EASEA Massively Parallel Platform for Cloud Computing – EASEA-CLOUD

The aim of the EASEA-CLOUD project is to exploit the massively parallel resources that are offered by clusters or a grid of modern GPU-equipped machines in order to find solutions to inverse problems whose evaluation function can be intrinsically sequential.

Exploiting massively parallel machines is very difficult because they typically implement several (4) levels of parallelism. Therefore, Petaflop machines exist, but the programs that can run on them in an efficient way are very rare.

Fortunately, massive parallelization of generic sequential problems can be achieved by evolutionary computation, that can efficiently exploit the parallel evaluation of thousands of potential solutions (a population) for optimization or machine-learning purposes.

The project consists in turning the existing EASEA (EAsy Specification of Evolutionary Algorithms, lsiit.u-strasbg.fr/easea) research platform into an industrial-grade platform that could be exploited by running in “cloud” mode, on a large grid of computers (ISC-PIF/CREA is the current manager of the French National Grid).

The necessary steps are to develop:

• a professional-grade API, development environment and human-computer interface for the existing academic EASEA platform,

• cloud-management tools (in order to launch an experiment on a grid of computers, monitor the experiment and bill the laboratories or companies that will be using EASEA-CLOUD for intensive computation,

• novel visualisation tools, in order to monitor an evolutionary run, potentially launched on several hundred heterogeneous GPU machines.

The consortium is made of thee partners: LSIIT/UDS (which is developing the EASEA platform), ISC-PIR/CREA (for its experience in grid and cloud computing), AVIZ/INRIA (for its experience in visualization tools for evolutionary computation) and two sub-contractors: LogXLabs (a software development company in order to create industrial-grade code and interfaces) and BIOEMERGENCE-IMAGIF, the “valorisation” department of CNRS Gif s/Yvette. Valorisation will take place in strong collaboration with UNISTRA VALO, the valorisation structure of Université de Strasbourg.