Speedups in Apache Lucene: A 2019 Perspective

Search
06/17/2019 - 12:20 to 13:00
Frannz Salon
long talk (40 min)
Intermediate

Session abstract: 

Apache Lucene is the basis of an overwhelming number of search applications, either used directly or through the use of Apache Solr or Elasticsearch. Up until recently, performance improvements in Apache Lucene over the years were incremental and fairly modest. However, recently, a lot of speedup has been realized since Lucene 7.0 and 8.0.

This talk is about some of these major performance improvements. Notable among them is the implementation of Block-Max WAND, which has introduced significant throughput and latency efficiency [0]. Also, this talk will cover some exciting ongoing work on GPU based acceleration of parts of Lucene for scoring and faceting [1].

Key takeaway for the audience would be the awareness of the scenarios where such improvements can benefit their search applications and the constraints under which they are effective.

[0] - LUCENE-8135 [1] - LUCENE-7745