Visiting researcher application: Massive Multiprocessor Architectures

点击翻译按钮获取中文摘要

The proposal is to support an extended visit to the Computer Laboratory at Cambridge from Professor Andrew Brown, from Southampton. The headline objective is to undertake a feasibility study and design a strategy for an extended collaboration between the Universities of Cambridge, Southampton and Manchester.Southampton and Manchester have a grant entitled A scalable chip multiprocessor for large-scale neural simulation (EP/D079594/01). A year into this, it has become clear that the potential application arena for the architecture being developed extends far beyond the initial mandate of neural simulation. Cambridge are building a FPGA based hardware /software infrastructure to enable manycore parallel computer architectures to be simulated in (near) real time. They have a recently awarded grant entitle C3D - Communication Centric Computer Design (EP/F018649/1). Conventional parallel processing systems are large and expensive, and provide hundreds or thousands of processing nodes. Large scale distributed systems (seti@home et al) potentially provide millions of cores, but with extremely limiting bandwidth constraints.The work at Cambridge/Southampton/Manchester has the potential - for the first time - to make available a tightly-coupled, scalable, million core+ architecture. Brushing aside for the moment such fundamental issues as how one might even design a pan-processor compiler for such a system, one of many unanswered questions is What might we do with it if we had one? Although applications have a way of migrating themselves towards useful architectures, attempting to foresee the sphere of use as early as possible in the design stage has obvious benefits to the design process at many levels of granularity.Potential application arenas include:* Neural simulation (the focus of the existing Southampton/Manchester grant)* Discrete and continuous system/network simulation* Finite element analysis* Ray-tracing and rendering* Protein folding prediction* Data-mining, pre-emptive disaster predictionThis visit will support the construction of a detailed roadmap describing the research to be undertaken; one particular application (protein folding) pursued in a little more detail; and the feasibility of bringing together the two threads of activity (the Manchester/Southampton scalable architecture and the Cambrudge FPGA farm) assessed in some detail.