Otto J. Anshus, John Markus Bjørndalen, Ole Martin Bjørndalen, Daniel Stødle, Ken Arne Jensen

University of Tromsø

February 18th 2003

We propose to use a distributed and parallel robot system as an instrument to demonstrate the principles and practice of distributed computing in general, and high performance parallel computing in particular.

Overall, the system is small enough to be portable, but complex enough to provide for an interesting distributed system with specialized services including a cluster for high performance computing. The system is visual, and invites the audience to speculate how it is done. This will give many opportunities to explain how the system is built, and why it is built as it is. In particular, the role of the compute cluster will be simple to demonstrate by running two demonstrations, one with the cluster, the other without. The behavior of the robots will dramatically change when they must wait longer for solutions to time critical computations.

Associate Professor Otto J. Anshus and staff at the Department of Computer Science, University of Tromsø, have built a working prototype of the proposed robot arena. The goal of this project is to create an enhanced but portable version to be used both in university courses, and at exhibitions, open days and school visits by NOTUR and the University of Tromsø.

The prototype has been used for mandatory student projects and in three robot competitions at the University of Tromsø. It is our experience that the robot system creates a great deal of attention from people both outside and inside of the academia, and furthers the interest in the established solution and in the technology behind. An enhanced portable version will provide NOTUR, the University of Tromsø, and involved vendors with a unique opportunity to demonstrate the visions, demands and technology that make distributed, and in particular high performance computing, so vital for the future.

The High Performance Distributed Computing (HPDC) group at the Department of Computer Science, University of Tromsø, does research and teaches courses on operating systems, computer architecture, distributed systems, and parallel cluster architectures and programming. The research is “systems research” where actual systems are designed, implemented, and experimented on to identify and characterize new ways of configuring and understanding the behavior of complex distributed and parallel systems. In particular, the scaling of applications to utilize hundreds of processors and the behavior of such systems has been intensively investigated the last few years. Several advanced courses have been developed to both support the research efforts, and produce students with deep understanding of distributed and parallel systems. As a part of the advanced course on cluster architecture and programming, we have the last three years developed a system where the students program robots, and supporting computers. The students use computers with different performance characteristics, and must learn to utilize the resources in a distributed system in practice. In particular, they must find ways of utilizing a compute cluster to support the robots with compute performance. The robots need lots of processing cycles to solve the given tasks, including finding the position of other robots when their position has been encrypted to avoid detection.

Robots with small on-board computers cooperate or compete inside an arena according to a set of rules. Separate support computer, one per robot, enhances the robots low processing, memory, and I/O performance. A compute cluster and a file server provide even higher performance. The robots are under constant surveillance by a positioning computer with a video camera. The camera positioning system enhances the robots very limited on-board sensors, and provides the robots with positioning information. To report the progress and the state of the system, a scoreboard computer monitors many activities, and report them on a scoreboard through a projector. A control and management computer starts, stops, and manipulates the system according to user input. Finally, an infrastructure computer makes the system independent of other infrastructures and networks by providing network services (DNS, DHCP, gateway to other networks).

Figure 1: Technical solution

The robots communicate using infrared. As collisions will corrupt the communication, the camera position computer closely controls all infrared communication. All robots and laptops must send their infrared communication to this computer for multi or broadcast relay. All computers except the robots communicate through a wireless network.

An enhancement is to use small handheld computers (PDA), and let the audience participate through these. The PDAs will communicate with individual robots through the control and management computer.

The goal of this project is to establish a fully functional portable robot arena environment comprised of:

·        Six LEGO robots (including one backup).

·        Five laptops, one per robot.

·        One laptop for the camera.

·        One laptop for the scoreboard.

·        One laptop for control and management.

·        One laptop as file server.

·        One laptop for DNS, DHCP, and gateway network services.

·        Four laptops as development systems for all software.

·        A blade cluster comprised of 20 blades

·        A high definition camera, mounted in the frame centered over the arena.

·        A 100BaseT switch or router based networking backbone.

·        An 11 Mbps radio LAN base station.

·        Six handheld computers (including one backup) with radio LAN capabilities.

·        Two portable video projectors (for the scoreboard).

·        A collapsible lightweight frame in aluminum with a fixed mounting-point centered 2.20-2.40 meters over the arena.

·        Specially fitted, light and strong carrying cases for easy transportation (Type “Tenba Air Cases”).

·        Spare high capacity batteries for all laptops.

The proposed system is visually distributed. In our opinion, this is vital in being able to make it clear for an audience the distributed and parallel nature of the system. Even if a single powerful server can handle some functions, this is contra productive with regards to making the system both technically and pedagogically interesting and useful.

The proposed system has a synergistic effect on several activities and goals:

·        The system demonstrates the cooperation between computers with low, medium and high performance. The system demonstrates that clusters are needed to enable lower performance computers to function better.

·        It is easy to raise the audience’s interest in other sciences including mathematics. The robots navigate using simple mathematics, and they utilize the cluster to do more complicated computations like parallel matrix multiplication and differential equations. The roots demonstrating an actual use of mathematics answering the typical question from pupils: “why do I have to study mathematics?”

·        It attracts students to work and develop the system. This both supports the CS Department, and provides students with the necessary knowledge for demonstration purposes, and with deep knowledge about cluster programming.

·        The system is fun both to work on, and to look at. This benefits both developers and the audience.

·        The system is visual. The computers used will get a lot of attention, and suitable posters identifying their brand, nature and tasks will certainly be beneficial.

The system can visually demonstrate robots cooperating or competing. It is easy to point out the division of labour between the individual computers, and discuss distributed and parallel principles and practice.

The scoreboard has certain components of animation, and this can be further expanded. Most computer games today have a very good graphic component. However, what is not common is a combination of graphics, actual physical robots, and distributed and parallel computing. The proposed system is unique and is bound to create interest.

The technical description of the advanced Robo-2002 competition at the Department of Computer Science last year is enclosed. A video from Robo-2001 has been enclosed to demonstrate what was done in 2001. Both documents and more can be found at http://www.cs.uit.no/~otto/robot/ selected from the web pages for the course given in 2001 and 2002.

o       All laptops should at minimum have XGA, P4, 2GHz, 1GB RAM, USB, firewire, 40GB HDD, 10/100BaseT, DVD.

o       All laptops should at minimum have 802.11b wireless network (up to 11Mbit/sec, 2.5GHz). A wireless 802.11a network is even better (up to 54Mbit/sec, 5GHz).

o       1-way with Intel Xeon 2 GHz processor.

o       1 MB of memory.

o       40 GB of storage.

o       10/100baseT Ethernet connection.

o       Linux compliant.

For additional information selected from the course pages in 2001 and 2002 about the robot trials, including a video-presentation done at the ROBO 2001 trial, check out the following web page:

[../photogallery/photo26003/real_p.htm]

This site was last updated 05/06/03