Mikhail Chalabine
Programming Environments Laboratory (PELAB)
Dept. of Computer and Information Scinence (IDA)
tel.: +46 13 28 66 18
email: mikch@ida.liu.se
September 2005
M.Sc. Thesis in Modeling and Simulations, Computer
Networks and Parallel Programming (20p.)
Parallelization of a Centralized Spanning Tree Discovery Algorithm
C
ontemporary routing protocols predominantly seek routes minimizing hop count. Such a
strategy appears misleading, however, under intermittent decline in link quality caused by
traffic bursts or other transient factors. The shortest path, in this case, may oblige to transmit
over the links with the highest congestion. The issue is crucial, e.g., in multicasting where erroneous
route discovery results in sub-optimal spanning trees.
Contemporary approaches optimize multicasting either for energy consumption, improving packet loss
ratio or spurring maintenance in ad-hoc networks. In your work you will address an exigent and yet
abandoned problem of multicasting optimized for throughput. We have suggested a centralized greedy
spanning tree discovery algorithm by relaxation that circumvents links with the lowest bandwidth on
demand. This can be compared to the state of the art protocols based on Prim’s and Dijkstra’s methods.
In your work you will:
1. Develop a simulation of the local (IDA) department network (+random networks) in the
NS -II simulator based on the suggested algorithm. You will then discuss possible gain/loss
in throughput as a function of links’ capacity variations by running the simulation and
comparing the results to the ones given by the contemporary techniques.
2. Adapt an existing parallel implementation of the algorithm and test it on a supercomputer
located at the National Supercomputer Center. In these experiments you will measure
gain/loss in speed compared to the sequential version of the algorithm.
Further information: Mikhail Chalabine, mikch@ida.liu.se; Dr. Christoph Kessler, chrke@ida.liu.se
Prerequisites: Background in Parallel programming, e.g., TDDC78, Concurrent programming, Data
structures and algorithms so as good knowledge of C / C++.
Post address:
Linköpings Universitet
IDA, PELAB, 581 83 LINKÖPING
Telephone:
013 28 6618
Telefax:
013 14 22 31
Email:
mikch@ida.liu.se