Counting With Tinytable: Every Bit Counts!

Abstract

Bloom filters are space efficient data structures that support approximate membership queries. They are easily extensible but incur significant overheads when extended to support additional functionality, such as removals or counting. This paper shows that fingerprint-based hash tables offer a much better tradeoff between accuracy and space. We present TinyTable that supports set membership, removals, and multiplicity queries. TinyTable reduces the required memory by as much as 28% compared to Bloom filter-based variants for the set membership and by as much as 60% for counting and statistics. It is more compact than Bloom filters as long as the false positive ratio is less than 1%.

Keywords

• Bloom filters
• approximation algorithms
• storage systems
• database
• networks
• datastructure