Pseudorandom graphs in data structures

Abstract

We prove that the hash functions required for several data structure applications could be instantiated using the hash functions of Celis et al. (SIAM J. Comput., 2013). These functions simultaneously enjoy short description length as well as fast evaluation time. The applications we consider are: (1) Cuckoo Hashing, (2) Cuckoo Hashing with Stash and (3) the Power of Two Choices paradigm for load balancing. Our analysis relies on a notion of sparse pseudorandom graphs that are similar to random graphs in having no large connected component and no dense subgraph. Such graphs may be of independent interest. Relating pseudorandom graphs to the two-choice paradigm relies on a very simple new proof we give (at the price of somewhat worse parameters). © 2014 Springer-Verlag.