HIPERFITHIPERFIT seminar in February postponed
It is time to inform about our HIPERFIT seminar in February. However, you still have more than one week, as we are again postponing the seminar by one week.
The seminar will be held on Tuesday, February 10, 3pm (moved by one week) in Mødelokale A/B at DIKU. It features two talks by faculty members:
-
Marcos Vaz Salles: Making Time-stepped Applications Tick in the Cloud
-
Cosmin Oancea, Jost Berthold: On Real-World Pricing Code and its Parallelisation
The first talk reports Marcos’ joint work with Cornell fellows (presented at SoCC 2011). The second talk is work in progress on essentials of, and insights into, real-world pricing code using a Monte-Carlo method. Abstracts are included below. We hope you will enjoy the talks!
Best regards from HIPERFIT
Abstracts
Marcos Vaz Salles: Making Time-stepped Applications Tick in the Cloud
Scientists are currently evaluating the cloud as a new platform. Many important scientific applications, however, perform poorly in the cloud. These applications proceed in highly parallel discrete time-steps or “ticks” using logical synchronization barriers at tick boundaries. We observe that network jitter in the cloud can severely increase the time required for communication in these applications, significantly increasing overall running time. In this paper, we propose a general parallel framework to process time-stepped applications in the cloud. Our framework exposes a high-level, data-centric programming model which represents application state as tables and dependencies between states as queries over these tables. We design a jitter-tolerant runtime that uses these data dependencies to absorb latency spikes by (1) carefully scheduling computation and (2) replicating data and computation. Our data-driven approach is transparent to the scientist and requires little additional code. Our experiments show that our methods improve performance up to a factor of three for several typical time-stepped applications.
Cosmin Oancea, Jost Berthold: On Real-World Pricing Code and its Parallelisation
HIPERFIT’s main mandate is to investigate language and compiler technology to allow financial software to be (i) elegantly and readily written and (ii) effectively optimized to exploit hardware parallelism. In this context, we report our insights from work in progress: evaluating a reasonably-complex kernel of a partner’s financial library, comprising about 6200 lines of low level C.
First, we experiment with parallelization solutions in the C-language domain and show that speed-ups as high as 22x and 34x can be achieved on a commodity GPU (Quadro2000M), for double and single precision floats, respectively. Second, we distinguish two main sources of optimizations: (i) classical transformations that adapt the memory usage to the hardware parameters, and (ii) the higher-level ones that exploit algorithmic invariants, which need to be synthesized at interface level. Finally, to support this higher-level perspective, we are in the process of rewriting the library in a higher-level language (Haskell) with good support for mathematical abstraction, higher-order functions, modularity and code reuse. Not surprisingly, the functional approach makes the data-parallel nature of the computation much more evident than the original C version.
We expect that the C-Haskell performance gap can be substantially reduced by using modern libraries for data-parallelism in our – so far sequential – code. Furthermore, our results indicate that the optimizations that were identified in the C code carry over to the Haskell domain and can even be partially automated (by means of rewrite rules). Our work in progress on extensions and optimizations is an important step towards a more generic codebase and domain-specific abstractions of pricing code.