Thursday 12 Jan, 1:15PM-2PM, room: G45
Rob Stewart:
Achieving fault tolerance in high performance computing applications
The applicability of Moore's law, for single core CPUs at least, is all but retired for modern computing systems. There's no such thing as a free lunch, for software engineers riding the wave of increasing CPU clock speeds - so instead, paradigms and skeletons for encapsulating *parallel* evaluation is needed. My PhD focuses on distributed-shared memory architectures - that is, a large number of connected multicore machines, evaluating programs in parts, as parallel skeletons permit. As the number of utilized compute nodes increases, failures per unit remains constant. So there is no free lunch, either, for those software engineers who want to run long running complex programs, on large compute platforms, and expect to ever get a result they'd hope for - or indeed to get a result at all.
The contribution of my PhD is to borrow well defined fault tolerant behaviours and techniques from successful languages such as Erlang, and to build these into a shallow embedded language in Haskell. My ambition is to implement fault tolerance in a distributed parallel Haskell, exposing fault tolerant skeletons and supervision behaviours, to enable Haskell to run wild as a successful dependable language on high performance computing platforms.
David Robb
Crowds: Visual Co-design via Rich Crowd-sourcing
This project aims to connect designers to thousands of potential customers and :
- enable crowds to provide emotive feedback in a visual language
- give crowds a sense of ownership of the design process and provide enthusiastic target markets
- significantly reduce time-to-market and the risks of producing unwanted product
Problems to be investigated and overcome include capturing meaningful visual feedback via an Internet application,
and the distillation or summary of the visual feedback for the designer whilst preserving its meaning.
Christoph Strassmair
[tba]
No comments:
Post a Comment