Email:
dfs -> (cs.umd.edu)
Research Advisor:
Dr. James Purtilo
The Scoop:
Life after over-education:
Ho finito!
Education:
Ph.D. Computer Science, University of
Maryland College Park.
M.S. Computer Science, University of
Maryland College Park.
B.S. Astronomy, University
of Maryland College Park.
Dissertation Research:
- Dynamic reconfiguration and
customization of distributed services to meet application-specific
requirements using
virtual
services.
Former Research:
-
Software frameworks for augmenting programmability, administration,
and usability of clustered computing systems.
-
Application of object oriented software development methodologies to
parallel programming.
-
Analysis of the impact of caches in multiprocessor systems
on the performance and scaling of large-scale scientific applications.
-
Performance evaluation of the Convex SPP-1000 for
CESDIS, NASA
Goddard.
-
Performance evaluation of the Beowulf clustered workstation.
-
Development of supercomputers from commodity off the shelf parts and
open source operating systems (i.e.,
Beowulf).
I was lucky enough to have worked with Thomas Sterling, Don Becker,
and Phil Merkey at Goddard Space Flight Center in developing the
first Beowulf cluster.
The Beowulf Project was partly responsible for motivating companies
such as Compaq, IBM, and Cray to produce COTS Linux clusters that
deliver supercomputer performance at a fraction of previous prices.
Beowulf also either spawned or encouraged startups such as Linux
Networx, Penguin Computing, and Scyld to develop cluster hardware and
software products.
Seven of the top ten supercomputers on the 2003 Top 500 supercomputer list are now
clusters as well as 208 of the top 500 systems, many of them
Beowulf-class. This result is unfortunate in some respects because
clusters have many limitations; and supercomputing is sorely in need of
new architectures if it is to reach sustained PetaFLOPS levels of performance.