"A Bloom filter is a space-efficient probabilistic data structure that is used to test whether an element is a member of a set. False positive matches are possible, but false negatives are not. In other words, a query returns either "possibly in set" or "definitely not in set". Elements can be added to the set, but not removed," says Wikipedia.
What's Bloom filter in a nutshell:
- Optimization for memory. It comes into play when you cannot put whole set into memory.
- Solves the membership problem. It can answer one question: does an element belong to a set or not?
- Probabilistic (lossy) data structure. It can answer that an element probably belongs to a set with some probability.
libraryDependencies += "com.github.alexandrnikitin" %% "bloom-filter" % "latest.release"// Create a Bloom filter
val expectedElements = 1000000
val falsePositiveRate = 0.1
val bf = BloomFilter[String](expectedElements, falsePositiveRate)
// Put an element
bf.add(element)
// Check whether an element in a set
bf.mightContain(element)
// Dispose the instance
bf.dispose()You can read about this Bloom filter and motivation behind in my blog post
Here's a benchmark for the String type and results for other types are very similar to these:
[info] Benchmark (length) Mode Cnt Score Error Units
[info] alternatives.algebird.StringItemBenchmark.algebirdGet 1024 thrpt 20 1181080.172 â–’ 9867.840 ops/s
[info] alternatives.algebird.StringItemBenchmark.algebirdPut 1024 thrpt 20 157158.453 â–’ 844.623 ops/s
[info] alternatives.breeze.StringItemBenchmark.breezeGet 1024 thrpt 20 5113222.168 â–’ 47005.466 ops/s
[info] alternatives.breeze.StringItemBenchmark.breezePut 1024 thrpt 20 4482377.337 â–’ 19971.209 ops/s
[info] alternatives.guava.StringItemBenchmark.guavaGet 1024 thrpt 20 5712237.339 â–’ 115453.495 ops/s
[info] alternatives.guava.StringItemBenchmark.guavaPut 1024 thrpt 20 5621712.282 â–’ 307133.297 ops/s
[info] bloomfilter.mutable.StringItemBenchmark.myGet 1024 thrpt 20 11483828.730 â–’ 342980.166 ops/s
[info] bloomfilter.mutable.StringItemBenchmark.myPut 1024 thrpt 20 11634399.272 â–’ 45645.105 ops/s
[info] bloomfilter.mutable._128bit.StringItemBenchmark.myGet 1024 thrpt 20 11119086.965 â–’ 43696.519 ops/s
[info] bloomfilter.mutable._128bit.StringItemBenchmark.myPut 1024 thrpt 20 11303765.075 â–’ 52581.059 ops/s
Basically, this implementation is 2x faster than Google's Guava and 10-80x than Twitter's Algebird. Other benchmarks you can find in the "benchmarks' module on github
Warning: These are synthetic benchmarks in isolated environment. Usually the difference in throughput and latency is bigger in production system because it will stress the GC, lead to slow allocation paths and higher latencies, trigger the GC, etc.