A Robust Bloom Filter

A Bloom filter is a space-efficient randomized data structure representing a set for membership queries. Faults in Bloom filters, however, cannot guarantee no false negatives. In this paper, we present a simple redundancy scheme for detecting false negatives and tolerating false positives induced by faults in Bloom filters during normal operation. A spare hashing unit with a simple coding technique is employed in the robust Bloom filter to tolerate faults without significant hardware overhead. Re-computing with shifted operands is used to locate the faulty unit upon detecting a fault. Both permanent and transient faults are considered in developing the fault diagnosis scheme. In the event of a permanent fault, the proposed Bloom filter is to operate in a degraded mode with some increase in false positive probability.