Information Technology Reference
In-Depth Information
Setup search
38
MOLPAK
MOLPAK
200
200
DMAREL
DMAREL
DMAREL
DMAREL
Save &
visualise
Save &
visualise
Finish
FIGURE 8.6
Overview of the polymorph-search process.
In a nutshell, the chemists rely on two Fortran application programs to
perform this two-stage search. MOLPAK can generate up to 200 molecule
packings in accordance with all 38 packing types known to scientists
beforehand. Each of the packings will then be further optimized by
DMAREL to determine whether a hypothetical structure is thermody-
namically possible. During the execution of a full-scale search, a data vol-
ume in the region of 6 GB is usually produced. All MOLPAK and DMAREL
calculations are independent of each other with separate input i les, so full
parallelization can be engineered respectively for all 38 and 7,600 execu-
tions (although no DMAREL jobs should go forth without the preceding
MOLPAK calculation having been completed).
The initial system of the polymorph search was not built on the grids.
Instead, it relied on a dedicated high-end server. The idea was straight-
forward. Around the two executables, a number of shell scripts could be
coded to enable the automation of the search process, although the degree
of manual interactions required was still so high that PhD students or
chemical engineers were often assigned to maintain searches. Depending
on the individual molecule, a MOLPAK calculation normally takes a few
hours for completion, while in contrast a DMAREL optimization only
lasts a few seconds or, occasionally, a couple of minutes. Because of the
limited computing power and the lack of full automation and parallel
processing of the system, it could easily take a few weeks to run a full-
size search, which can be done now in hours with grid computing and
BPEL enabled. Later on, it was possible for a user interface to be imple-
mented for scientists to provide initial search parameters for the backend
Search WWH ::
Custom Search