Evolving Cooperative Teams of Unmanned Aerial Vehicles
(UAVS)
Darrell Whitley
Department of Computer Science
Monday, January 24, 4:10 p.m.
107 Guggenheim Hall
We use simulated evolution, specifically Genetic Programming, to evolve behaviors for teams of Unmanned Aerial Vehicles (UAVS). The problem can also be posed as one of evolving adaptive teams of cooperative agents. Agents must display cooperative interactions to carry out collaborative missions in an uncertain and/or hostile environment. Behaviors are controlled by program trees constructed from a set of low-level sensors. Evolved program trees are robust to changes in initial mission parameters. The evolved programs are also provably near optimal compared to optimal time-to-completion under static conditions. Strategies are evolved using the Genitor steady-state genetic algorithm. Empirical results indicate that Genitor outperforms the traditional generational genetic algorithm for this domain. Videos of the evolved behaviors often show surprising and novel results, as teams of agents cooperatively carry out missions to fly over target areas.
Poster Design for the CSU Computer Science Research Symposium
The CSU Computer
Science Research Symposium will take place on April 18th. All CS
graduate students are strongly encouraged to submit a poster to this
symposium. Everyone will gain valuable experience in preparing and
presenting a poster of their research in a friendly supportive
atmosphere. The talk today will introduce the symposium and
demonstrate how to prepare a poster using Microsoft Powerpoint and
OpenOffice. Advice on what to include on a poster and how to format
it will be discussed. Common errors made in poster design will also
be discussed.
Patterns before and beyond the Object-Oriented Paradigm:
Interpreting the Ideas and Methods of Michael Jackson
After summarizing Jackson's contributions to software engineering and
common threads in his thinking, I plan to illustrate Jackson's work
on components and patterns in two areas:
I will also demonstrate a CASE tool, Jackson Workbench, that supports
program design using JSP and suggest that CASE tools to support
Jackson's JSD and Problem Frames would be useful in teaching and
research in software engineering.
I will conclude by suggesting some topics for educational projects and papers.
Predicting
Malicious Attacks from Authorized Insiders
One of the most important challenges in computer security research is
the problem posed by malicious insiders. These breaches cause more damage
to the system than attacks from outsiders. This is because malicious
insiders are authorized users to begin with. So they do not have to take
extra steps to gain access to the system. In addition they are able to
exploit certain knowledge about system configuration and system
vulnerabilities that are not readily available to outsiders. Most
currently available intrusion detection systems (which include both
anomaly and misuse detection system) fail to address this problem.
A second concern is that most intrusion detection systems operate on
an after-the-fact basis. This may be too late in many cases.
In this work we report on our efforts to develop a system
to predict malicious insider threats. Our work builds on work by
others on user intent analysis. The system we propose analyzes various
user activities in real time and generates an opinion about the
possibility of an attack. This opinion offers the system
administrator a window of opportunity in which to prepare adequately.
Our system is not intended to replace existing intrusion detection and
prevention technology, but rather is intended to complement
current and future technology
Techniques for Model-Based Analysis and Implementation Synthesis of
Real-Time Embedded Software
As real-time embedded systems become increasingly complex, the main
concern for an embedded software developer is no longer to optimize
performance at low levels of coding, but to ensure high-level
modularity, maintainability and correctness at the expense of some
performance loss. There is a recent trend to elevate the level of
abstraction from writing code to building models that generate code,
in order to enable analysis of system properties at early design
stages. As part of the DARPA Model-Based Integration of Embedded
Software project, an end-to-end toolset for component-based real-time
embedded software has been developed collaboratively by researchers
from multiple institutions, with avionics mission computing from
Boeing as an application example. In this talk, I first discuss
model-based analysis techniques developed as part of the toolset,
including analysis techniques for system static structural properties,
and application of model-checking to verify system dynamic behavioral
properties. I then consider the more general class of component-based
software models with interaction style of buffered asynchronous
message passing between components with ports, of which avionics
mission computing is a special case. After building a logical software
model, it is necessary to synthesize a multi-threaded implementation
that runs on a given target hardware platform and satisfies timing
constraints. I discuss real-time scheduling techniques for
implementation synthesis and design space exploration of such software
models. Finally, I discuss future research directions and some ideas
on teaching and curriculum development.
Representation-Independent Compiler Analysis
Program analysis is necessary in many application domains
including software engineering, high performance computing,
scientific computing, data mining, and operating systems.
However, reusing analysis implementations is difficult
because they are typically coupled to a particular
intermediate representation (IR). The goal of the
OpenAnalysis toolkit is to separate analysis from program
representation in a way that allows the orthogonal
development of compiler infrastructures and program
analyses. The separation of analysis from program
representation enables researchers to leverage prior work
and make direct comparison amongst analyses.
In this talk, I describe how insufficient analysis support
is problematic in the context of automatic differentiation.
Analysis-specific interfaces we have developed as part of
the OpenAnalysis package make it possible to decouple
analysis from program representation and therefore implement
analysis that can be reused in any compiler infrastructure
with an intermediate representation for imperative programming
languages.
Human Graphics: Imagery That Works For Its Users
The history of graphics to date is largely a technical history,
focused on physics, mechanics, material science and hardware. This is
now changing: graphics imagery must work for its users in perceptual,
cognitive, social and practical terms. My work spearheads this basic
shift. I will briefly summarize the scope of my work in this vein, and
then delve into two components of that work in depth.
The first component is a basic investigation of human visual
sensitivity and graphical detail (LOD). Many are already familiar with
the contrast sensitivity function (CSF), which describes the
relationship of the thresholds of perceptibility to viewed spatial
frequency and contrast. LOD systems base fidelity on the CSF, despite
the fact that their manipulations take place well above perceptibility
thresholds. We find strong evidence that supra-threshold LOD control
should not be based on threshold perceptibility. Indeed, we find that
the spatial frequency of detail should play only a minor role in
supra-threshold LOD control, and surprisingly, that LOD should often
be increased (not decreased) in low contrast or peripheral display
regions.
The second is a new graphics renderer that is extremely adaptive to
the user's view. While previous renderers were spatially adaptive, our
renderer is both spatially and temporally adaptive. Closed loop
feedback guides sampling to image regions that change significantly
over space or time. Adaptive reconstruction emphasizes older samples
in static settings, resulting in sharper images; and new samples in
dynamic settings, resulting in images that may be blurred but are
up-to-date. Compared to a standard full-resolution 60-Hz rendering,
our renderer's output is of comparable quality (as measured by
root-mean-squared error), but is generated using an order of magnitude
fewer samples. The renderer already adapts immediately to view and
object motion, and is an ideal platform for other perceptual
adaptations such as fidelity reduction in the view periphery.
Biography: Benjamin Watson is an Assistant Professor at the Computer
Science department of Northwestern University. There his group
focuses on human graphics, ensuring that graphics works for its human
users. His research interests are graphics and perception, adaptive
rendering and modeling, procedural modeling of human artifacts
(particularly cities), and graphical interfaces and applications. His
work has been applied to digital entertainment and training, marketing
and financial intelligence, medical therapy and assessment, and
education. Watson earned his Ph.D. at Georgia Tech's GVU Center,
co-chaired the Graphics Interface 2001 conference, chaired the IEEE
Virtual Reality 2004 conference, and will chair the ACM Interactive 3D
Graphics and Games conference in 2006. He is a coauthor of "Level of
Detail for 3D Graphics", published by Morgan Kaufman.
Cooperative Edge Cache Grid: Concepts, Architecture and Techniques
Dynamic content on the World Wide Web (WWW) has experienced an
exponential growth in recent years. According to statistics collected
by BrightPlanet Corporation in 2001, the number of dynamic web pages
was estimated to be at least 500 times the number of static web pages,
holding approximately 7,500 terabytes of information in contrast to
about 19 terabytes of information contained by static web pages. This
massive increase in the dynamic web content has posed a serious
challenge to its efficient and timely delivery. Techniques which are
effective for static web content delivery such as search engines and
proxy caches are not directly applicable to searching, cataloging, and
delivery of dynamic web content.
In recent years, caching on the edge of the Internet has received
considerable attention as a promising solution to challenges posed by
the tremendous growth of dynamic web content. The underlying
philosophy of edge caching is to move data, and possibly some parts of
the application, closer to the users. However, building an edge
network that can efficiently deliver dynamic web content poses a
number of research challenges such as: (1) What is the appropriate
granularity of caching and delivery of dynamic web content, and how to
discover cost-effective cache units in dynamic web pages?, (2) How to
design structures and mechanisms necessary for effective cooperation
among the caches of an edge network?, and (3) Can failures of
individual servers and caches be gracefully handled, thereby ensuring
high data and service availability?
In this talk I will present the design and evaluation of Cooperative
Edge Cache Grid - a scalable, efficient and failure-resilient
cooperative edge cache network that organizes individual edge caches
into cache clouds. I will first introduce the concept of cache clouds,
and present an Internet landmark-based scheme to accurately cluster
the caches of an edge network to yield high quality cache clouds. Then
I will describe the design architecture of individual cache clouds,
including a dynamic hashing-based mechanism for document lookups and
updates, which not only supports efficient one-hop document lookups
and updates, but also balances the document update and lookup loads
among the caches within a cache cloud. In addition, I will give an
overview of a utility-based scheme for placing documents within the
edge caches of a cache-cloud, and discuss the experimental evaluation
of the EC Grid system. Our initial experiments indicate that the
cooperative EC Grid, and its dynamic load management and update
propagation algorithms can significantly improve the performance of
edge cache networks on critical parameters like latency, network-load,
server-load, and hit-rates. I conclude my talk with a brief overview
of other related research projects that I am involved in during my
years at Georgia Tech and the topics of my future interests.
Biography: Lakshmish Ramaswamy is a Ph.D. candidate at College of
Computing, Georgia Tech. His research interests are in the areas of
large-scale distributed systems, Internet data management, Web
performance, overlay networks including grids and peer-to-peer
systems, and distributed databases. Lakshmish's research has been
published in a dozen of international journals and conferences. He is
one of the recipients of the best paper award at the 2004
International Conference on World Wide Web (WWW 2004). Lakshmish's
work at IBM has resulted in three US patents being filed in the last
three years.
A kernel method for predicting protein-protein interactions
Most proteins perform their function by interacting with other proteins.
Therefore, information about the network of interactions that occur in a
cell can greatly increase our understanding of protein function. We
present a kernel method for predicting protein-protein interactions using
a combination of data sources, including protein sequences, Gene Ontology
annotations, local properties of the network, and interactions in
different species. We propose a pairwise kernel that provides a similarity
between pairs of proteins, and illustrate its effectiveness in conjunction
with a support vector machine classifier. We obtain improved performance
by combining several sequence-based kernels based on k-mer frequency,
motif and domain composition and by further augmenting the pairwise
sequence kernel with features that are based on additional sources of
data. In yeast, at a false positive rate of 1%, the classifier retrieves
close to 80% of a set of trusted interactions, demonstrating the ability
of our method to make accurate predictions despite the sizable fraction of
false positives that are known to exist in interaction databases.
Biography: Asa Ben-Hur is currently a postdoctoral fellow in the Department of Genome
Sciences in the University of Washignton in Seattle.
His research interests are in applying machine learning in bioinformatics.
He received a PhD in information systems from the Industrial Engineering
Faculty of the Technion in Israel. He was then recruited to Biowulf, a
biotech startup that sought to apply machine learning to biomedical data.
Following its bankruptcy he continued in bioinformatics as a postdoc at
the Department of Biochemistry at Stanford.
Marker Gene Selection and Cancer Classification in Microarray Data Analysis
Recent high-throughput DNA microarray technologies, such as gene expression
arrays and Single-Nucleotide Polymorphism (SNP) arrays, have afforded
bio-medical researchers with the unprecedented capability to interrogate genes
at the whole genome scale. Two important and closely related issues in
microarray data analysis for cancer research are marker gene selection and
cancer classification. These issues pose a major computational challenge
because the number of probes on a microarray is typically several orders of
magnitude greater than the number of tissue samples.
In the first part of my talk, I will present our novel HykGene approach for
selecting marker genes for phenotype classification using microarray gene
expression data. Our approach is capable of selecting relatively few marker
genes while offering the same or better leave-one-out cross-validation
accuracy
compared to approaches that use top-ranked genes directly for classification.
In the second part of my talk, I will present our recent results on cancer
classification using Loss of Heterozygosity (LOH) data and DNA copy number
aberrations derived from SNP microarrays.
Joint work with Fillia Makedon, Justin Pearlman, and James Ford.
Bio Sketch: Yuhang Wang is currently a Ph.D. candidate in the
Department of Computer Science at Dartmouth College. His research
interests include bioinformatics, medical image analysis, data mining,
machine learning, pattern recognition, multimedia information
retrieval, and computational geometry. He received the B.E. degree in
1996, and the M.E. degree in 1999, both from Tongji University in
Shanghai, China. Most recently, he received a Bioinformatics
Postdoctoral Fellowship from Harvard Medical School.
A Cultural
History of the Original JPEG Standard
The "J" in JPEG (Joint Photographic Experts Group) acknowledges its
two main parent organizations, ISO (International Organization for
Standardization) and CCITT (International Telegraph and Telephone
Consultative Committee) which is now called ITU (International
Telecommunications Union). The JPEG committee was one of the first
times these organizations formed a single committee for the purpose of
creating a joint standard despite their different histories, cultures
and procedures for settling disputes. ISO is a decentralized
self-regulating group aimed at industry standards. This non-treaty
agency of the United Nations tends to collect experts (hence the term
"experts group") to work on standards that are for the "good of the
industry." The CCITT is an intergovernmental treaty organization
sponsored by the United Nations. Official voting was done by
nations. It issues "recommendations" that each country can voluntarily
adopt. However, frequently administrations demand rigid conformance
with these recommendations for their telecommunica-tions equipment.
Rather than trying to meld its parents' cultures or deciding which set
of rules to use for settling conflicts, the original JPEG committee
adopted a 100% consensus rule. This nontechnical talk will share
stories illustrating how achieving complete consensus influenced the
definition of this widely-used standard.
Dr. Joan Mitchell graduated from Stanford University with a B.S. in
physics in 1969. She received her M.S. and PhD. degrees in physics
from the University of Illinois at Champaign-Urbana in 1971 and 1974,
respectively. She joined the Exploratory Printing Technologies group
at the IBM T. J. Watson Research Center immediately after completing
her PhD. She was a manager there for nine years. She then worked for
three years in IBM Marketing before returning to the IBM Research
Division in 1991 to work again in the Image Technologies group as a
manager. From 1987 through 1994, she was a member of the ISO and CCITT
international Joint Photographic Experts Group which standardized the
color image JPEG compression algorithm. She was the final editor of
JPEG Part 1, and in 1992, coauthored a book about JPEG. In 1994, she
took a two year leave of absence from IBM during which she coauthored
a book on MPEG, consulted for IBM Burlington, and was a visiting
professor at the University of Illinois for six months. She returned
to the IBM T.J. Watson Research Center as a Research Staff Member in
the Image Applications Department. For the last three years she was on
temporary assignment with the IBM Printing Systems Division in
Boulder, CO, transferring there permanently in 2002. Since 1976, Joan
has worked in the field of image processing and data compression. She
received IBM Outstanding Innovation Awards for Two-Dimensional Data
Compression in 1978, for Teleconferencing in 1982, for Image View
Facility in 1985, for Resistive Ribbon Thermal Transfer Printing
Technology in 1985, for Speed-Optimized Software Implementations of
Image Compression Algorithms in 1991, and for the Q-coder in
1991. December 2001, she was awarded an Outstanding Technical
Achievement Award for Algorithms for Improved Printer Performance
Transferred to IBM's Printing Systems Division and her Twenty-first
Invention Achievement Plateau Award. She was elected to the IBM
Academy of Technology in 1997, and became an IEEE Fellow in 1999. She
is a member of APS, IEEE, IS&T, and Sigma Xi and co-inventor on 40
patents. She was made an IBM Fellow in 2001. In 2002, she initiated a
Master Inventor program for PSD in Boulder and became a Master
Inventor there. She recently became a certified PADI Dive Master.
Title TBA
Dr. Wolfgang Gentzsch is the managing director of MCNC Grid Computing
& Networking Services, a member of the MCNC family of companies. After
leading global grid technology development at Sun Microsystems
(NASDAQ: SUNW), Gentzsch joined MCNC in April 2004 with more than 25
years of experience in grid computing, software development,
computational engineering, computer architecture, and teaching.
At MCNC, Gentzsch directs the organization's grid strategy and
technology development, including the development of one of the
nation's first statewide research and education grids. Prior to
joining MCNC,
Gentzsch became senior director of grid computing for Sun
Microsystems. He was responsible for Sun's grid computing vision,
strategy and technology development. Gentzsch joined Sun in 2000 when
it acquired GRIDWARE, a distributed computing software company that he
co-founded in 1999. Gridware's technology is the foundation for the
Sun Grid Engine, the world's leading distributed resource management
software used in over 10,000 departmental and enterprise grids
worldwide. Sun Grid Engine was a finalist at LinuxWorld 2002, earned
the Excellence in Cluster Technology Award at ClusterWorld 2003, and
was honored with the Frost & Sullivan Excellence in Technology Award
in 2004.
Gentzsch was also a professor of mathematics and computer science at
the University of Applied Sciences in Regensburg, Germany, and served
as the head of computational fluid dynamics and supercomputing at the
German Agency for Aerospace and Aeronautics. Throughout his career,
industry leaders including IBM, Cray Computers and Digital Equipment
Corporation have sought his consulting skills on distributed computing
and supercomputing projects.
He is a widely published author of more than 150 articles about
computer science, numerical algorithms, engineering applications, and
grid computing, and he has spoken extensively at conferences around
the world. He is an adjunct professor at Duke University, North
Carolina State University and the University of North Carolina at
Charlotte.
A Tool Supported Approach to Testing UML
Designs
For Model Driven Development approaches to succeed, there
is a need for model validation techniques. We present an
approach to testing designs described by UML class diagrams,
interaction diagrams, and activity diagrams. We describe a
prototype tool that transforms UML design models into
executable forms with test infrastructure, executes
tests, and reports failures.
Quantum
Information and Quantum Computing - Why the Excitement?
Quantum information has special properties: the state of a quantum
system cannot be measured or copied without disturbing it; quantum
state can be entangled, two systems have a definite state though
neither has a state of its own; superposition - we cannot reliably
distinguish non-orthogonal states of a quantum system.
The potential impact of quantum computing and quantum information
theory is astounding. Reversible quantum computers avoid logically
irreversible operations and can, in principle, dissipate arbitrarily
little energy for each logic operation. By contrast, solid state
devices vintage year 2000 require some 3 x 10^-18 Joules/switching
operation. Quantum parallelism has the potential to support "exact"
simulation of complex systems. Quantum key distribution protocols
allow detection of an intruder with a very high probability.
To exploit the immense power of a quantum computer we need to develop
quantum algorithms. In 1944, Peter Shor found a polynomial time
algorithm for the factorization of n-bit numbers on quantum
computers. In 1996, Grover described a quantum algorithm for searching
an unsorted database containing N items in a time of order sqrt(N) while on
a classical computer the search requires a time of order N.
Only few quantum algorithms have been discovered so far; quantum
computing requires a substantial adjustment to our way of
thinking. Quantum effects are counterintuitive; we need to understand
the physics and be familiar with the sophisticated mathematical
apparatus of quantum mechanics.
In this presentation we introduce basic concepts and some of the
applications of quantum computing and quantum information theory. We
present quantum gates and quantum circuits used to transform the state
of a quantum system and thus to process information. Then we discuss
in some depth the concept of quantum parallelism. We illustrate
quantum parallelism with the example of an "oracle" capable of
establishing if a binary function is balanced or not. If time permits,
we shall discuss concepts from quantum information theory, and provide
some insights into dense coding and quantum teleportation.
The roadblocks on the way to quantum computing are staggering. The
immense costs to develop quantum computers could only be justified if
new quantum algorithms and applications of quantum computing and
communication are discovered. The discovery of new algorithms and
applications of quantum computing requires computer scientists to play
an increasingly more significant role in this interdisciplinary
research field. We should also ponder how to educate our students and
allow them to contribute to the development of this exciting field.
Bio:
Dr. Dan C. Marinescu (http://www.cs.ucf.edu/~dcm) joined the Computer
Science Department at University of Central Florida in August 2001 as
Professor of Computer Science. He has been an Associate and then Full
Professor of Computer Science at Purdue University, in West Lafayette,
Indiana, since 1984. He is conducting research in parallel and
distributed systems, computational biology, ubiquitous computing,
Petri nets, and quantum computing and has published more than 160
papers in journals and refereed conference proceedings in these
areas. He is the author of "Internet-Based Workflow Management:
Towards a Semantic Web," published by Wiley in 2002, and has co-edited
"Process Coordination and Ubiquitous Computing." The book "Approaching
Quantum Computing," co-authored with Gabriela M. Marinescu, was
published in September 2004 by Prentice Hall.
The Microprocessor in the Year 2015: Issues, Challenges,
Potential Avenues to Solutions
The first time I gave this talk (not really THIS talk), the year was
1985 and the title was "The Microprocessor in the Year 1995." Moore's
law was promising 10 million transistors on a chip and the question
was what to do with all that potential. Some gurus said "Give it
up. We've reached the end of the line as far as microarchitectural
improvements are concerned." But, we didn't give up, and ten years
later, we had the Pentium Pro, for example. Time passes, Moore's Law
continues to be alive and well, and we continue to look ten years
out. At the High Performance Computer Architecture conference less
than two months ago, a senior Intel computer architect promised 10 to
50 billion transistors by 2015. Even in the near term, we will soon
see one billion transistors on a chip, running at a frequency in
excess of 10 GHz. With such potential, a lot of problems that were
minor in the past are now starting to drive chip design. Like power
consumption, off chip delay to memory, on chip delays, effectiveness
of deep pipelines, etc. And, still some gurus say, "Give it up." This
talk looks at the above, explains the issues and suggests avenues that
could make a difference.
Bio: Dr. Yale Patt is Professor of
Electrical and Computer Engineering and the Ernest Cockrell,
Jr. Centennial Chair in Engineering at The University of Texas at
Austin. He directs the research of 13 PhD students in high performance
computer architecture and implementation, and enjoys teaching both
large undergraduate classes and small advanced graduate seminars. He
has, for more than 35 years, combined an active research program with
extensive consulting in industry and a strong commitment to
teaching. The research he has conducted with his students has had
major impact on the microprocessor industry. HPS (the integration of
wide-issue, speculative, out-of-order execution, and in-order
retirement), the two-level branch predictor, and SSMT (more commonly
called helper threads) are three examples. Dr. Patt has received many
awards for his research and teaching, including the IEEE Emmanuel
R. Piore Medal in 1995, the IEEE/ACM Eckert-Mauchly Award in 1996, the
IEEE Wallace W. McDowell Award in 1999, and the ACM Karl V. Karlstrom
Outstanding Educator Award in 2000. His vital concern for how we
introduce computing to computer science and engineering majors has led
to "Introduction to Computing Systems, from Bits and Gates to C and
Beyond," co-authored with Professor Sanjay Patel of Illinois. Yale
Patt earned his BS at Northeastern and MS and PhD at Stanford, all in
electrical engineering. He is a Fellow of both the IEEE and of the
ACM.
Understanding Complex Adaptive Systems:
An Evolutionary Agent-based Approach
The world is full of complex adaptive systems, such as computer
networks, stock markets, biological systems, and economies. There
continues to be a need to understand the underlying dynamics of these
systems in order to better design, predict, respond to, and modify
them. However, because they invariably consist of a large number of
components that interact in non-linear ways, they are extremely
difficult for the human mind to grasp. Furthermore, they generally
defy formal mathematical analysis without making unrealistic
assumptions regarding linearity and independence.
A promising alternative approach is the use of modeling and simulation
to augment the human cognitive system. To be effective, the modeling
and simulation tools brought to bear must be both effective in
capturing the non-linear dynamics and scalable to real-world
problems. We believe that this is now possible via the synergistic
blending of two mature technologies: agent-based modeling and
evolutionary computation. The agent-based modeling technology allows
one to effectively model the non-linear dynamics as an emergent
property of the interactions among agents. The evolutionary
computation technology endows these agents with the ability to adapt
their behavior over time, creating a co-evolutionary dynamic capable
of generating important insights into realistic, yet previously
unencountered scenarios.
This talk will describe the tools and techniques we have developed for
this purpose, and illustrate their application to computer network
security and treatments for inhalation anthrax.
Professor De Jong coined the term "Genetic Algorithms" while doing his
Ph.D. under John Holland at the University of Michigan in 1975 and did
the first research exploring the use of genetic algorithms for
functions optimization. Over the last 30 years, he has continued to
explore the frontier of innovation in genetic algorithms and
evolutionary computation.
Adaptive Resource Management in Multithreaded
Architectures
In order to leverage increasingly many transistors, designers are
moving toward including multiple multi-threaded processor cores on a
single chip die. As these systems increase in size, designers will
face new challenges in simultaneously managing current swings (di/dt),
strictly limiting power consumption, and efficiently managing on-chip
cache memory resources to meet system performance demands and
real-time deadlines. Unfortunately, existing general-purpose
operating systems and run-time systems are not adequately designed to
support these architectures. Nevertheless, by integrating run-time
monitoring and management techniques to dynamically adjust system
resource allocation to applications characteristics, unparalleled
advances in computing systems will be enabled.
We propose a full system approach for resource management of
multithreaded multi-core systems. In such systems, opportunities to
improve system behavior via adaptation occurs at three time scales -
the microarchitecture controls activities that occur in the 10s-100s
of cycles, the runtime system controls activities on the scale of
1000s-1000000s of cycles, and the operating system controls resources
and activities at larger time scales. We present our initial
investigation into novel techniques for each of these components that
operate synergistically to enable the full potential of future
systems.
Bio: Dan Connors is an Assistant Professor at the University of Colorado at
Boulder. He received his Ph.D. in Computer Engineering from the
University of Illinois at Urbana-Champaign in 2000. His research
explores the interaction of compilers, run-time optimizers, and
operating systems in modern architectures. He directs the DRACO
research group, which investigates run-time optimization and compiler
technologies that enable optimization, power efficiency, prefetching,
and thread scheduling in future multi-threaded multi-core systems.
Sustained Petaflop
and Beyond Can Parallel Computing Systems Meet The
Challenges?
Breakthrough-quality scientific discoveries in the new millennium
(such as those expected in computation biology and others), along with
optimal engineering designs, have created a demand for High-End
Computing (HEC) systems with sustained performance requirements at a
petaflop scale and beyond. Despite the very pessimistic (if not
negative) views on parallel computing systems that have prevailed in
1990s, there seems to be no other viable alternatives for such HEC
systems.
In this talk, we present a fresh look at the problems facing
the design of petascale parallel computing systems. We review several
fundamental issues that such HEC parallel computing systems must
resolve. These issues include: execution models that support dynamic
and adaptive multithreading, fine-grain synchronization, and global
name-space and memory consistency. Related issues in parallel
programming, dynamic compilation models, and system software design
will also be discussed. Present solutions and future direction will be
discussed based on (1) application demand (e.g. computation biology
and others) , (2) the recent trend as demonstrated by the HTMT, HPCS,
and the Blue-Gene Cyclops (e.g. Cyclops-64) architectures, and (3) a
historical perspective on influential models such as dataflow, along
with concepts learned from these models.
Biography
Professor Gao's
main research interests include high-performance architectures and
systems, their programming models, system software (particularly
compilers), and applications such as computational biology and
bioinformatics. He has devoted most of his time in searching for a
scalable parallel program execution/architecture model that can serve
as a basis for high-end parallel supercomputers. To this end,
Professor Gao has many research publications in these fields. He is an
active participant in professional activities.
Dr. Guang R. Gao
received his Ph.D. degrees in Electrical Engineering and Computer
Science from the Massachusetts Institute of Technology. Currently, he
is a Professor of the Department of Electrical and Computer
Engineering at University of Delaware, where he has been the founder
and director of the Computer Architecture and Parallel Systems Lab,
and the Director of the Bioinformatics Center at Delaware
Biotechnology Institute (2001-2004). Prior to the above, he has been a
faculty member of the School of Computer Science at McGill University,
Montreal, Canada.
Born in Beijing, Gao is among the first wave of
graduate students from People's Republic of China to attend graduate
school in the U.S. in the early 1980s. In fact, he is the first
student from PRC to receive a MS and Ph.D. degree in Computer Science
from MIT.
Dan Massey, Sudipto Ghosh, and Chuck
Anderson
Department of Computer Science, Colorado State University
Monday, February 7, 4:10 p.m.
107 Guggenheim Hall
Nick Ourusoff
Department of Computer Information Science at
Western State College of Colorado
Monday, February 14, 4:10 p.m.
107 Guggenheim Hall
Nayot Poolsappasit
Department of Computer Science, Colorado State University
Monday, February 21, 4:10 p.m.
107 Guggenheim Hall
Dr. Zonghua (Sam) Gu
Department of Computer Science, University of Virginia
Tuesday, February 22, 9:00 am - 9:50 am
Clark A-102
Dr. Michelle Strout
Argonne National Laboratory, Mathematics and Computer Science Division
Thursday, February 24, 9:00 am - 9:50 am
Engineering E203
Dr. Benjamin
Watson
Department of Computer Science, Northwestern University
Monday, February 28, 9:00 am - 9:50 am
Rockwell 39
Dr. Lakshmish Ramaswamy
College of Computing, Georgia Tech
Monday, February 28, 11:00 am - 11:50 am
Engineering D-104
Asa Ben-Hur
Noble Laboratory, University of Washington, Seattle
Thursday, March 3, 9:00 am - 9:50 am
Engineering E104
Yuhang Wang
Department of Computer Science, Dartmouth College
Thursday, March 3, 4:10 pm - 5:00 pm
Guggenheim 107
Dr. Joan Mitchell
IBM Printing Division, Boulder, CO
Monday, March 7, 4:10 p.m.
Glover 130
Wolfgang Gentzsch
Managing Director at MCNC Grid Computing and Networking Services
Monday, March 21, 4:10 p.m.
Lory Student Center
Trung T. Dinh-Trong, Nilesh R. Kawane, Sudipto
Ghosh, Robert B. France
Department of Computer
Science, Colorado State University
and Anneliese A. Andrews
Department of Electrical
Engineering and Computer Science, Washington State University
Monday, March 28, 4:10 p.m.
107 Guggenheim Hall
Dan Marinescu
University of Central Florida
Abell Distinguished Lecture in Computer
Engineering, in conjunction with the Computer Science Department
Seminar
Friday, April 1, 4:10 p.m., Reception at 3:30.
203/205 Lory Student Center
Yale
Patt,
Department of Electrical and Computer Engineering,
University of Texas at Austin.
Joint CS/ECE seminar sponsored by ISTeC.
Monday, April 11, 4:10 p.m. Refreshments at 3:30.
Ammons Hall
Ken De Jong
Department of Computer Science, head of Evolutionary Computation
Laboratory, faculty member of the Krasnow Institute
George Mason University,
Monday, April 25, 12 noon
Wager 232
Dan Connors
Department of Electrical and Computer Engineering and Department of Computer Science
University of Colorado,
Boulder, Colorado
Monday, April 25, 4:10 p.m.
107 Guggenheim
Guang R. Gao
Computer Architecture and Parallel System Laboratory,
Dept. of Electrical and Computer Engineering, University of Delaware
Abell Distinguished Lecture in Computer
Engineering, in conjunction with the Computer Science Department Seminar.
Monday, May 2, 4:10 p.m.
107 Guggenheim Hall