Gil Einziger

Bloom filters are a very popular and efficient data structure for approximate set membership queries. However, Bloom filters have several key limitations as they require 44% more space than the lower bound, their operations access multiple memory words and they do not support removals. This work presents TinySet, an alternative Bloom filter construction that is more space efficient than Bloom filters for false positive rates smaller than 2.8%, accesses only a single memory word and partially supports removals. TinySet is mathematically analyzed and extensively tested and is shown to be fast and more space efficient than a variety of Bloom filter variants. TinySet also has low sensitivity to configuration parameters and is therefore more flexible than a Bloom filter.