| |
Research Statement
As an undergraduate at the Indian School of Mines, I worked on the detection and characterization of congestion in Delay Tolerant Networks using a cognition based mechanism. In future I wish to continue exploring issues that lie at the intersection of networking and artificial intelligence, and contribute in meaningful ways to emerging network architectures.
Project Mentors:
Scott Burleigh(NASA/JPL), Adrian Hooke(NASA/JPL), and Dr Vint Cerf(JPL).
|
|
Outline
As part of my undergraduate research, I have been exploring congestion issues in Delay Tolerant Networks, with particular emphasis on the InterPlanetary Internet. I have focused on answering the following questions:
1. How do we define congestion in a Delay Tolerant Network (DTN)?
2. What are the possible causes of congestion in DTNs?
3. What are the parameters that could be used to predict congestion in a DTN?
4. How can the parameters be related to obtain a function that gives the likelihood of congestion?
5. What are the steps to be taken to avoid congestion? What steps have to be taken by a DTN to recover once congestion has set in?
6. Can congestion be avoided, or at least mitigated, by backward propagation of hop-by-hop flow control?
7. In what way would the congestion and flow control policy differ for different types of DTNs such as InterPlanetary Internet, battlefield tactical networks, sensor networks and undersea networks?
8. At what layer should hop-by-hop flow control be exercised?
9. What kind of non-local information could be used by a node for computing the likelihood of congestion?
Due to large speed of light delays and an extreme environment in interplanetary space, information at a node is likely to be delayed, incomplete and even conflicting at times. Besides, it is highly desirable that nodes in space be able to anticipate and fix problems in the network with minimum human intervention. This has encouraged me to explore a cognitive approach to congestion detection and characterization. An intelligent mechanism could allow the network to be tolerant of erroneous information and learn from failures and successes.
My research work on congestion issues in DTNs started in Dec 2002. I started by studying past research on congestion and flow control. During the summer of 2003, I came up with a proposal to use a cognitive approach for congestion detection and characterization in DTNs. The proposal was peer reviewed on the DTN-INTEREST mailing list.
Since then, I have tested the effectiveness of different machine learning algorithms in predicting congestion using sample data generated by a computer program. (The machine learning algorithms implementation is available online as open-source software called Weka).
|
|
| | |