Hi, I'm trying to get a grip on what seems to be a multitude of options for running Phenix in multiprocessor/parallel mode. Some of the options/keywords are obvious and easy to find in documentation but other elude me and the posts on the phenixbb make the whole issue even more complicated. So, here is my humble request: Dear Phenix Developers could you please prepare a short summary containing the info concerning running Phenix in multiprocessor/parallel environment for both stable and self compiled developer versions? Even a table with required keywords per program would be a great help... Best regards, Andrzej Lyskowski -- Dr. Andrzej Lyskowski (andrzej[.]lyskowski[@]acib[.]at) ACIB Austrian Centre of Industrial Biotechnology GmbH Petersgasse 14, A-8010 Graz, Austria http://acib.at/ c/o IMB Institute of Molecular Biosciences, University of Graz Humboldtstraße 50,III/194, A-8010 Graz, Austria http://strubi.uni-graz.at/ +43 316 380 1993 tel +43 316 380 9897 fax
On Wed, Mar 23, 2011 at 5:27 AM, Jendrek
I'm trying to get a grip on what seems to be a multitude of options for running Phenix in multiprocessor/parallel mode. Some of the options/keywords are obvious and easy to find in documentation but other elude me and the posts on the phenixbb make the whole issue even more complicated.
Dear Phenix Developers could you please prepare a short summary containing the info concerning running Phenix in multiprocessor/parallel environment for both stable and self compiled developer versions? Even a table with required keywords per program would be a great help...
I will try to throw something together later this week. However, the short answer is that only AutoBuild and a handful of smaller programs (find_tls_groups, and the structure comparison GUI) have parameters to make them run in parallel. phenix.refine and Phaser can be parallelized at a very low level by compiling with OpenMP support, but this requires building Phenix from source, and is incompatible with the GUI. -Nat
Is it possible to incorporate (separate) OpenMP builds in future nightlies? I'm a command line person so I would benefit from this. F On Mar 23, 2011, at 9:57 AM, Nathaniel Echols wrote:
phenix.refine and Phaser can be parallelized at a very low level by compiling with OpenMP support, but this requires building Phenix from source, and is incompatible with the GUI.
--------------------------------------------- Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder gpg --keyserver pgp.mit.edu --recv-keys 67BA8D5D 8AE2 F2F4 90F7 9640 28BC 686F 78FD 6669 67BA 8D5D
On Wed, Mar 23, 2011 at 9:12 AM, Francis E Reyes
Is it possible to incorporate (separate) OpenMP builds in future nightlies?
This may be difficult to support, since it is only possible with relatively recent versions of GCC (or Intel's C++ compiler, which few people will have installed). It is effectively impossible on Mac, since the versions of Xcode that we need to use have no OpenMP support whatsoever, and even in XCode 3.2 or 4, the bundled GCC has a (according to Ralf) broken implementation. (You can build and run a newer version of GCC with functional OpenMP on Mac, but we don't like distributing Phenix binaries that depend on non-standard software being already installed.) I'm not sure what the situation on Linux is, although it looks like Fedora 11 and Ubuntu 10.04 both have GCC 4.4.
I'm a command line person so I would benefit from this.
If your lab is still hanging on to its G5 Macs, you'd benefit far more from throwing those out and buying a multicore Intel/AMD system. -Nat
participants (3)
-
Francis E Reyes -
Jendrek -
Nathaniel Echols