Science Gateway for the Serbian Condensed Matter Physics Community - Science Gateways for Distributed Computing Infrastructures - page 201

15.2 Building the Science Gateway

Before the AEGIS CMPC science gateway was built and put into production, the
default interface to DCI within the AEGIS CMPC was based on gLite technology
provided by the European Middleware Initiative (EMI 2014). This is a command
line interface, and it requires a personal certi cate issued by one of recognized

certi cate authorities, membership in a virtual organization (VO) that supports the

applications, and access to a user interface (UI) machine. A personal certificate is
obtainable from the national AEGIS CA, while the Serbian AEGIS National Grid
Initiative provides the VO membership service. A central UI machine is available at
the IPB, but within the community personal UI virtual machines are used as well.
In this approach, the use of AEGIS CMPC applications requires preparation of a
parameter input

file that describes the physical system of interest. In addition, the
DCI description of computational task has to be prepared as well. In a gLite- based
UI machine this is typically done using the Job Description Language (JDL).
Depending on machine architecture and the software stack provided by DCI in
terms of available compilers, a CMPC application could be compiled locally at a UI
machine or remotely at a DCI, on a target worker node. In the first case, a statically

linked executable is submitted to the prede ned architectures, while in the latter

case the source code is transferred to the computation node, where one of the
various predeployed compilers performs the compilation. A CMPC application with
a physical system description from the parameter input file and with a DCI

description of the computational task from the JDL file is submitted from the UI

machine to the DCI by the gLite command line interface. Further job management
(job cancelation, monitoring, and output retrieval) is done by the gLite command
line interface as well.
This approach demands several skills: knowledge of CMPC application source
codes and their parallelization techniques, knowledge of different compiler tech-
nologies, knowledge of Linux operating system and its common tools, and
knowledge of gLite command line interface and DCI technology. This variety of
skills, which is required for solving even trivial problems, creates a strong barrier
that prevents ready use of CMPC applications within the community, and adversely
affects attracting new people who are interested in numerical simulations based on
the available set of codes. Since the knowledge of the underlying physics (quantum
theory and Bose Einstein condensation) and understanding of CMPC applications

should be the only real requirements, the development of a scienti

-

c gateway

effectively solves the identi ed problems and enables new users to perform

numerical simulations in this field without much technical knowledge of the

operating system, compiler technology, low-level DCI interface, and DCI itself.
Furthermore, AEGIS CMPC SG promotes sharing of the results of simulations
within the community, and in this way increases understanding of the behavior of
particular physical systems of interest.
The desired level of automation of CMPC applications is achieved through the
use of WS-PGRADE/gUSE (gUSE 2014) workflows. For example, typical usage of