Testing some open source bloom filter implementations

SCALA: https://github.com/alexandrnikitin/bloom-filter-scala

GUAVA: https://guava.dev/releases/23.0/api/docs/com/google/common/hash/BloomFilter.html

ORESTES: https://github.com/Baqend/Orestes-Bloomfilter

LOVA: https://github.com/lovasoa/bloomfilter

RESULTS:

SCALA      NumAllocations: 96;         FalsePositiveRate: 0.04839999973773956; QPS: 1244379.4491231728
GUAVA      NumAllocations: 1920000136; FalsePositiveRate: 0.04899999871850014; QPS: 817545.5744868225
ORESTES    NumAllocations: 800000080;  FalsePositiveRate: 0.04569999873638153; QPS: 831717.514132752
LOVA       NumAllocations: 112;        FalsePositiveRate: 0.04839999973773956; QPS: 1218425.5675395832

All 4 implementations are correct in that none of them exceed the assigned false positive rate of 0.05

But the GUAVA and ORESTES implementations are doing lots of allocations internally (creating new objects)

The SCALA and LOA versions use less allocations and have higher QPS than the others (likely due to less allocations)

Dev

To run this test, first do a clean build:

 mvn clean install

Then run with below VM params:

-javaagent:/Users/rbanikaz/.m2/repository/com/google/code/java-allocation-instrumenter/java-allocation-instrumenter/3.1.0/java-allocation-instrumenter-3.1.0.jar

Output gets printed to console