Prevalence of HIV among adults per country at the end of 2005. Images courtesy of Wikipedia

DEISA, the Distributed European Infrastructure for Supercomputing Applications, and GridAustralia-APAC, joined by Monash University, have demonstrated the interoperation of their HPC infrastructures with distributed simulations in both continents on the effectiveness of drugs against mutants of the Human Immunodeficiency Virus.

Both DEISA and GridAustralia-APAC use different, incompatible underlying main middleware platforms; however, they were able to spread the compute tasks associated with HIV drug simulations over their different grids at a recent demonstrate at the SC07 conference in Nevada, U.S.

The simulations aim to assess the drug-binding energies of anti-HIV drugs bound to HIV protease, including drug-resistant versions. It is hoped that this technique can give an accurate assessment of the likely effectiveness of each antiviral drug with respect to any given HIV mutant strain. Given this information, patients who develop HIV drug resistance can be prescribed the most effective drug for their HIV strain while reducing the danger of further resistance enhancements.

In order to assess the drug binding interactions, each drug and respective HIV mutant enzyme is run as a short molecular dynamics simulation to obtain an averaged energy of interaction between the drug and the HIV protease strain. The huge number of calculations required is well suited for distributed processing, and time-to-solution can be significantly reduced by simultaneously employing several supercomputers in a suitable HPC grid.

Technical overview

DEISA is based on UNICORE 5 as far as job submission is concerned, while APAC makes use of the Globus Tool Kit. Following different approaches in job management, Globus and UNICORE are not interoperable in their currently established versions. In addition to DEISA’s option of data management via a continental global file system, however, both infrastructures support data transfer via GridFTP, usable both in Globus and in UNICORE.

Input data sets were provided in Australia by the Australian Research Group on an APAC storage server. Series of workflow jobs were submitted as shell scripts at the client site both to DEISA and to APAC through infrastructure specific interfaces, using DESHL for DEISA (a UNICORE command line tool), and Globus WS-Gram client for APAC. The parts of the input data required by each job were automatically moved to the respective DEISA or APAC execution sites in Europe or Australia via GridFTP.

A number of simulations were run with the drug Amprenavir (Apv) acting on various HIV strains. The trajectory data was post-processed to measure the energy of interaction between the drug and each HIV strain. The simulation results were later automatically uploaded on that APAC storage server for post-processing and visualization by the researchers.

By this transparent linking of compute resources in Australia and in Europe, and by offering reliable, automated bidirectional data transfer between both infrastructures, this project-oriented interoperation of DEISA and APAC could successfully be demonstrated for the first time

- DEISA