In this video from the Dell EMC HPC Community meeting in Austin, Thomas Lippert from the Jülich Supercomputing Centre describes their pending 5 Petaflop Booster system.
This will be the first-ever demonstration in a production environment of the Cluster-Booster concept, pioneered in DEEP and DEEP-ER at prototype-level, and a considerable step towards the implementation of JSC’s modular supercomputing concept,” explains Prof. Thomas Lippert, Director of the Jülich Supercomputing Centre.
Dell EMC plans to develop and deploy a next-generation modular supercomputing system at the Jülich Supercomputing Centre. Leveraging the experience and results gained in the EU-funded DEEP and DEEP-ER projects, in which three of the partners have been strongly engaged, the group will develop the necessary mechanisms required to augment JSC’s JURECA cluster with a highly-scalable component named “Booster” and being based on Intel’s Scalable Systems Framework (Intel SSF).
Modular supercomputing is a new paradigm directly reflecting the diversity of execution characteristics, found in modern simulation codes, in the architecture of the supercomputer. Instead of a homogeneous design, different modules with distinct hardware characteristics are exposed via a homogeneous global software layer that enables optimal resource assignment.
Code parts of a simulation that can only be parallelized up to a limited concurrency level stay on the Cluster – equipped with faster general-purpose processor cores – while the highly parallelizable parts are to run on the weaker Booster cores but at much higher concurrency. In this way increased scalability and significantly higher efficiency with lower energy consumption can be reached, addressing both big data analytics and Exascale simulation capabilities.
The JURECA Booster will use Intel Xeon Phi 7250-F processors with on-package Intel Omni-Path Architecture interfaces. The system will be delivered by Intel with its subcontractor Dell, utilizing Dell’s PowerEdge C6230P servers. Once installed, it will provide a peak performance of 5 Petaflops. The system was co-designed by Intel and JSC to enable maximum scalability for large-scale simulations.
The JURECA Booster will be directly connected to the JURECA cluster, a system delivered by T-Platforms in 2015, and both modules will be operated as a single system. As part of the project a novel high-speed bridging mechanism between JURECA’s InfiniBand EDR and the Boosters’ Intel Omni-Path Architecture interconnect will be developed by the group of partners. Together with the modularity features of ParTec’s ParaStation ClusterSuite, this will enable efficient usage of the whole system by applications flexibly distributed across the modules.
ParaStation MPI software from ParTec will be used to keep the new Booster system running efficiently. The resiliency-related extensions of ParaStation MPI support fine-grained recovery of parallel application tasks.
The next Dell EMC HPC Community meeting takes place June 22 in Frankfurt, Germany.
