Hit Rate
Below we compare LRU, ConcurrentLru (TU-Q) and ConcurrentLfu (W-TinyLfu) hit rate using several well-known traces from the literature.
See:
- The LRU-K Page Replacement Algorithm For Database Disk Buffering
- ARC: A Self-Tuning, Low Overhead Replacement Cache.
- LIRS: An Efficient Low Inter-reference Recency Set Replacement Policy to Improve Buffer Cache Performance
This replicates the test suite used in Java's popular Caffeine library.
The result of replaying 45 million web requests from the September 2007 Wikibench dataset is shown below. The trace files used are 1190153705, 1190157306, 1190160907, 1190164508, 1190168109.
This trace is provided by the authors of the LIRS algorithm and exhibits a looping access pattern.
The ARC database trace is available from Dharmendra S. Modha's IBM page and is mirrored here: DS1.lis.gz.
Results for the OLTP trace are show below. The OLTP trace was described as "a database server running at a commercial site running an ERP application on top of a commercial database."
The ARC search traces are available from Dharmendra S. Modha's IBM page and are mirrored here: S1.lis.gz, S2.lis.gz, S3.lis.gz.
Results for the S3 trace are shown below, the search traces are described as "disk read accesses initiated by a large commercial search engine in response to various web search requests."
The ARC OLTP trace is available from Dharmendra S. Modha's IBM page and is mirrored here: OLTP.lis.gz.
Results for the OLTP trace are show below. The OLTP trace was described as "a one hour page reference trace to a CODASYL database."
