Multicore Discussion 2010414

Chris suggests a three pronged approach

  1. algorithm improvements # libraries # programming techniques

Marc brought up a list of things or ways to organize.
We discussed application areas that might be of interest to
focus our attention:

  1. event processing # LQCD simulation # Accelerator modeling # Astrophysics galaxy formation simulations # Astrophysics image processing

Two other things came up:

  1. VM use with GPUs - how would this work? # VM performance issues # what about analysis on the desktop? (and other computing tasks)

Other: Is the model used by the astrophysics community regarding coding of
production quality algorithms the right way to do business? This is where
the scientist writes in a high-level language to sort of thing they want to
do and the programmer takes that and converts it to a C++ program. This
does simplify the problem of coding to some multicore standard.

Possible steps

Problem to focus on to frame thinking

  • mu2e triggerless DAQ and other DAQ systems. * accelerator modeling * sampling partons (secondary) * Bayesian analysis (secondary)

ask for talk on the above things

  • Rob K for Kalman filter or Hough transform * Jim A for accelerator modeling

Look in the CS environment for

  • direction of processors over next five years * tools (compilers mostly) * techniques (ways to code) * research groups to contact

Talk to groups we know

  • Argonne * ROSE people

What are tools and techniques

  • PGAS * Parallel I/O * functional programming

Marc has some interest in doing work with Chapel.

Goal: to do the groundwork necessary to get into grant writing
and writing papers.