Date: Thursday April 29, 204
Time: 11am-12:15pm
Location: Woodward 149
Scott Pike <pike@aya.yale.edu>
Department of Computer Science and Engineering
Ohio State University
Abstract:
Ideally, faults should be isolated within small, local neighborhoods of impact.
Failure locality quantifies impact as the radius of the largest set of processes
disrupted by a given fault. The locality radius of a distributed algorithm demarcates
a halo beyond which faults are masked. As such, fault-local algorithms are central
to engineering survivable distributed systems, because they protect against cascading
and epidemic failures.
My work makes theoretical and practical contributions to fault-localization for the generalized dining philosophers problem, subject to crash failures. The optimal crash locality for synchronous dining is 0, but this metric degrades to 2 for asynchronous dining. My work resolves the apparent complexity gap by constructing the first known dining algorithms to achieve crash locality 1 under partial synchrony.
The software-engineering context of my approach consists in augmenting existing dining algorithms with unreliable failure detection oracles. My extended results characterize optimal locality bounds for every detection oracle in the Chandra-Toueg hierarchy with respect to four practical models of mutual exclusion. Analysis of the resulting lattice identifies the weakest detector for solving each dining problem, and discovers two points of discontinuity indicating unresolved complexity gaps.
Bio:
Scott Pike is a Doctoral Candidate in Computer Science & Engineering at Ohio
State University. He received his M.S. in Computer Science from Ohio State in
2000, and his B.A. in Philosophy from Yale University in 1996. His research interests
focus on software engineering and distributed computing, and, more concretely,
on scalable approaches to building agile, adaptive, and survivable components
for distributed systems.
Date: Tuesday April 27, 2004
Time: 11am-12:15pm
Location: Woodward 149
Tao Li <tli3@ece.utexas.edu>
Department of Electrical and Computer Engineering
University of Texas at Austin
Abstract:
The Operating System (OS) which manages both hardware and software resources,
constitutes a major component of today's complex systems. Many
modern and emerging workloads (e.g., database, web servers and file/e-mail applications)
exercise the OS significantly. However, microprocessor
designs and (performance/power) optimizations have largely been driven by the
user-level applications. In this talk, I will present the advantages and benefits
of integrating OS component in processor architecture design.
In the first part of my talk, I will show how control flow prediction hardware, which is critically to deliver instruction level parallel (ILP) and pipelining performance on today's highly-speculative and deeply-pipelined machine, can be cost-effectively adapted to significantly improve its speculation accuracy on the exception-driven, intermittent OS execution. In the second part of my talk, I will address the adaptations of processor resources to reduce OS power on today's high-complexity processors, which exploit aggressive hardware design to maximize the performance across a wide range of targeted applications.
Bio:
Tao Li is currently a Ph.D. candidate (in Computer Engineering) at the Electrical
and Computer Engineering Department, University of Texas at Austin. His research
interests include: computer and system architecture, operating systems, energy-efficient
design, modeling, simulation and evaluation of computer systems and hardware
system prototyping.
Date: Tuesday April 20, 2004
Time: 11am-12:15pm
Location: Woodward 149
Andrew Lumsdaine <lums@cs.indiana.edu>
Director, Indiana University Pervasive Technology Labs, Open Systems Lab
Associate Professor, Computer Science Department
Abstract:
Many modern programming languages support basic generic programming, sufficient
to implement type-safe polymorphic containers. Some languages have moved beyond
this basic support to a broader, more powerful interpretation of generic programming,
and their extensions have proven valuable in practice, particularly for the development
of reusable software libraries. Fundamental to realizing generic algorithms is
the notion of abstraction: generic algorithms are specified in terms of abstract
properties of types and algorithms, the specification of which we call "concepts".
Although many languages today have support for "generics", they do
not directly support concepts, making it difficult to fully leverage the potential
of generic programming in modern software construction. This talk reports on
recent work to better understand concepts. Building on previous work by Kapur
and Musser, we provide a language-independent definition for concepts. We show
how this definition can be realized in the context of different programming languages
and discuss how it could be used to address the limitations of existing languages
for generic programming.
Date: Thursday April 15, 2004
Time: 11 am-12:15pm
Location: Woodward 149
Dan Greer
VP/Chief Scientist
Verdasys
Waltham, Massachusetts
Abstract:
The wonderful thing about a small town is that you know everyone; the terrible
thing is that they all know you. The wonderful thing about a national infrastructure
is that you are closely connected to everything; the terrible thing is that it
is all closely connected to you. At the national scale, what are the shared risks
and, hence, the shared solutions? What sort of collective action is desirable?
Are there tools and strategies we can borrow from other fields and do we have
the time to invent?
Date: Tuesday, April 13, 2004
Time: 11am-12:15pm
Location: Woodward 149
Jing Li <jing.li@email.ucr.edu>
Ph.D. Candidate, Department of Computer Science and Engineering, University of
California, Riverside
Abstract:
With the completion of the Human Genome Project, an (almost) complete human genomic
DNA sequence has become available. An important next step in human genomics is
to determine genetic variations among humans and the correlation between genetic
variations and phenotypic variations (such as disease status, quantitative traits,
etc.). The patterns of human DNA sequence variations can be described by SNP
(single nucleotide polymorphism) haplotypes. However, humans are diploid and,
in practice, haplotype data cannot be collected directly, especially in large
scale sequencing projects (mainly) due to cost considerations. Instead, genotype
data are collected routinely in large sequencing projects. Hence, efficient and
accurate computational methods and computer programs for the inference of haplotypes
from genotypes are highly demanded.
We are interested in the haplotype inference problem on pedigrees and haplotype-based association mapping methods for identifying disease genes. We study haplotype reconstruction under the Mendelian law of inheritance and the minimum recombination principle on pedigree data. We prove that the problem of finding a minimum-recombinant haplotype configuration (MRHC) is in general NP-hard. An iterative algorithm based on blocks of consecutive resolved marker loci (called block-extension) is proposed. It is very efficient and accurate for data sets requiring few recombinants. A polynomial-time exact algorithm for haplotype reconstruction without recombinants is also presented. This algorithm first identifies all the necessary constraints based on the Mendelian law and the zero recombinant assumption, and represents them using a system of linear equations over the cyclic group Z2. All possible feasible haplotype configurations could be obtained by adopting the Gaussian elimination algorithm. For genotypes with missing alleles, we develop an effective integer linear programming (ILP) formulation of the MRHC problem and a branch-and-bound strategy that utilizes a partial order relationship (and some other special relationships) among variables to decide the branching order. When multiple solutions exist, a best haplotype configuration is selected based on a maximum likelihood approach. The ILP algorithm works for any pedigree structures, regardless of the number of recombinants, and effective for any practical size problems. We have implemented the above algorithms in a software package called PedPhase and tested them on simulated data sets as well as on a real data set. The results show that the algorithms perform very well.
Haplotype information is much valuable for disease gene association mapping, which is a very important problem in biomedical research. We also develop a new algorithmic method for haplotype mapping of case-control data based on a density-based clustering algorithm, and propose a new haplotype (dis)similarity measure. The mapping regards haplotype segments as data points in a high dimensional space. Clusters are then identified using a density-based clustering algorithm. Z-score based on the numbers of cases and controls in a cluster can be used as an indicator of the degree of association between the cluster and the disease under study. Preliminary experimental results on an independent simulated data set, and on a real data set with the known disease gene location show that our method could predict gene locations with high accuracy, even when the rate of phenocopies is high.
Biography:
Jing Li currently is a Ph.D. candidate in the Department of Computer Science
and Engineering at the University of California - Riverside. He received a
B.S. in Statistics from Peking University, Beijing, China in July 1995 and
a M.S., in Statistical Genetics, from Creighton University in Aug. 2000. He
was a winner of the ACM Student Research Competition in 2003. Jing Li's recent
research interest includes Bioinformatics / computational molecular biology,
algorithms and statistical genetics. He is particularly interested in developing
algorithms for haplotype inference and haplotype-based disease gene mapping.
Date: Tuesday, April 6, 2004
Time: 11am-12:15pm
Location: Woodward 149
Dr. Nancy Andreasen, <nca@unm.edu>
Director, MIND Institute
Abstract:
For many years the cerebellum has been viewed as a cerebral organ primarily
dedicated to coordinating motor activity. During recent years, however, new evidence
has emerged that indicates that the cerebellum may also be a key cognitive organ
in the brain as well. Several models of cerebellar nonmotor or cognitive learning
have been proposed. One model (Keele and Ivry, 1990, 1997) argues that the cerebellum
functions as a clock or timekeeper, based primarily on lesion studies that indicate
cerebellar injury leads to an impairment in the ability to estimate time intervals
or imitate timed rhythm sequences. PET studies recently conducted in Iowa confirm
that a variety of timing tasks produce robust activation in the cerebellum, as
well as the thalamus and insula. Another model has been proposed by Ito (1997).
He argues from cerebellar anatomy and suggests that the cerebellum is composed
of large numbers of units that he refers to as microcomplexes. These provide
an error-driven adaptive control mechanism. Microcomplexes are subunits within
the cerebellum that facilitate its function. They receive dual inputs from mossy
and climbing fibers; the climbing fibers detect errors and act to reorganize
internal connections, while the mossy fibers "drive" the complex. The
microcomplexes function like computer chips or microprocessors-they can be used
to perform a vast array of different functions. They are connected to diverse
brain regions (e.g., multiple different cortical areas) and therefore can play
many diverse roles in brain function, including all types of cognitive processing.
In addition, evidence has also emerged that suggests that cerebellar function
is impaired in schizophrenia. Studies of schizophrenia using the tools of functional
imaging have found a relatively consistent pattern of abnormalities in distributed
brain regions that include the cerebellum. Abnormalities are seen in these studies
in both the vermis and in the cerebral hemispheres in patterns that are task-related.
Patients with schizophrenia have decreased blood flow in the cerebellum in a
broad range of tasks that tap into diverse functional systems of the brain, including
memory, attention, social cognition, and emotion. Vermal abnormalities are more
frequently noted in tasks that use limbic regions (e.g., studies of emotion),
while more lateral neocerebellar regions are abnormal in tasks that use neocortical
regions (e.g., memory encoding and retrieval). It is therefore highly plausible
that the symptoms and cognitive abnormalities of schizophrenia may arise because
of malfunctions in a group of distributed brain regions and that the cerebellum
is a key node in this malfunctioning group of regions.
Date: Thursday, April 1st, 2004
Time: 11am-12:15pm
Location: Woodward 149
Susanne Jul, sjul@umich.edu
Computer Science and Engineering, University of Michigan
Abstract:
Identifying which design constraints -- limitations on what constitutes an acceptable
solution to a design problem -- do or do not apply to a particular design situation
is key to how quickly and how well the design problem is solved. In this talk,
I present a case study that derives a set of design constraints for navigational
design from existing empirical evidence surrounding navigational and spatial
cognition. (I use "navigation" to mean the task of determining where
places and things are, how to get to them and actually getting there, and my
focus is on support for human navigation within the environment under design.)
The study yielded a set of design elements that are key to navigational cognition
along with a set of design principles describing how manipulations of these elements
affect navigational performance. During the talk, I will demonstrate a navigational
design for a spatial multiscale environment (Jazz) that is based on the derived
principles. I will also present results from user testing of this design that
show significant increases in navigational performance, along with significant
decreases in effort, on a directed search task.
The immediate value of identifying design elements and constraints upon them
lies in their explicit use in design. However, I anticipate that the greater
benefit will lie in embedding them in software development tools, such as application
frameworks, so that they are used implicitly by both developers and designers
(future work).
Date: Tuesday, March 30th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Stefan Wolfgang Pickl,
Department of Mathematics, Center for Applied Computer Science Cologne
University of Cologne
Current address:
Department of Computer Science
University of New Mexico
Abstract:
This talk will give an introduction into the challenging field of the optimization
of biosystems applying discrete structures and suitable algorithms. Many optimization
problems can be described and solved with the aid of polytopes exploiting their
geometrical and combinatorial structure. The presentation describes two cases
where polytopes determine feasible sets (Example 1 - economathematics) and where
polytopes are suitable tools ("keys") for optimization techniques (Example
2 - data analysis in the lifesciences). In these fields, a special representation
form of polytopes may be used to construct an algorithm which is able to analyze
and optimize a nonlinear time-discrete system. The underlying theory of the algorithm
bases on the use of polytopes and linear programming techniques such that, successively
only the extremal points of the polytope are regarded. Their topological behaviour
can be used to get suitable decision criteria. Theoretical results are as well
presented as numerical solutions. As an example, the project TEMPI (Technology
Emissions Means Process Identification) is discussed.
Date: Thursday, March 25th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Godmar Back, <gback@stanford.edu>
Department of Computer Science, Stanford University
Abstract:
Single-language runtime systems, such as Java virtual machines, are widely deployed
platforms for executing untrusted code. These runtimes provide some of the features
that operating systems provide: inter-application memory protection and basic
system services. They do not, however, provide the ability to isolate applications
from each other, or limit their resource consumption.
In this talk, I will present KaffeOS, a Java runtime system that provides these features. The KaffeOS architecture takes many lessons from operating system design, such as the use of a user/kernel boundary, and employs garbage collection techniques, such as write barriers. It supports the OS abstraction of a process in a Java virtual machine. Each process executes as if it were run in its own virtual machine, including separate garbage collection of its own heap. The difficulty in designing KaffeOS lay in balancing the goals of isolation and resource management against the goal of allowing direct sharing of objects to enhance performance and scalability.
I will present performance results that show that KaffeOS can be used to effectively thwart denial-of-service attacks by untrusted or misbehaving code, and demonstrate the effectiveness of KaffeOS's sharing model. Finally, I will also discuss what I view should be the next steps in making type-safe language runtime systems ready for use in robust and scalable multi-process environments.
Biography:
Godmar Back works as a postdoctoral researcher with Professor Dawson Engler.
He received his PhD from the University of Utah in 2002. His research interests
lie at the intersection of systems and programming languages. He currently works
on MJ, a system for statically checking Java code. Before coming to Stanford,
he designed and implemented KaffeOS, a Java runtime system that provides process
isolation and resource management for multiple applications in a single JVM.
He has also worked on various OS projects, such as the Utah OSKit and the Fluke
microkernel.
Date: Thursday, March 11th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Jay McClelland, <jlm@cnbc.cmu.edu>
Department of Psychology, Carnegie Mellon University
and Center for the Neural Basis of Cognition Carnegie Mellon and the University
of Pittsburgh
Abstract:
Through the invention and proliferation of written language, it has become second
nature to view utterances as composed of words, words as composed of morphemes
and syllables, and syllables as composed of phonetic segments. In this talk I
will argue from the properties of connectionist models and from facts about language
that have been pointed out by Bybee and others that it might be best to view
all of these units, not as items with psychological reality as units per se,
but as contrivances that have proven useful for the construction of a notational
system that allows for the approximate transcription of spoken language. The
development of explicit theory that eschews all units is a task for the future,
a journey I at least am just beginning. I intend to explore in future work how
far such a journey may lead us and will use this talk to sketch out a few initial
steps in this direction.
Date: Tuesday, March 9th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Shuang (Sean) Luan, <sluan@cse.nd.edu>
Department of Computer Science and Engineering
University of Notre Dame
Abstract:
Computer-assisted radiotherapy is an emerging interdisciplinary area that applies
the state-of-the-art computing technologies to help the diagnosis, design, optimization,
and operation of modern radiation therapy. In this talk, we present some interesting
problems and their solutions in this exciting area.
Intensity-modulated radiation therapy (IMRT) is a modern cancer treatment technique, aiming to deliver a highly conformal dose to a target tumor while sparing the surrounding normal tissues and critical structures. A key to performing IMRT is the accurate and efficient delivery of discrete dose distributions using the linear accelerator and the multileaf collimator. The leaf sequencing problems arise in such scenarios, whose goal is to compute a treatment plan that delivers the given dose distributions in the minimum amount of time.
Existing leaf sequencing algorithms, both in commercial planning systems and in medical literature are all heuristics and do not guarantee any good quality of the computed treatment plans, which in many cases result in prolonged delivery time and compromised treatment quality. In this talk, we present some new MLC leaf sequencing algorithms and software. Our new algorithms, based on a novel unified approach and geometric optimization techniques, are very efficient and guarantee the optimal quality of the output treatment plans. Our ideas include formulating the leaf sequencing problems as computing shortest paths in a weighted directed acyclic graph and building the graph by computing optimal bipartite matchings on various geometric objects. Our new IMRT algorithms run very fast on real medical data (in only few minutes). Comparisons between our software with commercial planning systems and the current most well known leaf sequencing algorithm in medical literature showed substantial improvement. The treatment plan computed by our software not only takes much less delivery times (up to 75% less) but also has much better treatment quality. Our software has already been used for treating cancer patients at a couple of medical centers.
Date: Tuesday, March 2nd, 2004
Time: 11am-12:15pm
Location: Woodward 149
Ken Hopkinson, <hopkik@cs.cornell.edu>
Department of Computer Science, Cornell University
Abstract:
My thesis research has drawn upon the fields of simulation, networking, and distributed
computing to examine the inherent problems and potential solutions in using Internet
technology in real-time environments with a particular focus on the electric
power grid. As recent events have dramatized, the electric power grid is under
increasing strain as demand for energy increases while the existing transmission
infrastructure has remained largely constant. This makes the power grid an attractive
target for study for a number of reasons. It is a large real-time system with
well-established operating requirements. The power grid is also appealing due
to the interest that its constituents have shown in augmenting their infrastructure
with communication to improve its operation. Recent standards such as the Utility
Communication Architecture (UCA) and research efforts such as the use of Wide
Area Measurement Systems (WAMS) in the Western United States are just two major
examples of the active interest in the electric power industry for adopting a
private Utility Intranet based on Internet technology to improve the efficiency
and reliability of the power grid. Early effort has assumed that TCP would be
the underpinnings of any solutions due to its widespread adoption without regard
to the protocol’s performance in real-time situations. This is a problem
representative of a larger issue. The affordability and ubiquitous nature of
the Internet makes its protocols and equipment an obvious choice for any undertaking
involving communication, but Internet technology has not been created with real-time
requirements in mind. In my thesis work, I have focused on four main areas to
gain a better understanding of the issues involved in the use of Internet technology
in real-time applications and their potential solutions:
My talk will discuss an overview of the major results the have resulted from these research initiatives and will present potential directions for future work.
Date: Thursday, February 26th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Constantine "Dino" Pavlakos, <cjpavla@sandia.gov>
Manager, Visualization and Data Department
Sandia National Laboratories
Abstract:
Sandia National Laboratories, together with other DOE laboratories, industry
and academic partners, is developing advanced technologies that are setting a
standard for high performance data exploration and visualization. This work is
driven by requirements for the analysis of very large-scale (multi-terabyte)
complex scientific application data and is closely coupled to laboratory efforts
to provide scalable computing. The strategy for delivering advanced data analysis
and visualization capabilities is based on the themes of scalability, use of
commodity-based clustered hardware, accessibility from the office/desktop, ease
of use, and advanced interfaces and work environments. While working toward a
complete end-to-end system of high performance services that can be accessed
from the desktop, the current focus is on delivery of scalable data services
and scalable rendering, together with the deployment of high-end infrastructure
and facilities.
This talk will present an overview of such activities and related accomplishments at Sandia, together with a discussion of related issues and an architectural characterization of environments that effectively support high performance computing (including a brief description of the environment planned for Sandia's ASCI Red Storm machine).
Date: Tuesday, February 24th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Victor Winter, <vwinter@mail.unomaha.edu>
Computer Science Department,University of Nebraska at Omaha
Abstract:
The control mechanisms offered by strategic programming have been successfully
used to address a variety of problems relating to confluence and termination.
However, the application of strategic programming to problems of increasing complexity
has raised another issue, namely how auxiliary data fits within a strategic framework.
The distributed data problem characterizes the problem posed by auxiliary data.
This problem arises from a discord between the semantic association of terms
within a specification and the structural association of terms resulting from
the term language
definition.
My research is based on the premise that higher-order rewriting provides a
mechanism for dealing with auxiliary data conforming to the tenets of rewriting.
In a higher-order framework, the use of auxiliary data is expressed as a rule.
Instantiation of such rules can be done using standard (albeit higher-order)
mechanisms controlling rule application (e.g., traversal). Typically, a traversal-driven
application of a higher-order rule will result in a number of instantiations.
If left unstructured, these instantiations can be collectively seen as constituting
a rule base whose creation takes place dynamically. However, such rule bases
again encounter difficulties with respect to confluence and termination. In order
to address this concern the notion of strategy construction is lifted to the
higher-order as well. That is, instantiations result in rule bases that are structured
to form strategies. Nevertheless, in many cases, simply lifting first-order control
mechanisms to the higher-order does not permit the construction of strategies
that are sufficiently refined. This difficulty is alleviated though the introduction
of the transient combinator. The interplay between transients and more traditional
control mechanisms enables a variety of instances of the distributed data problem
to be elegantly solved in a higher-order setting. These
ideas are formalized in a higher-order strategic programming language called
TL.
Bio
Victor Winter is an assistant professor in the Computer Science department
at the University of Nebraska at Omaha. He received his Ph.D. in Computer Science
from the University of New Mexico in 1994. From 1995--2001 Dr. Winter worked
at Sandia National Laboratories where his research efforts focused on high-assurance
software development.
Date: Thursday, February 19th, 2004
Time: 11am-12:15pm
Location: Woodward 149
David Kempe, <kempe@cs.washington.edu>
Department of Computer Science and Engineering, University of Washington
Abstract:
A social network - the graph of relationships and interactions within a group
of individuals - plays a fundamental role as a medium for the spread of information,
ideas, and influence among its members. An idea or innovation will appear - for
example, the use of cell phones among college students, the adoption of a new
drug within the medical profession, or the rise of a political movement in an
unstable society - and it can either die out quickly or make significant inroads
into the population.
The resulting collective behavior of individuals in a social network has a long history of study in sociology. Recently, motivated by applications to word-of-mouth marketing, Domingos and Richardson proposed the following optimization problem: allocate a given "advertising" budget so as to maximize the (expected) number of individuals who will have adopted a given product or behavior.
In this talk, we will investigate this question under the mathematical models of influence studied by sociologists. We present and analyze a simple approximation algorithm, and show that it guarantees to reach at least a 1-1/e (roughly 63%) fraction of what the optimal solution can achieve, under many quite general models. In addition, we experimentally validate our algorithm, comparing it to several widely used heuristics on a data set consisting of collaborations among scientists.
(joint work with Jon Kleinberg and Eva Tardos)
Date: Tuesday, February 17th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Karen Devine, <kddevin@sandia.gov>
Discrete Algorithms and Math Department
Sandia National Laboratories
Abstract:
Efficient and effective dynamic load balancing is critical in parallel, unstructured
and/or adaptive computations. Dynamic load-balancing algorithms
assign work to processors during the course of a computation to reflect changing
processor workloads (e.g., due to mesh refinement in adaptive finite element
simulations) or changing data locality (e.g., due to deformation in crash simulations).
The Zoltan library provides a suite of parallel, dynamic load-balancing algorithms,
allowing application developers to easily compare the effectiveness of different
strategies within their applications. Zoltan also provides tools commonly needed
by parallel applications, including data migration functions, distributed data
directories, unstructured communication utilities, matrix ordering, and dynamic
memory debugging utilities. Zoltan's object-based interface and data-structure-neutral
design make it easy to use Zoltan in a variety of applications. This talk will
describe Zoltan's dynamic load-balancing algorithms, design, and usage in applications.
Results from applications using Zoltan will also be presented.
Date: Tuesday, February 10th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Prof. Lawrence Clark, <ltclark@ece.unm.edu>
Electrical and Computer Engineering, UNM
Abstract:
Programmable digital signal processors (DSP) are key enablers of rapidly expanding
wireless markets. The primary handset market, subject to strict
power constraints, also creates substantial process scaling barriers. Nonetheless,
DSP's in particular and hand-held IC's in general, are uniquely
situated to utilize emerging circuit and SOC design techniques to their advantage,
allowing more aggressive processes despite power barriers. This
talk discusses some of these challenges and design opportunities.
Date: Thursday, March 4th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Christian Poellabauer, <chris@cc.gatech.edu>
Department of Computer Science, Georgia Institute of Technology
Abstract:
The increasing number of network-enabled systems and the growing complexity of
distributed applications pose numerous challenges for the management and provision
of Quality of Service (QoS). Particularly in resource-scarce environments, such
as mobile wireless systems, adaptation of applications and system-level resource
management are used to provide users with the performance and qualities they
need. The distributed management of Quality of Service has been the focus of
intensive research efforts and has led to a multitude of techniques at the hardware,
network, system, or application layer. However, if multiple such techniques are
deployed in a system, an integrated approach to QoS management has to be chosen,
to ensure optimal results and to prevent adverse effects resulting from competing
techniques.
In this talk, I will address how the adaptation of applications and the management
of multiple system constraints can be coordinated to efficiently provide users
with the QoS they need. This coordination is supported by Q-Fabric, a collection
of operating system extensions, which provide the framework to deploy feedback-based
integrated QoS management for distributed applications. Specifically, my talk
will focus on the integrated approach to distributed energy management for mobile
multimedia applications. The goal is to efficiently deploy and coordinate novel
energy
management techniques at different layers of a system, while carefully balancing
the user-perceived QoS with the system's energy consumption.
SHORT BIO:
Christian Poellabauer is a Ph.D. candidate in Computer Science at the Georgia
Institute of Technology, expecting to graduate in May 2004. He received his Master's
degree in Computer Science from the Vienna University of Technology. His research
interests are in the area of experimental systems, including real-time systems,
operating systems, pervasive computing, and mobile systems. He is a recipient
of an IBM Ph.D. Research Fellowship.
Date: Thursday, January 29th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Patrick Bridges, <bridges@cs.unm.edu>
Department of Computer Science, UNM
Abstract:
Effective technical writing is an essential skill for every graduate student
and virtually every student has had at least one class as an undergraduate on
writing essays of various sorts. Unfortunately, these classes generally teach
very little that is of practical relevance to writing technical papers and reports.
In this colloquium, I will give students a brief introduction to the issues of
technical writing and how they differ dramatically from the essay writing most
students learned as undergraduates. In particular, I will focus on practical
tips for organizing technical papers and presenting technical information, give
example of well-organized and poorly-organized techncal writing, and also give
some advice on effective language use in technical writing. In addition, I will
give a few tips on presenting technical work to large groups, another essential
skill that is frequently not taught directly.
Date: Tuesday, January 27th, 2004
Time: 11am-12:15pm
Location: Woodward 149
Bernard Moret, <moret@cs.unm.edu>
Department of Computer Science, UNM
Abstract:
Graduate students are engaged in both learning and research. In each area, they
have a responsibility to themselves and to their peers -- a responsibility to
conduct themselves in a professional manner according to accepted principles
of ethics. We will review the guiding principles in the university setting, their
reason for existence, their consequences in terms of daily behavior in the classroom,
in the laboratory, and in the scholarly community in general. We will also briefly
review some of the many documents that the University provides in this area for
students and faculty that can help guide their behavior in teaching and research.
News & Events | People | Academics | Research | Jobs | Companies | Visitors | Computer Facilities | Home
