- Native (MRI/C) counting bloom filter
- Redis-backed getbit/setbit non-counting bloom filter
- Redis-backed set-based counting (+TTL) bloom filter
Bloom filter is a space-efficient probabilistic data structure that is used to test whether an element is a member of a set. False positives are possible, but false negatives are not. For more detail, check the wikipedia article. Instead of using k different hash functions, this implementation seeds the CRC32 hash with k different initial values (0, 1, ..., k-1). This may or may not give you a good distribution, it all depends on the data.
Performance of the Bloom filter depends on a number of variables:
- size of the bit array
- size of the counter bucket
- number of hash functions
- Determining parameters: Scalable Datasets: Bloom Filters in Ruby
- Applications & reasons behind bloom filter: Flow analysis: Time based bloom filter
MRI/C implementation which creates an in-memory filter which can be saved and reloaded from disk.
require 'bloomfilter-rb'
bf = BloomFilter::Native.new(:size => 100, :hashes => 2, :seed => 1, :bucket => 3, :raise => false)
bf.insert("test")
bf.include?("test") # => true
bf.include?("blah") # => false
bf.delete("test")
bf.include?("test") # => false
# Hash with a bloom filter!
bf["test2"] = "bar"
bf["test2"] # => true
bf["test3"] # => false
bf.stats
# => Number of filter bits (m): 10
# => Number of filter elements (n): 2
# => Number of filter hashes (k) : 2
# => Predicted false positive rate = 10.87%
Uses getbit/setbit on Redis strings - efficient, fast, can be shared by multiple/concurrent processes.
bf = BloomFilter::Redis.new
bf.insert('test')
bf.include?('test') # => true
bf.include?('blah') # => false
bf.delete('test')
bf.include?('test') # => false
- 1.0% error rate for 1M items, 10 bits/item: 2.5 mb
- 1.0% error rate for 150M items, 10 bits per item: 358.52 mb
- 0.1% error rate for 150M items, 15 bits per item: 537.33 mb
Uses regular Redis get/set counters to implement a counting filter with optional TTL expiry. Because each "bit" requires its own key in Redis, you do incur a much larger memory overhead.
bf = BloomFilter::CountingRedis.new(:ttl => 2)
bf.insert('test')
bf.include?('test') # => true
sleep(2)
bf.include?('test') # => false
Tatsuya Mori valdzone@gmail.com (Original C implementation: http://vald.x0.com/sb/)
MIT License - Copyright (c) 2011 Ilya Grigorik