Computer 03

High Performance Computing & Networking

Wednesday, June 2, 2010, 9:00am – 10:30am, Hall B2.1

>> View detailed schedule

Chair

  • Prof. Dr. Georg Carle, Professor for Network Architectures & Services, Technische Universität München, Germany

Advances in networking are an important driving factor for High-Performance Computing data centers. Interconnection speeds of 40 Gbit/s become possible with Infiniband as a specific Interconnection technology, but also with Ethernet technology.

Interconnection performance is a key factor for the applicability of cloud concepts to high-performance computing applications. Virtualisation approaches, including network virtualisation, are important technological components of cloud computing. Among the features supported by network virtualisation is the capability of transparent live migrating of virtual machines.

The session aims to address the potential of network innovations and virtualisation technology in support of high-performance computing applications. Also limiting factors such as overhead from virtualisation are discussed. A key question in this context is which evolution of networking approaches for cloud technology is needed, and to which extent this is feasible, for addressing the needs of high performance computing.

The session will review recent proposals for low-latency communication, such as low-latency Ethernet. The session will also present and review functionalities, such as the capability of migrating Virtual Machines across multiple networks or data centers, for purposes such performance optimisation, or load balancing within and across data centers. This can be realized by an innovative networking infrastructure, such as using virtual switches / routers consisting of multiple, distributed forwarding elements that are managed by a logically centralized controller entity. An example of such a concept is the Openflow switch consortium initiated by Stanford University.

The session will be opened by the chairmen introducing the context. It will be followed by perspectives from a supercomputer center, and from IT vendors of networking solutions for HPC data centers and cloud computing, followed by a panel discussion open to questions from the audience.