Different simulation results under Windows and Linux?

Moderator: wwlytton

Post Reply
Skyfire79
Posts: 21
Joined: Mon Sep 27, 2010 5:37 pm

Different simulation results under Windows and Linux?

Post by Skyfire79 » Wed Sep 04, 2013 6:28 pm

Hi, Ted:

I have a strange question to ask you. I previously run my network simulation under Windows XP (a virtual machine under CentOS) and recently I started to run my simulations in the Linux CentOS. The NEURON versions for both systems are the same (V7.1). Surprisingly, I found the simulation results I got from XP and from Linux are different, for the identical code (exactly the same model). Later I found out the random spike trains generated by the NetStim object are different. However, even after I turned off these random background inputs, I still got different results for the same network model (I just copied the model over). I really got confused because I expect the results from Windows and Linux should be the same. Otherwise, which one is correct? Any explanations for such strange observations? If you want, I can send my model to you so you can test it.

Thank you very much!

Best regards,
Guoshi

ted
Site Admin
Posts: 5108
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Different simulation results under Windows and Linux?

Post by ted » Thu Sep 05, 2013 10:17 am

There have been lots of bug fixes between 7.1 and now. Could you try with the most recent (7.3) versions for MSWin and Linux and see if the difference disappears?

Skyfire79
Posts: 21
Joined: Mon Sep 27, 2010 5:37 pm

Re: Different simulation results under Windows and Linux?

Post by Skyfire79 » Thu Sep 05, 2013 11:23 am

Hi, Ted:

Thank you for your suggestions. I will update my NEURON to the latest version 7.3 and let you know if the difference disappears or not.

Best,
GL

Skyfire79
Posts: 21
Joined: Mon Sep 27, 2010 5:37 pm

Re: Different simulation results under Windows and Linux?

Post by Skyfire79 » Thu Sep 05, 2013 1:37 pm

Hi, Ted:

I have installed the latest version (v7.3) under the WinXP. When I run the hoc code, an error message window pops up:

Neuron.ext – Entry Point Not Found
The procedure entry point_except_handler4_common could not be located in the dynamic link library msvcrt.dll


After I click "OK", the code starts to run normally. Do you know why the error message comes out and any solution to fix it? Besides, I notice the results from the v7.3 simulation is different from those from the v7.1 simulation (all under WinXP).

Also, I tried to install V7.3 under the Linux system, but did not succeed. The following errors come out:

[root@grendel guoshi]# rpm --install nrn-7.3.x86_64.rpm
error: Failed dependencies:
libc.so.6(GLIBC_2.11)(64bit) is needed by nrn-7.3-rel849.x86_64
libc.so.6(GLIBC_2.14)(64bit) is needed by nrn-7.3-rel849.x86_64
libc.so.6(GLIBC_2.15)(64bit) is needed by nrn-7.3-rel849.x86_64
libc.so.6(GLIBC_2.7)(64bit) is needed by nrn-7.3-rel849.x86_64
libreadline.so.6()(64bit) is needed by nrn-7.3-rel849.x86_64
libstdc++.so.6(GLIBCXX_3.4.11)(64bit) is needed by nrn-7.3-rel849.x86_64
libstdc++.so.6(GLIBCXX_3.4.15)(64bit) is needed by nrn-7.3-rel849.x86_64
libstdc++.so.6(GLIBCXX_3.4.9)(64bit) is needed by nrn-7.3-rel849.x86_64
libtinfo.so.5()(64bit) is needed by nrn-7.3-rel849.x86_64



How can I install those missing components?

Thank you very much for your help!

Best,
GL

ted
Site Admin
Posts: 5108
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Different simulation results under Windows and Linux?

Post by ted » Thu Sep 05, 2013 6:16 pm

Skyfire79 wrote:I have installed the latest version (v7.3) under the WinXP.
Before you did this, did you completely uninstall the previous version of NEURON? The correct procedure is to use the Uninstaller that is part of the NEURON program group, then delete any C:\nrn* directory that may persist.
I tried to install V7.3 under the Linux system, but did not succeed. The following errors
First question is whether your Linux is 32 bit or 64 bit. If it's 32 bit, use the 32 bit rpm.

If that's not the problem, then the problem is with your libc.so. On my machine,
rpm -qa | grep glibc
returns
glibc-common-2.12-1.107.el6_4.4.i686
glibc-headers-2.12-1.107.el6_4.4.i686
glibc-2.12-1.107.el6_4.4.i686
glibc-devel-2.12-1.107.el6_4.4.i686
so my guess is that I wouldn't be able to use the rpm installer for NEURON if it really needs glibc 2.11.

I wouldn't advise trying to install additional versions of these files unless you're very sure that you're doing it in a way that doesn't break your Linux installation. Instead, first make sure that your Linux installation is up to date. Then try the rpm again. If it still complains of failed dependencies, you may have to install NEURON from source code. If you were able to compile mod files under NEURON 7.1, you probably have all the software you need to do that. Follow the instructions at
http://www.neuron.yale.edu/neuron/download
to get the gzipped tar files for Interviews and NEURON and compile them.

Skyfire79
Posts: 21
Joined: Mon Sep 27, 2010 5:37 pm

Re: Different simulation results under Windows and Linux?

Post by Skyfire79 » Fri Sep 06, 2013 6:20 pm

Hi, Ted:

Thank you for your suggestions. I have tried to completely uninstall NEURON again using the Uninstaller and re-install the latest version (v7.3) in WinXP, but the error message still came up. Anyway, the program seems to run normally after I clicked OK to the message.

I was able to install NEURON 7.3 under Linux using the source code as you suggested. Then I run the same network under both systems and compare the results. The good news is that the random spike trains generated from the NetStim object are identical now. However, the simulation results are still different. I simulated a total of 3 second network time. I compare the voltage data recorded from multiple cells and found the voltage is initially the same till about 1.1 sec. I really don't know why difference occurs after that. Can I send the network to you so you can test it?

Thanks again!

Best,
GL

ted
Site Admin
Posts: 5108
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

Re: Different simulation results under Windows and Linux?

Post by ted » Fri Sep 06, 2013 8:38 pm

Skyfire79 wrote:I have tried to completely uninstall NEURON again using the Uninstaller and re-install the latest version (v7.3) in WinXP, but the error message still came up. Anyway, the program seems to run normally after I clicked OK to the message.
I don't like that. Something is not right.
Can I send the network to you so you can test it?
Sure, just zip up the necessary hoc, mod, and ses files--no dll, please--and email to
ted dot carnevale at yale dot edu

Post Reply