Probabilistic Analysis of Message Forwarding Louise Moser and Michael Melliar-Smith University of California, Santa Barbara
Message Forwarding Direct multicasting of messages to many nodes is expensive and might be infeasible, if the network is large Source node transmits its message to a small number of nodes Each such node retransmits the message to other nodes, chosen at random Spreads the load across multiple nodes Introduce a probability of message forwarding Limit the number of levels of message forwarding It is easy to calculate an upper bound on the number of nodes reached, when duplicate nodes are ignored July-August 2013ICCCN 20132
Forwarding in a Finite Network n 11 n 12 n 13 n 14 n1n1 n2n2 n3n3 n4n4 n 21 n 22 n 23 n 24 n0n0 July-August 2013ICCCN 20133
4 The pdf Algorithm Input of the algorithm: – n: number of nodes in the network – c: number of nodes to which a node forwards a message – f: probability with which a node forwards a message – l : number of levels of message forwarding Output of the algorithm: – pdf for the number of nodes reached at level l – expected number of nodes reached at level l – pdf for the number of nodes reached up through level l – expected number of nodes reached up through level l July-August 2013ICCCN 2013
Probability of a Duplicate Node Let N be a set of nodes having cardinality n A be a subset of N having cardinality a B be a subset of N having cardinality b where a ≤ b The pdf p(k), 0 ≤ k ≤ a, that A ∩ B has cardinality k, is given by: July-August 2013ICCCN 20135
Calculating the pdfs July-August 2013ICCCN D l -1 S l -1 C l k SlSl DlDl sprev[j] scurr[j] pdf of number of distinct nodes at level l -1 and all prior levels pdf of number of new nodes at level l -1 pdf of number of new nodes
Upper Bound Analysis Upper bound on the number of nodes reached at level l UB s = min(c l, n) Upper bound on the number of nodes reached up through level l UB d = min(1 + c + c 2 + … + c l, n) = min( (c l+1 – 1) / (c – 1), n) if c > 1 July-August 2013ICCCN 20137
Nodes Reached Upper Bound vs. pdf Analysis July-August 2013ICCCN c=4
pdfs for Total Nodes Reached Varying c July-August 2013ICCCN l = 10
pdfs for Total Nodes Reached Varying l July-August 2013ICCCN c=4
pdfs for Total Nodes Reached Varying f July-August 2013ICCCN c = 4 l = 10
Related Work S. E. Deering and D. R. Cheriton, “Multicast routing in datagram internetworks and extended LANs,” ACM Trans. Computer Systems, vol. 8, no. 2, pp , 1990 Z. J. Hass, J. Y. Halpern and L. Li, “Gossip-based ad hoc routing,” IEEE/ACM Trans. Networking, vol. 14, no. 3, pp , June 2006 S. M. Hedetniemi, S. T. Hedetniemi and A. L. Liestman, “A survey of gossiping and broadcasting in communications networks,” Networks, vol. 18, pp , 1988 D. Shah, “Gossip algorithms,” Foundations and Trends in Networking, vol. 3, no. 1, pp , 2008 July-August 2013ICCCN
Conclusions The expected numbers of nodes reached up through a given level, and at a given level, are substantially less than those for the upper bound analysis The pdfs for the number of nodes reached up through a given level, and at a given level, exhibit a wide range, particularly for smaller values of the – Degree of forwarding – Probability of forwarding As the forwarding probability decreases, the expected number of nodes reached decreases quite rapidly July-August 2013ICCCN
Future Work Neighborhoods – fully connected within neighborhoods but only partially connected between neighborhoods Faulty nodes and links Time to reach a certain number of nodes Energy and power consumption at nodes Application to existing iTrust system – decentralized publication, search and retrieval July-August 2013ICCCN
Questions? Comments? Louise Moser - Michael Melliar-Smith - This work was supported in part by National Science Foundation Grant NSF CNS July-August 2013ICCCN