Section: Partnerships and Cooperations

International Initiatives

The Joint Laboratory on Extreme-Scale Computing is jointly run by Inria, UIUC, ANL, BSC, JSC and RIKEN/AICS. It has ben created in 2014 as a follow-up of the Inria-UIUC JLPC, the Joint Laboratory for Petascale Computing.

The KerData team is collaborating with teams from ANL and UIUC within this lab since 2009 on several topics in the areas of I/O, storage and in situ processing and cloud computing. This collaboration has been initially formalized as the Data@Exascale Associate Team with ANL and UIUC (2013–2015) followed by Data@Exascale 2 Associate Team with ANL (2016–2018).

Since 2015, Gabriel Antoniu serves as a topic leader for Inria for the I/O, Storage and In Situ Processing topic.

• Project Acronym: Data@Exascale 2.

• Project Title: Convergent Data Storage and Processing Approaches for Exascale Computing and Big Data Analytics.

• International Partner:

  • Argonne National Laboratory (United States) — Mathematics and Computer Science Division (MCS) — Rob Ross.

• Start year: 2013.

• URL: http://www.irisa.fr/kerdata/data-at-exascale/.

• Description: In the past few years, countries including United States, the European Union, Japan and China have set up aggressive plans to get closer to what appears to be the next goal in terms of high-performance computing (HPC): Exaflop computing, a target which is now considered reachable by the next-generation supercomputers in 2020-2023. While these government-led initiatives have naturally focused on the big challenges of Exascale for the development of new hardware and software architectures, the quite recent emergence of the Big Data phenomenon introduces what could be called a tectonic shift that is impacting the entire research landscape for Exascale computing. As data generation capabilities in most science domains are now growing substantially faster than computational capabilities, causing these domains to become data-intensive, new challenges appeared in terms of volumes and velocity for data to be stored, processed and analyzed on the future Exascale machines.

  To face the challenges generated by the exponential data growth (a general phenomenon in many fields), a certain progress has already been made in the recent years in the rapidly-developing, industry-led field of cloud-based Big Data analytics, where advanced tools emerged, relying on machine-learning techniques and predictive analytics.

  Unfortunately, these advances cannot be immediately applied to Exascale computing: the tools and cultures of the two worlds, HPC (High-Performance Computing) and BDA (Big Data Analytics) have developed in a divergent fashion (in terms of major focus and technical approaches), to the detriment of both. The two worlds share however multiple similar challenges and unification now appears as essential in order to address the future challenges of major application domains that can benefit from both.

  The scientific program we propose for the Data@Exascale 2 Associate Team is defined from this new, highly-strategic perspective and builds on the idea that the design of innovative approaches to data I/O, storage and processing allowing Big Data analytics techniques and the newest HPC architectures to leverage each other clearly appears as a key catalyst factor for the convergence process.

• Title: DataCloud@Work.

• International Partner:

  • Polytechnic University of Bucharest (Romania), Computer Science Department, Nicolae Tapus and Valentin Cristea.

• Duration: 4 years.

• Start year: 2013. The status of IIP was established right after the end of our former DataCloud@work Associate Team (2010–2012).

• URL: https://www.irisa.fr/kerdata/doku.php?id=cloud_at_work:start.

• Description: Our research topics address the area of distributed data management for cloud services, focusing on autonomic storage. The goal is explore how to build an efficient, secure and reliable storage IaaS for data-intensive distributed applications running in cloud environments by enabling an autonomic behavior.

• National University of Singapore (NUS): We collaborate on resource management for workflows in the cloud and optimizing graph processing in geo-distributed data-centers.