21-27 March 2009
Prague
Europe/Prague timezone

User-friendly Parallelization of GAUDI Applications with Python

24 Mar 2009, 15:00
20m
Club A (Prague)

Club A

Prague

Prague Congress Centre 5. května 65, 140 00 Prague 4, Czech Republic
oral Software Components, Tools and Databases Software Components, Tools and Databases

Speaker

Dr Pere Mato (CERN)

Description

GAUDI is a software framework in C++ used to build event data processing applications using a set of standard components with well-defined interfaces. Simulation, high-level trigger, reconstruction, and analysis programs used by several experiments are developed using GAUDI. These applications can be configured and driven by simple Python scripts. Given the fact that a considerable amount of existing software has been developed using serial methodology, and has existed in some cases for many years, implementation of parallelization techniques at the framework level may offer a way of exploiting current multi-core technologies to maximize performance and reduce latencies without re-writing thousands/millions of lines of code. In the solution we have developed the parallelization techniques are introduced to the high level Python scripts which configure and drive the applications, such that the core C++ application code requires no modification, and that end users need make only minimal changes to their scripts. The developed solution leverages from existing generic Python modules that support parallel processing. Naturally, the parallel version of a given program should produce results consistent with its serial execution. The evaluation of several prototypes incorporating various parallelization techniques are presented and discussed.

Primary authors

Mr Eoin Smith (CERN) Dr Pere Mato (CERN)

Presentation Materials