Speaker
Dr
Peter Elmer
(Princeton University (US))
Description
Modern HEP software stacks, such as those used by the LHC experiments
at CERN, involve many millions of lines of custom code per experiment,
as well as a number of similarly sized shared packages (ROOT, Geant4,
etc.) Thousands of people have made contributions over time to these
code bases, including graduate students, postdocs, professional
researchers and software/computing professionals. Elaborate software
integration, testing and validation systems are used to manage the
resulting workflow.
HEP has also been a poster child for "Big Data" science with more
than 100 PetaBytes of event data stored around the world. Its
applications however typically need event data of order 1MB in
memory at a given time plus some tens of MB of calibration data.
The resulting data processing, as well as its Monte Carlo simulations,
is "embarrassing parallel". The hardware needed for this type of
"High Throughput" computing is relatively unspecialized, in the
category of "low-end" CPU servers, although great numbers are needed.
Most of the code executed is arguably non-numerical, with floating
point operations corresponding to only a small fraction of the total time.
These are particular points in a large phase space of scientific
applications. Other characteristics like MPI-style parallelism, small
code bases, the need for "High Performance" computing (GPU's,
specialized interconnects or large memory needs) are not uncommon.
Many are true numerical codes and the data requirements vary greatly.
This presentation will cover the results of investigations of the
software stacks and applications of a variety of other scientific
fields. Where are the commonalities and with whom? Or do we inhabit
a small niche in the world of scientific computing? Particular
attention will be placed on the characteristics and needs of other
scientific projects which will require similar or greater amounts
of resources than HEP in the next ten years.
Author
Dr
Peter Elmer
(Princeton University (US))
