NEURON7.1
r=0.64354531 - r=0.82241427 - r=0.64354531
r=0.16762013 - r=0.008886788 - r=0.16762013
r=0.36580191 - r=0.15383027 - r=0.36580191
r=0.25242554 - r=0.63635059 - r=0.25242554
r=0.92979875 - r=0.84966693 - r=0.92979875
r=0.8741479 - r=0.6223758 - r=0.8741479
r=0.39690984 - r=0.5343043 - r=0.39690984
r=0.46282246 - r=0.8210683 - r=0.46282246
r=0.95720508 - r=0.15385635 - r=0.95720508
r=0.74672166 - r=0.52208868 - r=0.74672166

According to this and if I don't misunderstand the result, the constructor of the generator uses differently the seed to generate the numbers stream on windows and on linux and it doesn't depend on the 32 or 64-bit architecture. Is that right Ted? if so/not could you post here some references about this point?

Thanks a lot,

OJAG

[ted]

Very strange. I just tried this very simple program

Code: Select all

objref r
r = new Random()
print r.uniform(0,1)
for i=1,10 print r.repick()

under Centos 5.5 (32 bit, Linux), Windows 2000 (32 bit), and, on a different machine, Windows 7 (64 bit), and got the same results each time:
0.5340179
0.36924247
0.65320583
0.63449432
0.35163532
0.92944949
0.30661334
0.0043154789
0.85162027
0.8414458
0.90359852

Then I switched to the MCellRan4 generator by inserting
r.MCellRan4()
right after the
r = new Random()
statement, and this time the result was
0.86599038
0.004729443
0.023549526
0.80911988
0.030454614
0.88158147
0.66938499
0.62269264
0.16661648
0.30357997
0.4439619
under all 3 operating systems.

So whereever the bug is--in NEURON, or in your hoc code--it's probably very subtle. If you email me just the files that are necessary to reproduce the problem, I'll see what I can find out.
ted dot carnevale at yale dot edu

[ted]

Trying your code with NEURON under Win 2K (32 bit), Win 7 (64 bit), and 32 bit CentOS 5.5 I get the same results as you did with Win XP and 7. But with NEURON under 64 bit CentOS 5.5 the results were the same as you got with 64 bit Ubuntu.