Package org.apache.commons.collections4.bloomfilter

package org.apache.commons.collections4.bloomfilter

Collects extensible Bloom filter classes and interfaces.

Background:

The Bloom filter is a probabilistic data structure that indicates where things are not. Conceptually it is a bit vector or BitMap. You create a Bloom filter by creating hashes and converting those to enabled bits in the map. Multiple Bloom filters may be merged together into one Bloom filter. It is possible to test if a filter B has merged into another filter A by verifying that (A & B) == B.

Bloom filters are generally used where hash tables would be too large, or as a filter front end for longer processes. For example most browsers have a Bloom filter that is built from all known bad URLs (ones that serve up malware). When you enter a URL the browser builds a Bloom filter and checks to see if it is "in" the bad URL filter. If not the URL is good, if it matches, then the expensive lookup on a remote system is made to see if it actually is in the list. There are lots of other uses, and in most cases the reason is to perform a fast check as a gateway for a longer operation.

Some Bloom filters (e.g. CountingBloomFilter) use counters rather than bits. In this case each counter is called a cell.

BloomFilter

The Bloom filter architecture here is designed for speed of execution, so some methods like merge, remove, add, and subtract may throw exceptions. Once an exception is thrown the state of the Bloom filter is unknown. The choice to use not use atomic transactions was made to achieve maximum performance under correct usage.

Nomenclature

• BitMap - In the bloomfilter package a BitMap is not a structure but a logical construct. It is conceptualized as an ordered collection of long values each of which is interpreted as the enabled true/false state of 64 continuous indices. The mapping of bits into the long values is described in the BitMaps javadoc.
• Index - In the bloomfilter package an Index is a logical collection of ints specifying the enabled bits in the BitMap.
• Cell - Some Bloom filters (e.g. CountingBloomFilter) use counters rather than bits. In the bloomfilter package Cells are pairs of ints representing an index and a value. They are not Pair objects.
• Extractor - The Extractors are FunctionalInterfaces that are conceptually iterators on a BitMap, an Index, or a collection of Cells, with an early termination switch. Extractors have names like BitMapExtractor or IndexExtractor and have a processXs methods that take a Predicate<X> argument (e.g. processBitMaps(LongPredicate) or processIndicies(IntPredicate)). That predicate is expected to process each of the Xs in turn and return true if the processing should continue or false to stop it.

There is an obvious association between the BitMap and the Index, as defined above, in that if bit 5 is enabled in the BitMap than the Index must contain the value 5.

Implementation Notes

The architecture is designed so that the implementation of the storage of bits is abstracted. Rather than specifying a specific state representation we require that all Bloom filters implement the BitMapExtractor and IndexExtractor interfaces, Counting-based Bloom filters implement CellExtractor as well. There are static methods in the various Extractor interfaces to convert from one type to another.

Programs that utilize the Bloom filters may use the BitMapExtractor or IndexExtractor to retrieve or process a representation of the internal structure. Additional methods are available in the BitMaps class to assist in manipulation of BitMap representations.

The Bloom filter is an interface that requires implementation of 9 methods:

• BloomFilter.cardinality() returns the number of bits enabled in the Bloom filter.
• BloomFilter.characteristics() which returns an integer of characteristics flags.
• BloomFilter.clear() which resets the Bloomfilter to its initial empty state.
• BloomFilter.contains(IndexExtractor) which returns true if the bits specified by the indices generated by IndexExtractor are enabled in the Bloom filter.
• BloomFilter.copy() which returns a fresh copy of the bitmap.
• BloomFilter.getShape() which returns the shape the Bloom filter was created with.
• BloomFilter.merge(BitMapExtractor) which merges the BitMaps from the BitMapExtractor into the internal representation of the Bloom filter.
• BloomFilter.merge(IndexExtractor) which merges the indices from the IndexExtractor into the internal representation of the Bloom filter.

Other methods should be implemented where they can be done so more efficiently than the default implementations.

CountingBloomFilter

The counting Bloom filter extends the Bloom filter by counting the number of times a specific bit has been enabled or disabled. This allows the removal (opposite of merge) of Bloom filters at the expense of additional overhead.

LayeredBloomFilter

The layered Bloom filter extends the Bloom filter by creating layers of Bloom filters that can be queried as a single Filter or as a set of filters. This adds the ability to perform windowing on streams of data.

Shape

The Shape describes the Bloom filter using the number of bits and the number of hash functions. It can be specified by the number of expected items and desired false positive rate.

Hasher

A Hasher converts bytes into a series of integers based on a Shape. Each hasher represents one item being added to the Bloom filter.

The EnhancedDoubleHasher uses a combinatorial generation technique to create the integers. It is easily initialized by using a byte array returned by the standard MessageDigest or other hash function to initialize the Hasher. Alternatively a pair of a long values may also be used.

Other implementations of the Hasher are easy to implement, and should make use of the Hasher.Filter and/or Hasher.FileredIntConsumer classes to filter out duplicate indices when implementing Hasher.uniqueIndices(Shape).

References

1. https://www.eecs.harvard.edu/~michaelm/postscripts/tr-02-05.pdf
2. https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/utils/BloomFilter.java#L60

Since:

4.5

Related Packages

Package	Description
org.apache.commons.collections4	Interfaces and utilities shared across all packages.

Class	Description
ArrayCountingBloomFilter	A counting Bloom filter using an int array to track cells for each enabled bit.
BitMapExtractor	Produces bit map longs for a Bloom filter.
BitMaps	Contains functions to convert int indices into Bloom filter bit positions and visa versa.
BloomFilter	The interface that describes a Bloom filter.
BloomFilterExtractor	Produces Bloom filters from a collection (e.g.
CellExtractor	Some Bloom filter implementations use a count rather than a bit flag.
CellExtractor.CellPredicate	Represents an operation that accepts an <index, count> pair.
CountingBloomFilter	The interface that describes a Bloom filter that associates a count with each bit index rather than a bit.
EnhancedDoubleHasher	A Hasher that implements combinatorial hashing as described by Krisch and Mitzenmacher using the enhanced double hashing technique described in the wikipedia article Double Hashing.
Hasher	A Hasher creates IndexExtractor based on the hash implementation and the provided Shape.
IndexExtractor	An object that produces indices of a Bloom filter.
IndexFilter	A convenience class for Hasher implementations to filter out duplicate indices.
LayeredBloomFilter<T extends BloomFilter>	Layered Bloom filters are described in Zhiwang, Cen; Jungang, Xu; Jian, Sun (2010), "A multi-layer Bloom filter for duplicated URL detection", Proc.
LayerManager<T extends BloomFilter>	Implementation of the methods to manage the layers in a layered Bloom filter.
LayerManager.Builder<T extends BloomFilter>	Builder to create Layer Manager
LayerManager.Cleanup	Static methods to create a Consumer of a List of BloomFilter perform tests on whether to reduce the collection of Bloom filters.
LayerManager.ExtendCheck	A collection of common ExtendCheck implementations to test whether to extend the depth of a LayerManager.
LongBiPredicate	Represents a function that accepts a two long-valued argument and produces a binary result.
SetOperations	Implementations of set operations on BitMapExtractors.
Shape	The definition of a Bloom filter shape.
SimpleBloomFilter	A bloom filter using an array of bit maps to track enabled bits.
SparseBloomFilter	A bloom filter using a TreeSet of integers to track enabled bits.
WrappedBloomFilter	An abstract class to assist in implementing Bloom filter decorators.