Re: [Chimera-users] Chimera on x86_64?
Hi Ben, There's no technical obstacle to us making regular 64-bit linux Chimera distributions. The only issue is having time to do it. The work involves compiling the ~30 third party libraries that Chimera uses. I did this is the past and it took about one day of fiddling. Our Chimera build scripts compile all of the third party libraries automatically and my past hand fiddling to build on 64-bit linux would have to be put into our scripts. That could take a 1 to 5 days of work. We have a 64-bit linux machine to do the work on. So it all comes down to when one of the 4 developers (Tom Goddard, Eric Pettersen, Greg Couch, or Conrad Huang) has time to work on that. I don't think I will have time until January at the earliest. Could you give me a better sense of how much need there is in the Sali lab? Are you really saying you would not install 32-bit Linux system libraries except for Chimera? I would think there are a fair number of applications out there with binary distributions only for 32-bit. The other problem you point out is that 32-bit Chimera Python cannot use your own 64-bit Python modules. Are you using additional C Python modules within Chimera that do not come with Chimera? That is interesting -- what are they? Tom
From: Ben Webb <ben@salilab.org> To: Eric Pettersen <pett@cgl.ucsf.edu> Date: Wed, 29 Nov 2006 12:51:27 -0800 Subject: Chimera on x86_64?
I wonder if you could give me a rough idea of when you plan to make regular builds of Chimera on x86_64 (if at all)? All I've been able to glean from the download page and the mailing list archive is that you do make occasional builds, but recommend the 32-bit version.
Memory isn't so much of an issue for us in particular (although we are increasingly looking at large EM maps) but the requirement to have 32-bit libraries is starting to become a problem. All of our new desktops are x86_64 machines, and it's a pain to require the 32-bit libraries just for Chimera. Additionally, our users want to mix-and-match Python modules (including Modeller, of course) and Chimera is unable to import any module which has a C backend, since they're compiled for x86_64 and the Python interpreter in Chimera is 32-bit. (Of course, we could recompile all of our other Python modules for 32-bit, but that would cause problems with x86_64 apps.)
If you need people to test x86_64 Chimera builds, we certainly have some demanding users here to break them. ;)
Ben -- ben@salilab.org http://salilab.org/~ben/ "It is a capital mistake to theorize before one has data." - Sir Arthur Conan Doyle
Thomas Goddard wrote:
There's no technical obstacle to us making regular 64-bit linux Chimera distributions. The only issue is having time to do it. ... Could you give me a better sense of how much need there is in the Sali lab? Are you really saying you would not install 32-bit Linux system libraries except for Chimera? I would think there are a fair number of applications out there with binary distributions only for 32-bit.
Yes, of course there are a handful of other applications (mostly ancient and unsupported codes) which require a 32-bit glibc. But it's really only Chimera which needs all of the X libraries etc. to get pulled in.
The other problem you point out is that 32-bit Chimera Python cannot use your own 64-bit Python modules. Are you using additional C Python modules within Chimera that do not come with Chimera? That is interesting -- what are they?
We've already discussed using our MODELLER package in combination with Chimera. The next public release of MODELLER will primarily be made available as a Python module, and we are exploring the options for using this directly with Chimera (rather than having it generate input files and run a subprocess). This works just fine for 32-bit, but not for x86_64 unless we build Modeller as a 32-bit application. Given the large performance boost we've seen for our x86_64 builds vs. i386, we are rather reluctant to do that. Ben -- ben@salilab.org http://salilab.org/~ben/ "It is a capital mistake to theorize before one has data." - Sir Arthur Conan Doyle
participants (2)
-
Ben Webb
-
Thomas Goddard