[HADOOP-11829] Improve the vector size of Bloom Filter from int to long, and storage from memory to disk

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Invalid
• Component/s: util
• Assignee: Hongbo Xu
• Reporter: Hongbo Xu
• Created: 13/Apr/15 11:40
• Updated: 26/May/17 01:28
• Resolved: 22/Apr/15 01:30

Description

org.apache.hadoop.util.bloom.BloomFilter(int vectorSize, int nbHash, int hashType)
This filter almost can insert 900 million objects, when False Positives Probability is 0.0001, and it needs 2.1G RAM.
In My project, I needs established a filter which capacity is 2 billion, and it needs 4.7G RAM, the vector size is 38340233509, out the range of int, and I does not have so much RAM to do this, so I rebuild a big bloom filter which vector size type is long, and split the bit data to some files on disk, then distribute files to work node, and the performance is very good.
I think I can contribute this code to Hadoop Common, and a 128-bit Hash function (MurmurHash)