High Performance Computing Symposium (HPC 2012)
March 16, 2012
Welcome to the 2012 High Performance Computing Symposium!
This is the 20th special symposium devoted to the impact of high
performance computing and communication on scientific computing and
simulation. The symposium encompasses a wide variety of topics with
a focus on tools and applications for the simulation of physical and
engineering systems. The goal of the symposium is to encourage
innovation and synergistic advances in these areas. It promotes the
exchange of ideas and information between universities, industry, and
national laboratories about new developments in system modeling, high
performance computing and communication, and scientific computing and
simulation.
A particular focus this year is on adapting algorithms to work on some
of the most current architectures. Fourteen high quality papers and two
tutorial sessions will be presented in six technical sessions.
The first tutorial is "Ingredients for good parallel performance on multicore
systems" presented by Georg Hager of the University of Erlangen-Nurember.
The second tutorial is on "Vienna CL", an open source linear algebra library
that allows use of both GPUs and multicore CPUs, presented by Karl Rupp from
the Vienna University of Technology.
The Symposium features three keynote talks from highly regarded experts
in the field. The first keynote talk is entitled
"Designing multiple-fault tolerant RAIDS: Using graphs and hypergraphs".
Prof. Narsingh Deo, Millican Chair and Director of the
Center of Parallel Computation at the University of Central Florida.
Abstract. Redundant Arrays of Independent Disks (RAID) systems are in widespread
use because of their enhanced I/O bandwidths, large capacities, and low cost.
However, the increasing demand for greater array capacities at low cost has
led to the use of arrays with larger and larger number of disks, which increases
the likelihood of the concurrent occurrence of two or more random disk failures.
Hence the need for RAID systems to tolerate two or more random disk failures
without compromising disk utilization. In this talk, we will present a novel
algorithm based on the perfect 1-factorization of the complete graphs KP and
K2P - 1 for placing data and parity in two-disk fault-tolerant arrays with
(P - k) and (2P - 1 - k) disks respectively, where P is a prime number and k
greater than or equatl to 1. Furthermore, we determine the fraction of space
used for storing parity in such arrays and show that this fraction has the
optimal value when k = 1.
The second keynote talk is "Accelerating linear system solutions on new
parallel architectures", presented by Prof. Marc Baboulin, Inria Excellence
Research Chair at the University Paris-Sud.
Abstract. Recent years have seen an increase in peak "local" speed through
parallelism in terms of multicore processors and GPU accelerators. At the same
time, the cost of communication between memory hierarchies and/or between
processors have become a major bottleneck for most linear algebra algorithms.
In this presentation we explain how hybrid multicore+GPU systems can be used
efficiently to enhance performance of linear algebra libraries. We illustrate
this approach by considering hybrid factorizations where we split the
computation over a multicore and a graphic processor and where the amount
of communication is significantly reduced. We also show how mixed precision
algorithms can improve performance. Next we describe a randomized algorithm
that accelerates factorization of general or symmetric indefinite systems on
multicore or hybrid multicore+GPU systems. Randomization prevents the
communication overhead due to pivoting, is computationally inexpensive,
requires very little storage. The resulting solvers outperform existing
routines while providing us with a satisfying accuracy.
A third invited presentation
is "Exascale algorithms for synthesizing parameters of computational
models" by Prof. Sumit Kumar Jha of the University of Central Florida.
Abstract: The success of high performance computing has facilitated the rapid
development of increasingly complex models of natural and engineered systems
by biologists, physicists, chemists, and even financial engineers. While the
development of such models requires considerable domain knowledge and arguably
little knowledge of the science of computing itself, we survey a key problem
in computational modeling that cuts across boundaries of scientific disciplines
and motivates the development of new massively parallel high performance
algorithms: the synthesis of parameters for stochastic computational models.
The overall structure of stochastic computational models can often be obtained
from first principles by using our understanding and insight into the physical
system that is being modeled. However, several components of a computational
model are not readily obtained from first principles. Very often, model
designers incorporate such information in the model as parameters. The model
designer chooses these parameter values carefully so that the computational
model replicates the behavior of the natural or engineered system being
modeled. Naturally, a key problem in computational modeling is the
identification of such parameters.
While the insight of the domain knowledge expert was sufficient to estimate
parameters of small stochastic models, there is an urgent need to develop
massively parallel algorithmic techniques for synthesizing parameters of
large and complex stochastic models. In this talk, we will survey massively
parallel algorithms for synthesizing parameters of stochastic models from
semi-quantiative and qualitative specifications. We will study the use of
stochastic temporal logics for describing the expected behavior from complex
stochastic models, and argue that domain-oriented flavors for temporal logics
will facilitate the development of discipline-specific formal frameworks for
specifying correctness of computational models.
We look forward to hosting a very successful High Performance Computing
Symposium this year.
Gary Howell, General Chair
North Carolina State University
Fang(Cherry) Liu, General Vice-Chair
Ames Laboratory
Steven Seidel, Program Chair
Michigan Technological University
Rhonda Phillips, Program Vice-Chair
MIT Lincoln Laboratory
Karl Rupp, Publicity Chair
Vienna University of Technology.