The systems research
projects on High
Performance Distributed Systems at the Department of Computer Science, University of Tromsø, have researched
basic distributed, parallel and GRID-style systems since the mid 1980's. We have developed architectures, tools, and
mechanisms to configure, instrument, monitor, and visualize the
behavior and output of distributed and parallel applications. The
purpose is to understand and develop enabling technologies for making
distributed and parallel applications scale well with respect to
performance and failures when the size of the problem and the number of
participating computers increase. In our opinion this is critical to
the successful development and use of future really large scale systems
with thousand, and eventually millions of cooperating components.
A structured distributed shared memory
system, PastSet, has been designed, implemented and developed.
Low-level communication mechanisms have been studied extensively and
high-performance, highly efficient implementations have been developed.
Contrary to conventional wisdom, our shared memory system has been
shown to outperform current message-based designs for a variety of
applications. Recently, substantial work has been done on scheduling of
computations onto nodes, and of combining communications and
computations effectively. An experimental system, PATHS, has been
developed and experimented with using several applications. ELCIRC, a 3D model for the Columbia River
estuary and plume based on the solution of the shallow water equations
for continuity, momentum, and salt and heat balances, using
Eulerian-Lagrangian finite volume method has been parallelized for a
small grid (22 CPUs),
demonstrating a running time reduction of 93% over the sequential computation running
on one CPU. At EuroMicro 2002
we presented work that demonstrated the performance of several other
applications running in a wide-area grid setting. We are presently
investigating the scaling properties of the Dalton application, and how it can be
improved further.
Our approach is
primarily experimental in order to capture the complex interactions
between real workloads and modern technology. We design, implement,
measure, and evaluate software and hardware. We are presently using
clusters of commodity computers as test beds. This allows us to do
controlled, repeatable and non-intrusive experiments using hundreds and
even thousands of computers. In our opinion we need to understand such
small scale systems of up to a few thousands of computers before we can
start to understand larger systems of up to millions of computers.
We have secured funding for a cluster and a
large high-resolution scalable display wall from the Norwegian Research
Council (NFR) and the University
of Tromsø,
and will establish this research vehicle during 2003/2004. The display and cluster will be
used, among other things, for furthering our work on monitoring,
visualizing, debugging, and tuning applications that run on an
experimental Danish-Norwegian grid.
The research described here have been done
by three tenured faculty, an adjunct associate professor, and several
Ph. D., master, and undergraduate students.
We have published more than 20
papers and two dissertations over
the last five years; all in the area of design and implementation of
infrastructure architectures and mechanisms for small and large scale
distributed and parallel systems including grid-style of computing.
Research Areas and Topics
Our research and interests spans several areas and topics of
computer science, including Operating systems, Distributed systems,
Parallel systems, Cluster-based computing, Networks, Graphics, Vision,
Ubiquitous Computing,, Internet Systems and Infrastructure, Systems
Reliability, Dependability, Fault Tolerance, Recoverability,
Human Computer Interaction, Immersive systems, Robotics
Projects
Seminars and Workshops
Programs
As a part of our research we are looking for and pursuing
cooperation with universities, laboratories and industry in many fields
including:
- Petroleum related simulation and visualization
- Visualization of scientific data and large data sets
- Human-Computer interfaces for immersive computing using
display walls
- Application domains for display walls (including teaching,
architecture, tele-medicine, interactive displays/exhibitions)
- E-medicine
- Hydrological simulation and visualization
- Weather simulation and visualization
- Sensor Networks
- Robots
- Bio informatics
Computer Resources
A display wall and cluster project proposal
has secured funding from the Norwegian Research Council and the University of Tromsø, and is comprised
of:
- Tiled Display Wall: 28 projectors and a display
cluster of
29 PC' (28+1 Dell 370 P4 Prescott, 3.2GHz, 2GB RAM, 1Gbit Ethernet)
- VNC Server (for main
display wall): Dell PowerEdge 2800, with 2 Xeon 3.8GHz/2MB
800FSB, 8GB Dual Rank DDR2 Memory (4x2GB), 146GB SCSI Ultra320
(15,000rpm) 1in 80 pin Hard Drive x 2
- Small 2*2 LCD display: 4 Dell LCD displays, and a
display
cluster of 5 PC's (4+1 Dell 370 P4 Prescott, 3.2GHz, 2GB RAM, 1Gbit
Ethernet)
- Compute cluster: 40 Dell 360 Prescott 3.2GHz, 2GB
RAM,
1Gbit Ethernet, 48 port HP switch
- Compute cluster: 40 Dell 370 Prescott 64 EMT,
3.2GHz,
2GB RAM, 1Gbit Ethernet, 48 port HP switch
- NAS server:
vncnas00, about 2.8TB RAID disk.
In the Display
Wall Laboratory, a number of computers are used for development and
experiments, including:
- Dell 360 P4
Prescott, 3.2GHz, 2GB RAM, LCD display 1600*1200
- Dell 360 P4
Prescott, 3.0GHz, 2GB RAM, LCD display 1600*1200
- PowerMac
Dual G5, 2.5GHz, 4GB RAM, 2*30' displays 2560*1600
- Fujitsu
Tablet PC for handheld input on the display wall
- Wireless
keyboard(s) and mice
- A number of
PC's used as the laboratory infrastructure for services like shared
file storage, and control services for the display wall.
In addition we
have three older clusters:
- Cluster of four eight-way
computers, HP Server Pentium Pro
- Cluster of eight four-way
computers, HP Server Pentium Pro
- Cluster of twenty two-way
computers, HP Vectra
Finally, we have access to clusters in Denmark
and at the Computing Centre, University of Tromsø
People
Presently the following persons are involved in this part of
the departments systems and architecture research:
- Otto Anshus, professor
- John Markus Bjørndalen, associate professor
- Tore Larsen, associate professor
- Brian Vinter, Professor, Copenhagen University
- Phuong Hoai Ha, Post doctor
- Lars Ailo Bongo, University "stipendiat", Ph.D. student
- Daniel Stødle, NFR "stipendiat" (SHARE Project), Ph.D. student (also: NOTUR ET Cluster project, Robot system)
- Espen Johnsen, NFR "stipendiat" (SHARE Project), Ph.D. student (also: robots)
- Yong Liu, NFR "stipendiat" (SHARE Project), Ph.D. student
- Tor Magne Stien Hagen, "stipendiat", Ph.D. student.
Papers
List of papers (comprehensive list to be installed
soon, temporarily please see http://www.cs.uit.no/~otto
and http://www.cs.uit.no/~johnm/
)
Information about Grid, Parallel, Supercomputing, High
Performance Computing Resources
Ph. D. students, active
- Lars Ailo Bongo
- Daniel Stødle
- Espen Johnsen
- Yong Liu
- Tor-Magne Stien Hagen
Ph.D. students, graduated
- John Markus Bjørndalen, 2003
- Brian Vinter, 1999
Master and Diploma students, active
- Bård Fjukstad
- Thomas Heggelund, Multi-window cluster debug
- Thomas Samuelsen, Distributed Scene Graphs for Display Walls
Bachelor and Term projects, active
- Bjørn-Hugo Nilsen, Gesture Interface for a Large
High-Resolution Display
- Oleg Jakobsen, Mobile objects in display wall 3D
environments
- Ståle W. Nilsen, Mandelbrot-based tool for
demonstration of parallel and distributed computations
- Tom Arild Jakobsen, Competing Virtual Robots on a Display
Wall
- Webjørn Larsen, TileSnakes: Display wall control and
collaboration using mobile phones
Master and Diploma Students, graduated
to be expanded |
