Closed Hashing is Computable and Optimally Randomizable with Universal Hash Functions
Abstract
We have shown that in double hashing, a universal family of hash functions, that with high probability provides (c)log(n)-wise independence, will give optimal performance for any fixed load bound below 1. The notion of double hashing is generalized to include almost any reasonable probe scheme. Similarly, linear probing incurs no loss of performance when such hash functions are used. These optimality results apply to the expected [r]-th moment of the probe count, for any fixed [r]. Our proof technique analyzed local and global hashing interactions separately, and used analytic tools to measure complicated but weakly correlated events in terms of simpler independent processes. Bundling and thinning techniques were used to eliminate (spurious) combinatorial explosions from more nave counting formulations. Surely these methods can be applied to other probabilistic process that exhibit weak correlation or that might only be supported by a source of limited randomness.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1995
- Accession Number
- AD1020223
Entities
People
- Alan Siegel
- Jeanette P. Schmidt
Organizations
- Courant Institute of Mathematical Sciences, NYU