EECS500 Spring 2012 Seminar

David Johnston
Probabilistic Multicast Trees
Case Western Reserve University
White Bldg., Room 411
11:30am - 12:30pm
March 6, 2012

Smart phones, movies on demand, regulated industrial process information; the thirst for data access has never been greater and will only continue to grow. Our network infrastructure must continue to evolve to meet the ever increasing demand for data.  This is especially true when many devices demand the same data at the same time.  Since improvement in pure network bandwidth capabilities is only part of the solution; many researchers have investigated efficient transfer of
information and a variety of solutions have been proposed.  One popular solution is the multicast overlay network which can be described as a tree.  Many single multicast tree solutions and multiple multicast tree solutions have been developed; however, we still need to make these
solutions more efficient.
This research will focus on the use of multiple multicast trees.  This is often referred to as striping where one multicast tree is one stripe.  Typically in these models, each multicast tree is used equally; however, not every multicast tree has the same performance.  This new method, called Probabilistic Multicast Trees (PMT), will build upon other multiple multicasting models.  Given a number of multicast trees from source node to destination nodes using the multiple multicast
trees, a probability of usage is calculated for each of the multicast trees with the highest probability for packet transmission assigned to the most efficient tree.  Feedback is generated from destinations to source to provide the input for the probability calculations.  For a given packet transmission, one tree will be chosen randomly based on the tree’s collective probability distribution and the packet will be sent on this tree.  The trees’ probability distributions will be calculated and continually adjusted based on feedback of each tree’s performance.  Packet transmission continues with periodic adjustment to the multicast tree usage probability based on multicast tree feedback measurements.  It is the use of feedback and probability adjustments that makes PMT more efficient than other methods.


Mr. Johnston is a principle engineer at Rockwell Automation in Cleveland, Ohio.  He has over 32 years of experience in factory automation and is currently a communication architect for the Logix
Controllers family of products.  Mr. Johnston is named as co-inventor on 11 U.S. patents.  He has earned his Bachelor of Science in Electrical Engineering from the University of Toledo and his Masters of Computer and Information Science from Cleveland State University.  He has recently completed his Ph.D. in Computer Engineering at Case Western Reserve University.