TUM Informatics Colloquium: Computing at a million laptops per second
21 May 2012, 3.30 pm, Seminar Room 1, LRZ, Boltzmannstrasse 1, Garching
Exascale computing brings new challenges to the high performance playing
field but also creates new research opportunities. In this talk, Dr Baden
describes current work in his research group aimed at solving two
outstanding challenges: reducing the opportunity cost of communication
costs and curbing the overheads of abstraction in application libraries.
While both challenges are relevant to current technology, the quest for
Exascale performance increases the urgency of solving them.
To reduce communication costs Dr Baden´s reserach group overlaps
communication with computation by constructing a custom translator that
converts MPI source into a dataflow form that tolerates latency automatically.
Running on up to 96K processors of a Cray XE-6, their translator meets or
exceeds hand coded overlap variants of two important HPC application motifs:
a 3D stencil method and dense matrix multiplication.
To reduce library abstraction overheads, Dr Baden´s group built a custom
translator for an array class library they developed for turbulent flow data
discovery. The translator flattens library abstractions, reducing overheads by over
an order of magnitude in some cases.
Scott B. Baden is Professor of Computer Science and Engineering at the
University of California, San Diego, where he has been a member of the
faculty since 1990. He received the Ph.D. in Computer Science from the
University of California, Berkeley in 1987. Dr. Baden's research is in
high performance and parallel computation and focuses on
programming abstractions, domain-specific translation, performance
programming, adaptive and data centric applications and algorithm