COMP Superscalar (COMPSs)

A task-based programming models that is able to parallelize applications for distributed platforms.

Institution:

Research Group:

Workflows and Distributed Computing

Researcher/s:

Rosa Maria Badia, Javier Conejero, Raül Sirvent, Francesc Lordan, Daniele Lezzi

Description:

COMPSs lets developers annotate tasks in Python/Java; at run-time it builds a dependency graph and places tasks automatically on any mix of CPUs + GPUs across clusters or clouds. By abstracting the hardware it tackles the “single code-base across vendors” pain the portability trend highlights. The same abstraction is useful upstream and downstream of AI mega-model training pipelines, letting teams reuse workflow code when the underlying accelerator mix changes.

Value Proposition:

One code, any cluster; automates parallel execution and scaling.

Aplication areas:

Biomedical pipelines, engineering simulations, biodiversity analytics, chemistry, astrophysics, finance, telecom, manufacturing, earth-science modelling

Target market:

Public & sovereign super-computing centres needing multi-vendor portability; Cloud-HPC providers bundling managed workflow services; Independent Software Vendors (ISVs) whose customers deploy on diverse infrastructures; Large enterprise R&D groups (pharma, automotive, energy) building in-house HPC/AI stacks

Technology Readiness Level (1-9): 7

Protection:

Open source (apache v2)

More information

if you want to know more about this project do not hesitate to contact us