"O" for Linux Wanted!!

George M. Sheldrick gsheldr at shelx.uni-ac.gwdg.de
Tue Oct 14 02:52:21 EST 1997

We use XtalView, CCP4, Denzo/Scalepack, the Cambridge Structural
Database and SHELX (amongst other programs) extensively under Linux and
are very happy with their performance. A good hardware configuration is
a dual PP200 with 4MB Millenium graphics card and 64MB RAM, e.g. from
DELL (128MB is recommended if you plan to use refmac and it will
probably also speed up Denzo; Linux currently supports up to 16 CPUs
simultaneously, but 4 is the hardware limit for the PPro).  In the case
of SHELX, which I have tested most carefully and which is very
FPU-intensive, there were no precision problems or compiler errors with
the GNU g77 compiler, but both the Absoft and Siemens/Nixdorf FORTRAN
compilers gave minor (primarily optimization) problems.
Particularly impressive is the performance of the CSD under Linux
relative to IRIX, which I attribute to its better I/O and NFS support.
Although the CCDC took the decision to port it to Linux after intense
user pressure at the Seattle IUCr Meeting in August 1996, they included
it in their official September 1996 release, and we have not been able
to find any bugs in that Linux version.  It is clearly very clean
X-Windows code.
Although we were pushed into using XtalView because one of our SGIs died
and the (Linux PC) replacement was cheaper than the repair would have
been (and much faster too), we don't regret it. XtalView is much easier
for students to learn and handles disorder and anisotropic temperature
factors correctly (i.e. it can actually understand a standard Brookhaven
PBD file!!), and it interfaces well to SHELX, so we now use XtalView for
nearly all our routine work.  However we still keep O - in expert hands
the O macros are incredibly powerful (but incredibly frustrating if you
are using the wrong version), and the electron density plots from O are
a little more beautiful (the ones from XtalView can however be improved
by hand-editing the Postscript file).
Although the Linux kernel and much of the GNU software are of a standard
rarely found in commercial software, some of the other code that is
bundled with Linux may be at an early stage of development and still
needs to undergo the process of mutation and Darwinian selection.  I was
not impressed by the (Open)GL support (an extra software layer on top of
X-Windows) and suspect that this is the real problem in porting O (I can
sympathize - I have the same problem with a little program that I once
wrote for small molecule crystallographers called XP).
George Sheldrick

More information about the Xtal-log mailing list