/ 
FusqlFS-0.009
River stage zero No dependents
/ FusqlFS::Cache::Limited

NAME

FusqlFS::Cache::Limited - FusqlFS limited cache strategy implementation

SYNOPSIS

use FusqlFS::Cache::Limited;

our %cache;
tie %cache, 'FusqlFS::Cache::Limited', 10000;

DESCRIPTION

This is a limited by items number cache strategy implementation. This class is not to be used directly.

This cache strategy accepts single `threshold' parameter which must be integer greater than zero and determines maximum number of items to be stored in cache.

This cache strategy tries to keep memory usage low enough by limiting number of cache items. If on cache write number of items in cache exceeds given threshold, cache cleanup is forced, striving to decrease cached items number down to threshold plus 1/8 of total current number of items. So number of cache items varies between (7/8)*threshold and threshold in fact. It does cleanup this way to avoid cleanup situations appearance too often.

The cleanup process chooses items to remove quite cleverly: this class keeps items usage statistics and cleanup process uses this statistics to remove least used items first. It makes cleanup and cache hit try processes a little slower, but speeds up cache performance in whole, as resulting hit/miss ratio is better, than in case of naive cleanup algorithm.

This cache strategy is good if you are going to mount database with a lot number of records and you have memory issues because of this, as it focuses on keeping overall number of cached items low enough. If you need to mount database with moderate number of large records, like records with huge blob or text fields, you might consider using FusqlFS::Cache::File strategy.

I also recommend setting threshold for this cache strategy to at least 3/4 of total objects in your database (including all tables, sequences, views, data rows etc.), which will bring you about 60% cache hits (~45% for 1/2 and ~56% for 2/3). But this is just a basic recommendation based on educated guess and some tests with "entry" names generated with normally distributed random generator. Experiment is your best advisor in this case.

SEE ALSO