Topic:
Impact
of communication protocols and memory use, MPI-1, MPI-2 and one-sided
communication, shared memory and OpenMP, input-output
to disk storage from parallel programs.
Panel members:
Curt
Janssen
Tomasz
Janowski
Wibe de Jong
Thom Dunning
Topic:
Issues related to
parallel implementation of algorithms to solve coupled-cluster equations, what
is similar to and different from algorithms used in other large-scale parallel
software, e.g. molecular dynamics, finite element analysis, weather simulation.
Panel members:
Josef Noga
Karol Kowalski
John Watts
Topic:
Can we define a set
of problems that everyone can run with their software on their hardware and
report to the community? Similar to the Linpack
benchmark in linear algebra, we need some molecules and some properties,
energy, gradient, with a list of methods. Maybe also a molecule like a polymer
that can be made larger in a defined
way to provide more work as
the number of processors is increased
Panel members:
Peter Pulay
Sudhakar Parmidighantam
P. (Saday)
Sadayappan
Topic:
What algorithms can
we use to scale CC methods to 1,000, 10,000 and 100,000 processors? Do we need
all new algorithms? How do we measure performance? How do we debug?
Panel members:
Ed Valeev
So Hirata
Shawn Brown
Charles Patterson
Beverly Sanders
Discussion of the panel topics and formulation of conclusions and recommendations of the workshop. One of the goals of the workshop is to present a report that defines the present status and future needs and direction of parallelization of Coupled-Cluster methods.