BloomFilter(s) in Ruby

• 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

Resources

• Determining parameters: Scalable Datasets: Bloom Filters in Ruby
• Applications & reasons behind bloom filter: Flow analysis: Time based bloom filter

MRI/C API Example

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%

Redis-backed setbit/getbit bloom filter

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

Memory footprint

• 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

Redis-backed counting bloom filter with TTLs

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

Credits

Tatsuya Mori [email protected] (Original C implementation: http://vald.x0.com/sb/)

License

MIT License - Copyright (c) 2011 Ilya Grigorik