Instructions for parallel structure calculation

In the Python interface parallel structure calculation is
transparently handled by the simulationTools.StructureLoop class. No
modification of the script is necessary. The pdbTemplate argument of
the StructureLoop constructor should be a string which includes the
literal STRUCTURE, which is replaced by a different number for each
structure. An example can be found in eginput/gb1_rdc/anneal.py

Parallel structure calculation in the XPLOR interface is more
involved. Scripts should be modified along the lines of
eginput/protG/anneal.inp. One should access the environment variables
XPLOR_PROCESS and XPLOR_NUM_PROCESSES, and make sure structure files
do not collide and that unique seed values are used for each
structure.

Parallel execution must be specified using flags to the top-level
xplor (or pyXplor) command. For distributed machines the appropriate
incantation is

   xplor -py -parallel -machines machines -o logfile.out script.py

where machines is a quoted list or the name of a file specifying host
names on which to execute the program. If a host has more than one
processor, its name can be listed multiple times. Each process writes
a output file to logfile.out_#. Requirements for proper operation are
that the current directory is on a filesystem shared by all hosts,
accessible using the same path, and that password-less login is
possible. By default ssh is used for login, but rsh can be used if one
sets the XPLOR_RSH environment variable to rsh before execution.

If the local computer has multiple CPUs, a simplified version
incantation is 

   xplor -py -smp # -o logfile.out script.py

where # is the number of processors to use. 

Additional parallelism is available when refining against an ensemble
of structures. Calculation for separate ensemble members can be
carried out almost independently- spread over NUM processers when the
-num_threads NUM command-line option is given. These processors are on
he same machine. The number of processors used in this way should
usually be an even divisor of the ensemble size. Both structure and
ensemble-based parallelization could be specified with the following
invocation:

 xplor -py -parallel -machines machine.list -num_threads 2 -o out.log file.py

Specifying that an ensemble calculation would be run on processors of
each node specified in machine.list.