N*-on-i'ntrusive N#*eighbor Pred'ict'ion in Sparse
Ovidiu Valentin Drugan · 2007
Toincrease theavailability ofmission critical services andinformation insparse MANETswithfrequent and/or longtermnetwork partitions, weaimtodevelop efficient replication andplacement algorithms. Thepredic- tionoftheneighborhood ofanodeisonecoreelement in these algorithms. Inthis paperwepresent aneighborhood prediction algorithm based ontheSequential MonteCarlo framework, i.e., recursive Bayesian filters usingasetof randomsamples, whichareupdated andpropagated bythe filter. Thealgorithm workswithout location information andextracts onlyinformation fromthelocal routing table topredict thefuture neighborhood ofthenode.We have performed extensive experiments toevaluate theaccuracy oftheprediction algorithm. Thepredicted connection and disconnection times follow closely thetrue distribution asregistered bytherouting protocol.