 |
||
|
||
Home > TERATEC FORUM 2010 > > Workshop 5
Wednesday, June 16 , 9h00 to 13h00
HPC Systems Architectures
We have just entered the Petascale era, and there is no doubt the coming months and years will see a tremendous increase of the number of very large size installed HPC systems. On our way towards the next, but still far horizon of Exascale, it is also clear that a sustainable consolidation of the current or near-existing technologies will more and more have to tackle the architecture and system issues as a whole.
Computing, communication, storage and data management elements will have to be integrated tightly in a balanced, energy-efficient and robust way to ensure the viability and acceptability of high end supercomputers. Building on efficient operations and system administration, seamless access to services will also be required to foster a widespread usage of HPC resources by scientific and industrial end users.
This workshop will address those global design, integration and services issues, through the enlightening visions of major players and leaders in the field of very high performance computing and data management systems.
The following workshop n°6 will elaborate on the challenges of harnessing these massive computing resources, focusing on programming models and tools for emerging and future HPC architectures.
Chairman: Jean-Philippe NOMINE, CEA |
|
|---|---|
09h00
|
Introduction We will introduce the differen t presentations of the session by highlighting issues and challenges of state-of-the-art high-end HPC, picking up examples from CEA experience in the design and operation of large supercomputing infrastructures for defence, research and industrial applications. |
09h20
|
Cray Next Generation System Architecture Cray's next generation systems build on the reliability and scalability of the Cray XT5 and XT6 supercomputers, boosting performance of both MPI and PGAS applications. We will present an overview of the Cray system, its processors, interconnect and software together with early performance data. |
10h00
|
Future exascale storage and evolution of file systems In this talk we will review some requirements for exascale file systems and their implications for the architecture of such storage infrastructure. We will illustrate this with some approaches taken in ClusterStor's Colibri file system. |
10h40 |
Break |
11h00
|
Big Challenges for Big Systems As HPC is moving towards Exascale Computing the suppliers and the users of such systems face quite dramatic challenges on various aspects of the hardware and software systems to be used to implement such systems . This talk will describe SGI’s approach to these challenges focusing especially on the system interconnect. |
11h40
|
Architecting a GPU Computing Solution In order to better adapt the design of an hybrid CPU/GPU based architecture, it is key to know the typology of the full applicative environment (code structure, data set size,…). This analyze will lead to an optimized design of the target architecture. HP has developed a wide portfolio of hardware solutions, allowing to chose within a server, the quantity and type of CPU cores, of GPUs, the type of PCI-E interface in between, and the server interconnection. |
