The Gaussian Bloom Filter

摘要: Modern databases tailored to highly distributed, fault tolerant management of information for big data applications exploit a classical structure reducing disk and network I/O as well managing distribution: The Bloom filter. This allows encode small sets elements, typically the keys in key-value store, into small, constant-size structure. In order reduce memory consumption, this suffers from false positives which lead additional operations are therefore only harmful with respect performance. With paper, we propose an extension filter construction facilitates use floating point coprocessors GPUs or main positives. proposed is compatible sense that can be extracted time linear size special case our construction. We show approach provides relevant gain positive rate. Implementations Apache Cassandra, C++, NVIDIA CUDA given support feasibility results approach.