AutoTuning and Adaptivity appRoach for Energy efficient eXascale HPC systems

Projektinformationen

ANTAREX

ID Finanzhilfevereinbarung: 671623

DOI

10.3030/671623

Projekt abgeschlossen

EK-Unterschriftsdatum 29 Juni 2015

Startdatum 1 September 2015

Enddatum 30 November 2018

Finanziert unter

• EXCELLENT SCIENCE - Future and Emerging Technologies (FET)

Gesamtkosten

€ 3 115 251,25

EU-Beitrag

€ 3 115 251,00

Koordiniert durch

POLITECNICO DI MILANO
Italy

Periodic Reporting for period 2 - ANTAREX (AutoTuning and Adaptivity appRoach for Energy efficient eXascale HPC systems)

Berichtszeitraum: 2017-03-01 bis 2018-11-30

Zusammenfassung vom Kontext und den Gesamtzielen des Projekts

The ANTAREX research project, coordinated by prof. Cristina Silvano from Politecnico di Milano, has been granted in the H2020 Future and Emerging Technologies programme on High Performance Computing. The project involves CINECA, the Italian Tier-0 Supercomputing Centre and IT4Innovations, the Czech Tier-1 Supercomputing Center. The Consortium also includes three top-ranked academic partners (ETH Zurich, University of Porto and INRIA), one of the Italian leading biopharmaceutical companies (Dompé) and the top European navigation software company (Sygic). Being one of the nineteen research projects in FET-HPC-2014, ANTAREX brings its partners on the forefront of the European research in HPC.

The main goal of the ANTAREX project is to provide a breakthrough approach to map, runtime manage and autotune applications for green and heterogeneous HPC systems up to the Exascale level.
The ground-breaking objectives of the project can be summarized as follows:

1. Autotuning. The proposed system-wide holistic approach will offer monitoring and autotuning capabilities to HPC applications.

2. Programming models and languages for self-adaptivity. One key innovation of the proposed approach consists of introducing a separation of concerns (self-adaptivity strategies vs. application functionalities) promoted by the definition of a Domain Specific Language (DSL) inspired by aspect-oriented programming concepts for heterogeneous systems.

3. Exploiting heterogeneous computing resources in green HPC platforms. The project will develop a software stack to support the smart integration of heterogeneous resources, while optimizing the energy efficiency and thermal profile.

Synergistic activities in application development, programming models, and runtime management proposed in the project have the potential to advance the state-of-the-art and provide the technology needed to reap the benefits of scalability, heterogeneity and dynamicity in future supercomputing platforms for HPC system up to the Exascale level.

Arbeit, die ab Beginn des Projekts bis zum Ende des durch den Bericht erfassten Berichtszeitraums geleistet wurde, und die wichtigsten bis dahin erzielten Ergebnisse

ANTAREX MAIN SIGNIFICANT ACHIEVEMENTS:


1. Compiler prototype (WP2 - Milestone MS4). Specifically, Clava, a source to source (C/C++ to C/C++) compiler, was entirely developed in WP2. It includes an aspect-oriented programming approach, implemented by an internal weaver and the technology provided by the LARA DSL, in order to describe source-to-source strategies, such as code transformations and code instrumentation.

2. Libraries of LARA strategies and DSL Runtime (WP2 – Milestone MS5). Another main achievement was the development of LARA DSL strategies to support the integration of the Examon and mARGOt APIs (developed in WP3), the integration of OpenCL kernels, the experiments with the support for MPI, the support of auto-parallelization via OpenMP, and to address the extra-functional requirements.

3. mARGOt application autotuning framework (WP3 - Milestone MS8) This achievement consists of the release of the final version of the mARGOt application autotuning framework. The mARGOt framework has been released as open-source and successfully integrated with the final prototypes of the two ANTAREX Use Cases.

4. Countdown power optimization framework (WP3 - Milestone MS9) Besides the release of the open-source Examon monitoring framework at the end of RP1 (successfully deployed on the 64-node Galileo system currently in production at CINECA), this achievement consists of the release of the Countdown power optimization framework. COUNTDOWN is a methodology and a tool for identifying and automatically reducing the power consumption of the computing elements during communication and synchronization primitives, filtering out phases which would impact the time to solution of the application.

5. Integrated and Validated Drug Discovery System Prototype (WP4 – Milestone MS12) A runtime tuneable version of the molecular docking application was developed for using in virtual screening experiments for the biopharmaceutical company Dompé and scaled out to the 10.4 petaFLOPS Marconi supercomputer at CINECA.

6. Integrated and Validated Navigation System Prototype (WP5 – Milestone MS15) A self-adaptive navigation system was developed by Sygic and IT4Innovations to mitigate traffic congestion in smart cities, using an innovative algorithm of road-balanced routing and deployed on the server-side at the IT4Innovations Czech supercomputing centre.

7. Awareness of final project foreground (WP6 -- Milestone MS18): Dissemination and exploitation activities generated awareness about the project outcomes externally to the Consortium to generate a significant impact.

Fortschritte, die über den aktuellen Stand der Technik hinausgehen und voraussichtliche potenzielle Auswirkungen (einschließlich der bis dato erzielten sozioökonomischen Auswirkungen und weiter gefassten gesellschaftlichen Auswirkungen des Projekts)

The main goal of the ANTAREX project is to provide a breakthrough approach to map, runtime manage and autotune applications for green and heterogeneous High Performance Computing systems up to the Exascale level. The goal is to support the parallelism, scalability and adaptability of a dynamic workload by exploiting the full system capabilities (including energy management) for emerging large-scale and extreme-scale systems, while reducing the Total Cost of Ownership (TCO) for companies and public organizations.

The ANTAREX project is driven by two use cases chosen to address the self-adaptivity and scalability characteristics of two highly relevant HPC application scenarios:

1. A runtime tuneable version of the molecular docking application was developed for using in virtual screening experiments for the biopharmaceutical company Dompé. This application was deployed and scaled out to the full size of the 10.4 petaFLOPS Marconi supercomputer (#19 in TOP500) at CINECA to screen a database containing a billion ligands – biochemical substances which bind to biological molecules –with the aim of targeting unresolved infective diseases. This represents the largest virtual screening experiment ever launched in terms of computational threads (up to one million) and size of the compound database (one billion ligands). Using the ANTAREX’s HPC technologies supporting autotuning, scalability and energy efficiency, Dompé is now able to optimize molecular docking to reduce the virtual screening process for the identification of new active compounds by two orders of magnitude.

2. A self-adaptive navigation system was developed by Sygic and IT4Innovations to mitigate traffic congestion in smart cities, using an innovative algorithm of road-balanced routing and deployed on the server-side at the IT4Innovations Czech supercomputing centre. Exploiting supercomputing power and ANTAREX code optimization technologies, we reach the point of being able to calculate routes for tens of thousands of drivers simultaneously and perpetually towards a global optimum, i.e. providing traffic-flow-optimized navigation to reduce total travel time. Using the ANTAREX technologies supporting autotuning and scalability, Sygic is now ready to adapt this innovative product for municipalities based on their specific cost/performance requirements.

ANTAREX Infographic

ANTAREX Group Photo at the Kickoff Meeting held in September 2015 at CINECA (Italy).