Tie::Cache - LRU Cache in Memory

 use Tie::Cache;
 tie %cache, 'Tie::Cache', 100, { Debug => 1 };   
 tie %cache2, 'Tie::Cache', { MaxCount => 100, MaxBytes => 50000 };
 tie %cache3, 'Tie::Cache', 100, { Debug => 1 , WriteSync => 0};

 # Options ##################################################################
 #
 # Debug =>      0 - DEFAULT, no debugging output
 #               1 - prints cache statistics upon destroying
 #               2 - prints detailed debugging info
 #
 # MaxCount =>   Maximum entries in cache.
 #
 # MaxBytes =>   Maximum bytes taken in memory for cache based on approximate 
 #               size of total cache structure in memory
 #
 #               There is approximately 240 bytes used per key/value pair in the cache for 
 #               the cache data structures, so a cache of 5000 entries would take
 #               at approximately 1.2M plus the size of the data being cached.
 #
 # MaxSize  =>   Maximum size of each cache entry. Larger entries are not cached.
 #                   This helps prevent much of the cache being flushed when 
 #                   you set an exceptionally large entry.  Defaults to MaxBytes/10
 #
 # WriteSync =>  1 - DEFAULT, write() when data is dirtied for 
 #                   TRUE CACHE (see below)
 #               0 - write() dirty data as late as possible, when leaving 
 #                   cache, or when cache is being DESTROY'd
 #
 ############################################################################

 # cache supports normal tied hash functions
 $cache{1} = 2;       # STORE
 print "$cache{1}\n"; # FETCH

 # FIRSTKEY, NEXTKEY
 while(($k, $v) = each %cache) { print "$k: $v\n"; }

 delete $cache{1};    # DELETE
 %cache = ();         # CLEAR

This module implements a least recently used (LRU) cache in memory through a tie interface. Any time data is stored in the tied hash, that key/value pair has an entry time associated with it, and as the cache fills up, those members of the cache that are the oldest are removed to make room for new entries.

So, the cache only remembers the last written entries, up to the size of the cache. This can be especially useful if you access great amounts of data, but only access a minority of the data a majority of the time.

The implementation is a hash, for quick lookups, overlaying a doubly linked list for quick insertion and deletion. On a WinNT PII 300, writes to the hash were done at a rate 3100 per second, and reads from the hash at 6300 per second. Work has been done to optimize refreshing cache entries that are frequently read from, code like CW$cache{entry}, which moves the entry to the end of the linked list internally.

Tie::Cache installs easily using the make or nmake commands as shown below. Otherwise, just copy Cache.pm to CW$PERLLIB/site/Tie

        > perl Makefile.PL
        > make
        > make test 
        > make install

        * use nmake for win32
        ** you can also just copy Cache.pm to $perllib/Tie

There is another simpler LRU cache implementation in CPAN, Tie::Cache::LRU, which has the same basic size limiting functionality, and for this functionality, the exact same interface.

Through healthy competition, Michael G Schwern got Tie::Cache::LRU mostly faster than Tie::Cache on reads & writes:

 Cache Size 5000       Tie::Cache 0.17  Tie::Cache::LRU 0.21
 10000 Writes             1.55 CPU sec          1.10 CPU sec
 40000 Reads              1.82 CPU sec          1.58 CPU sec
 10000 Deletes            0.55 CPU sec          0.59 CPU sec

Unless you are using TRUE CACHE or MaxBytes functionality, using Tie::Cache::LRU should be an easy replacement for Tie::Cache.

To use class as a true cache, which acts as the sole interface for some data set, subclass the real cache off Tie::Cache, with CW@ISA = qw( 'Tie::Cache' ) notation. Then override the read() method for behavior when there is a cache miss, and the write() method for behavior when the cache's data changes.

When WriteSync is 1 or TRUE (DEFAULT), write() is called immediately when data in the cache is modified. If set to 0, data that has been modified in the cache gets written out when the entries are deleted or during the DESTROY phase of the cache object, usually at the end of a script.

To have the dirty data write() periodically while WriteSync is set to 0, there is a flush() cache API call that will flush the dirty writes in this way. Just call the flush() API like:

  my $write_flush_count = tied(%cache)->flush();

The flush() API was added in the .17 release thanks to Rob Bloodgood.

 use Tie::Cache;

 # personalize the Tie::Cache object, by inheriting from it
 package My::Cache;
 @ISA = qw(Tie::Cache);

 # override the read() and write() member functions
 # these tell the cache what to do with a cache miss or flush
 sub read { 
    my($self, $key) = @_; 
    print "cache miss for $key, read() data\n";
    rand() * $key; 
 }
 sub write { 
    my($self, $key, $value) = @_;
    print "flushing [$key, $value] from cache, write() data\n";
 }

 my $cache_size   = $ARGV[0] || 2;
 my $num_to_cache = $ARGV[1] || 4;   
 my $Debug = $ARGV[2] || 1;

 tie %cache, 'My::Cache', $cache_size, {Debug => $Debug};

 # load the cache with new data, each through its contents,
 # and then reload in reverse order.
 for(1..$num_to_cache) { print "read data $_: $cache{$_}\n" }
 while(my($k, $v) = each %cache) { print "each data $k: $v\n"; }
 for(my $i=$num_to_cache; $i>0; $i--) { print "read data $i: $cache{$i}\n"; }

 # flush writes now, trivial use since will happen in DESTROY() anyway
 tied(%cache)->flush();

 # clear cache in 2 ways, write will flush out to disk
 %cache = ();
 undef %cache;

Many thanks to all those who helped me make this module a reality, including:

        :) Tom Hukins who provided me insight and motivation for
           finishing this module.
        :) Jamie McCarthy, for trying to make Tie::Cache be all
           that it can be.
        :) Rob Fugina who knows how to "TRULY CACHE".
        :) Rob Bloodgood, for the TRUE CACHE flush() API

Please send any questions or comments to Joshua Chamas at chamas@alumni.stanford.org

Copyright (c) 1999-2002 Joshua Chamas, Chamas Enterprises Inc. Sponsored by development on NodeWorks http://www.nodeworks.com

All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.