http:^^www.cs.umd.edu^grad^about.html

This data set contains WWW-pages collected from computer science departments of various universities

<H3>Chau-Wen Tseng
<!WA30><img src="http://www.cs.umd.edu/~tseng/tseng.gif" alt="" align=right>
</H3>
</a>

<P>Dr. Chau-Wen Tseng won an NSF CAREER award in Spring 1996,
which will allow him to pursue research in the area of efficient
machine-independent programming of high-performance multipro-
cessors.  Parallel computing can provide the next great leap in the
computation power scientists and engineers need to solve many
important problems.  Multiprocessor workstations are becoming
common and already provide a valuable resource for scientists in areas
such as physics, biology, and chemistry.  Experience has shown that
simply finding parallelism is not always sufficient for obtaining good
performance from today's multiprocessors.  The goal of this project is
to develop advanced compiler analysis of data and computation
decompositions, thread placement, communication, synchronization,
and memory system effects needed in order to take advantage of
performance-critical elements in modern parallel architectures.
Locality and interprocessor communication are the key performance
issues for multiprocessors.  To achieve high performance, the compiler
will apply communication analysis to determine sources of commu-
nication and guide optimizations for locality and communication.  The
compiler follows two basic guidelines.  First, it uses compilation
techniques for message-passing machines to retain most of the benefits
of explicit messages.  Second, it exploits architectural and operating
system support available in shared-memory multiprocessors to im-
prove flexibility and performance.  A novel characteristic of the
compiler will be its ability to take advantage of the multiple coherence
protocols and hybrid message-passing support found in software
Distributed-Shared-Memory (DSM) systems and Flexible-Shared-
Memory (FSM) machines.

<br clear=all>
<!WA31><a href="http://www.cs.umd.edu/~rich/">
<H3>Rich Gerber
<!WA32><img src="http://www.cs.umd.edu/~rich/gerber.gif" alt="" align=right>
</H3></a>

<P>The TimeWare group is currently carrying out projects in the areas of
real-time software development, automated verification and digital
video systems.  The real-time software project is called "end-to-end
design;" its objective is to automatically map high-level, end-to-end
timing requirements into a fully realized, multi-streamed
implementation.  Real-time designs are entered in terms of task graphs
possessing end-to-end requirements (i.e., delay, jitter, etc.); and
intermediate data rates and buffer sizes are then maintained
parametrically, in terms of equations based on the high-level design.
Once the hardware-specific details are known, the application is
integrated, and the intermediate parameters are automatically
calibrated to achieve the end-to-end requirements.  The payoff is that
software designers can have the buffer sizes and intermediate data rates
assigned for them -- thereby minimizing the degree of low-level tuning
required.  The verification project consists of automatically checking
large specifications for subtle safety and liveness errors by compiling
individual tasks into simple state-transition models and
compositionally checking the entire program for nonconformance to
its specification. Thus the verification is done in an iterative, piece-by-
piece manner, in that local analysis is first performed on the individual
tasks, and as tasks are composed, more analysis is carried out.  This
allows progressive deletion of states that are known to disprove the
specification, so that the generated state-space is kept to a minimum.
Work on media systems includes applying static and dynamic tuning
solutions to help master and then play back stored digital video.  Static
tuning takes place during the production phase; it is the process of
adjusting the video's intrinsic quality before it is exported.  The group
has studied results of many different static tuning alternatives by
altering key parameters and then charting their effects at playback.
Dynamic tuning occurs during playback itself; the idea is to process a
video stream as smoothly and deterministically as possible.  OS-level
software built by the group supports this; it periodically estimates the
playback requirements of a particular video, and then allocates buffers,
prefetch window sizes, IO bandwidth, and CPU utilization so that the
computer can best meet the video's requirements.  This technique
significantly outperformed the movie-playing procedures supplied by
Apple's Quicktime API.

<br clear=all>
<!WA33><a href="http://www.cs.umd.edu/~hjs/">
<H3>Hanan Samet
<!WA34><img src="http://www.cs.umd.edu/~hjs/samet.gif" alt="" align=right>
</H3></a>

<P>The representation of spatial data is an important issue in computer
graphics, computer vision, geographic information systems (GIS), and
image processing.  Once the representation has been chosen, users
must be given the ability to access it, and most importantly, perform
operations on it.  The utility of this data is maximized if it can be
integrated into a database management system.  This is a difficult
problem as most conventional systems in use today are mainly
designed to deal with alphanumeric data.  Our approach to solving this
problem is based on the observation that the problem is really one of
sorting. The difference from other approaches is the realization that the
geometric data must be sorted on the basis of its extent (i.e., the fact that
it occupies space) and with respect to the space that it occupies.  This is
instead of parametrizing the spatial data and treating it as points in a
higher dimensional space which is what is done by many researchers.
Representations that take the extent of the data into account enable us
to perform proximity queries efficiently. We are working on the
integration of non-point representations of spatial and image data as
well as nonspatial data into a conventional database management
system.  This research is backed up by the QUILT GIS which is a
working geographic information system and the SAND system for
integrating spatial and nonspatial data.  A principal goal is to be able to
extend this system to handle arbitrary spatial indices rather than just
arbitrary spatial data types.  Another goal is the development of a query
optimizer which takes into account characteristics of the spatial data
and chooses an efficient execution plan for the queries.  One of the
main results of this research has been the development of a browser
that enables posing queries that combine spatial and nonspatial data.
Most noteworthy is the user interface which enables spatial queries to
be specified graphically instead of requiring the use of SQL.  Work is
also being conducted in image databases to deal with symbolic image
and the development of data-parallel representations and algorithms
for spatial problems.

<br clear=all>
<!WA35><a href="http://www.cs.umd.edu/~tripathi/">
<H3>Satish Tripathi
<!WA36><img src="http://www.cs.umd.edu/~tripathi/tripathi.gif" alt="" align=right>
</H3></a>

<P><i>Mobile/Wireless Networking</i>:  Mobile computers equipped with wireless
communication devices frequently change point of attachment to the
network.  Providing continuous networking services to mobile hosts is a
challenging task.  We are developing communication protocols to provide
location independent networking services to such mobile hosts. This work
primarily involves designing MAC protocols for wireless channels,
developing packet routing schemes for mobile hosts and writing applications
for mobile clients.  <i>Multimedia Networking</i>:  Multimedia applications,
such as video-on-demand, video conferencing, etc., generate hundreds of megabits
of time sensitive data, posing serious problems to the current networking
infrastructure. We are developing protocols for real-time transport of
video and audio data over high-speed ATM networks. The work involves:  1)
design and implementation of protocols at all layers of the protocol stack
from network interface drivers to transport protocols, and 2) developing
multimedia applications such as multimedia conferencing, distributed
learning, etc.  <i>Testbed</i>:  The networking laboratory is equipped with
state of the art equipment. At present the testbed consists of Thinkpads 750C,
three RS/6000 power servers and several IBM PCs.  Network support includes
1MB/sec Infrared Wireless LAN, 100MB/sec high-speed ATM LAN and Ethernet.
The testbed is also equipped with specialized hardware for video/audio
capture, compression and playback.

<P><HR>

<H1 ALIGN=CENTER>Recent Seminars and Courses of Interest</H1>

<H2 ALIGN=CENTER>Distributed and Concurrent Systems</H2>
<H2 ALIGN=CENTER>Instructor:  Pete Keleher</H2>

<P>This course is intended to be a general systems survey course.
However, the central thread is high-performance distributed systems. What
hardware support do such systems need, should it all be in hardware? The
primary running examples that are used are distributed shared memory systems,
(mostly) software systems that present the abstraction of shared
memory to a collection of workstations connected by general-purpose
interconnect.  Such systems are becoming commonplace in the
research community, but have yet to achieve the kind of performance
and sophistication that causes the marketplace to listen.
<ul><li>First communication services and abstractions provided by
current operating systems are considered.  What are the important
elements?  What factors are no longer as important as when UNIX was
designed nearly 25 years ago? Do the demands of distributed
applications fit in with the concept of general-purpose machines?
<li>Then papers describing high-performance memory models and
several examples of software DSMs are read. Primary emphasis is on
the different approaches taken by different projects, and where we
think they will go in the future.
<li>Then several parallel tools are discussed, emphasizing the
compiler world's mistaken belief that they're going to put all of us out
of a job.
<li>At the end of the semester, there is a rapid survey of the most
interesting hardware projects currently going on in both industry and
academia.</ul>

<p><H2 ALIGN=CENTER>Architecture of Object-Oriented Database Systems</H2>
<H2 ALIGN=CENTER>Instructor:  Michael Franklin</H2>

<P>Recent years have seen a dramatic increase in research and development
activity in the area of object-oriented database management systems
(OODBMS).  There are now a number of commercial offerings in this area, and
these systems are beginning to gain real acceptance for certain classes of
commercial applications (e.g., CAD/CAM and CASE).  This emerging generation
of database management systems is being deployed primarily in distributed,
workstation/server-based environments.  The combination of distribution and
object-orientation gives rise to significant challenges and performance
opportunities in many areas, including:  distribution of function,
replication and caching, fault tolerance, concurrency control, clustering,
query processing, persistence, and programming language integration.  This
seminar began with a survey of some of the basic issues in distributed
databases and object-oriented databases.   The bulk of the seminar then
focused on the investigation of the state-of-the-art with respect to the
challenges listed above. Finally, some possible future directions, such as
the merging of object and relational technologies and the impact of mobile
computing, were discussed.

<P>Topics Covered:
<UL>
<LI>Overview of Object-Oriented Database Systems:  model wars - manifestos, evolution vs. revolution, etc.
<LI>Architectural Issues:  distribution of function;  cache consistency and concurrency control;  crash recovery;  client-server vs. peer-to-peer; alternative transaction semantics
<LI>Object Management:  object representation (e.g., pointer swizzling);  memory-mapped persistent architectures;  indexing; clustering; distributed garbage collection
<LI>Query Processing;  Performance and Benchmarks;  Existing Systems
<LI>Future issues:  mobility; object-relational systems; utilizing idle resources; dealing with huge databases
</UL>

<P>Readings:  Recent papers from SIGMOD and VLDB proceedings, etc.;some survey and background articles.


<p><H2 ALIGN=CENTER>Computer Graphics</H2>
<H2 ALIGN=CENTER>Instructor: Dave Mount</H2>

<P>This course provided an introduction to the principles of computer
graphics, that is, the creation and manipulation of computer generated
images.  The course covered a wide array of topics from the lowest level
issues of rasterization (how to draw lines and circles one pixel at a time)
up to shading and hidden surface removal.  Emphasis was placed on the
mathematics, data structures, and algorithms needed to perform these