Speaker
Dr
Alessandro Costantini
(University of Perugia)
Description
The increasing availability of computer power on distributed
platforms
makes it easier to perform molecular based simulations of
complex systems
in order to reproduce their macroscopic properties. In this
report we outline
the work carried out in our laboratory by performing a
molecular dynamics
simulation of the Propane bulk system in liquid and gas phase.
The calculations were aimed at estimating in an a priori
fashion some thermodynamics
properties of the system and to build the related phase
diagram [1]. To
this end the DL_POLY [2] software package was used for the
npt statistical
ensamble at different temperatures. Computational tasks were
distributed
using the EGEE-Grid platform [3].
To obtain preliminary indications on the performance of
the used platform,
a test case was run on six different EGEE-Grid clusters. In
order to
evaluate the elapsed time of each simulation and the related
speed-up for
each cluster, we ran the calculations sequentially on one
node and in parallel
on 2 and 4 nodes. Measured elapsed times and speedups will be
illustrated at the meeting. The parallel performance of some
clusters of the
EGEE-Grid is very close to the ideal value due to
their dedicated usage. Deviations from it occurring in the
other clusters are
mainly due to the time sharing regime adopted by them. This
means that
the parallel performances of each EGEE-Grid cluster strictly
depend on the
adopted regime. In order to evaluate more in detail the
parallel performance
of each cluser and the waiting time intercurring between the
scheduling and
the running of a process we restricted parallel calculations
to two nodes. To
carry out a statistical analysis we ran 50 parallel jobs. As
apparent from the table,
more than 70% of the jobs ran properly and only 26% was
aborted.
Abortion is due for 62% to comunications errors between the
nodes of the same cluster, for 23% to internal errors of DL
POLY occurred
during the running and for 15% to the scheduler.
The calculated value of the density of the system gets
closer to the data
given in literature (582 Kg m(E-3) at T=230 K at P=1.013 bar
[4]) when
going from 200 K to 230 K. At the same time we computed
high values of pressure and correlated statistical errors.
Possible ways out
of this problem can be an increase of the simulation time in
order to obtain
a better average value of the pressure and/or a small
modification of the
force field of the system in order to normalize the effect
of the pressure on
the system.
References
[1] Costantini, A., Lagana', A., Pirani, F.: Lecture Notes
in Computer Sci-
ence 3980 (2006) 738-713.
[2] Smith, W., Forester, T.R.: DL POLY2: a general-purpose
parallel
molecular dynamics simulation package. J. Mol. Graph. 14
(3) (1996)
136-141
[3] Storchi, L., Manuali, C., Gervasi, O., Vitillaro, G.,
Lagan, A., Tarantelli,
F.: Lecture Notes in Computer Science 2658 (2003) 297-306.
[4] Air liquid group website: http://www.airliquide.com
Author
Dr
Alessandro Costantini
(University of Perugia)
Co-authors
Prof.
Antonio Laganà
(University of Perugia)
Prof.
Osvaldo Gervasi
(University of Perugia)
