Skip to main content

Communication and architectural support for scalable parallel computing

Objective



Research objectives and content
In this project I would like to study the relations between the BSP programming model and the characteristics of the interconnection networks. There are three main themes of research that I would like to explore, related to different switching techniques and technological
characteristics.
The first thing is to bring to a mature state the results I have already obtained on wormhole networks in my PhD thesis, by analyzing the collective communication patterns originated by the BSP model, the total-exchange and the h-relations.
The second line of research that I want to pursue hinges on a new form of switching called wave switching. Wave switching is a hybrid technique for high performance routers that combines wormhole and circuit switching on the same architecture. This approach provides a drastic reduction in latency and increment in throughput when messages are long enough or there is some information available on the communication pattern. I think that this technique can be successfully applied to BSP computing to exploit all the information available on the global structure of the communication pattern.
The third and final line of research tends to explore one of the most promising and challenging subjects of the parallel architectures: the use of free space optics in the implementation of the interconnection networks. I plan to use an approach analogous to the wave pipelining to exploit all the potential of this technology. Part of the huge available bandwidth can be used to schedule the communication pattern establishing paths while the remaining part of the bandwidth can be dedicated to the actual message transmission.
Training content (objective, benefit and expected impact)
I expect the achievement of several results from this medium-term project. First of all, I would like to develop a reliable cost- calculus for different classes of architectures, topologies and technological solutions. I also expect to provide insight on the main characteristics that are at the basis of the implementation of new routing chips. Wave switching and free space optics are two very promising techniques that are still unexplored.
Links with industry / industrial relevance (22)
The outcome of this training should display a strong industrial relevance. I would like to provide new.results useful to develop the next generation of interconnection networks and routing chips of the PQE2000 project, a joint project that includes many European academic institutions and industries.

Funding Scheme

RGI - Research grants (individual fellowships)

Coordinator

University of Oxford
Address
11 Keble Road
OX1 3QD Oxford
United Kingdom

Participants (1)

Not available
Italy