Graphical models have been recognized as a state-of-the-art methodology for machine learning and statistical inference, where among others, the Belief Propagation (BP) algorithm is a powerful algorithm for solving certain large and otherwise intractable inference problems. Recently, another algorithm, the Survey Propagation (SP) algorithm, has been developed in graphical models for solving the classical NP-complete constraint-satisfaction problems, the k-SAT problems. As a celebrated success, SP is shown to be very efficient in solving random k-SAT problems in the hard regime. The algorithmic similarity between SP and BP insipres researchers to understand the connection between the two algorithms, where an earlier work from a Berkeley group (M. Wainright) shows that SP may be understood as an instance of BP for k-SAT problems. In this talk, we show that for general constraint satisfaction problems, SP may not be reducible from BP. We also establish the conditions under which such a reduction is possible. Along our development, we present a unification of the existing SP algorithms in terms of a probabilistically interpretable iterative procedure -- Weighted Probabilistic Token Passing.