SP7 - High Performance Computing
The overall goal of the High Performance Computing Platform subproject is to provide the HBP Consortium and the broader European neuroscience community with supercomputing, Big Data and Cloud capabilities at the exascale, as well as the system software, middleware, interactive computational steering and visualisation support necessary to create and simulate multi-scale brain models and to address the hard-scaling challenges of whole brain modelling.
The subproject will utilise innovative HPC technology as detailed in the Exascale Roadmap of the Strategic Research Agenda (SRA) of the European Technology Platform for High Performance Computing (ETP4HPC). In order to meet the specific challenging requirements of the HBP, the subproject will carry out additional specific research in hardware and software thus contributing to path-breaking ICT technology research. Key topics for research will include novel accelerator technologies addressing highly scalable computational challenges, the use of hierarchical storage-class memory to increase available memory by more than an order of magnitude per core, and the realisation of interactive supercomputing at the exascale level, in particular interactive computational steering, visualisation and Big Data integration.
The HBP hardware roadmap, aligned with the roadmap of the ETP4HPC, requires the exploitation of innovative energy-efficient technologies. The development system and the central architecture for simulations of the brain model will include hierarchical storage-class memory technologies with fully interactive computational steering and visualisation capabilities. It is planned to reach exascale performance by the end of 2022 using technologies complemented by brain-inspired communication and computing sub-systems.
The major objective of SP7 for the ramp-up phase is to establish the HBP High Performance Computing Platform and make it available to the Consortium initially and subsequently to groups from outside the Consortium.
During the ramp-up phase, the HBP will negotiate with the PRACE Tier-0 organisations GCS and GENCI, which have expressed their interest to add in-kind support to the HPC Platform. The plan is to complement the resources for specific simulations as virtual robotics and conceptual brain models. A further high priority goal is to establish a PRACE community access programme, to be negotiated in the ramp-up phase. This will allow access to the Tier-0 capability of the HPC Platform, reviewed by the HBP's International Access Board, via PRACE services. Formal agreements with PRACE that will benefit both sides will be established during the ramp-up phase.
A second, equally important goal will be to prepare the procurement of the HBP Pre-exascale-supercomputer. By 2017/18, Jülich plans to procure a Big Data-centred system with at least 50 PBytes of hierarchical storage-class memory, a peak capability of at least 50 PFlop/s and a power consumption <= 4 MW. The memory and computational speed of the machine will be sufficient to simulate a realistic mouse brain and to develop first-draft models of the human brain. (The rest of the hardware roadmap targets an exascale machine in 2021/2022 with a capability of 1 EFlop/s and a hierarchical storage-class memory of 200 PB).
Finally, Subproject 7 aims to research and develop the system software and middleware needed for the operational phase. During the ramp-up phase, these activities will be brought into alignment with the ETP4HPC. Key topics for research will include the development of novel mathematical methods, programming models and tools, in situ analysis of multi-Petabyte datasets, and real-time visualisation and visual computational steering of simulations. This novel kind of interactive supercomputing will become invaluable not just for brain simulation but also for a broad range of other applications in the life sciences and elsewhere. The integration of hierarchical storage-class memory in software will boost Big Data analytics and will widely benefit the HPC community providing resilience over millions of processing cores and communication devices. Mirror and checkpoint mechanisms exploiting the new capabilities will make it possible to move toward the final goal of system-wide virtualisation. The unprecedented scalability requirements of whole brain simulation will give a major boost to performance analysis and optimisation as well as to research into new numerical algorithms. In a long-term perspective, algorithmic research is expected to benefit from the novel capabilities of brain-inspired neuromorphic computing and communication.
As a general principle, HPC research in the HBP strives for platform independence through the provision of high-level APIs for application codes and for visualisation and steering middleware. This will be achieved through user-transparent programming paradigms, platform-optimised libraries, and, in the long run, virtualisation of the entire system including the communication sub-systems. This approach will avoid the danger of technology lock-in.
What People are Saying
Collaborate, collaborate, collaborate. This is our opportunity.
A key goal of the Human Brain Project is to construct realistic simulations of the human brain – this will require molecular and cellular information and from that we will be able to model and understand biological and medical processes. In addition, we will be able to use that information to design and implement new kinds of computers and robotics.
The Human Brain is the most complex system that we know of. We would like to develop some kind of ‘google' brain where we can zoom in and out, see it from different perspectives and understand how brain structure and function is related. The ultimate aim of the Human Brain Project is to understand the human brain. This is only possible when we understand the structural organization of the human brain.
The Human Brain Project will become a major driver of ICT in Europe.