An experimentation to add ORM on top of Tokyo Cabinet What Is Tokyo Cabinet? - Modern implementation of DBM(key/value hash style(HDB), but supports fixed length hash(FDB), and b tree(BDB)) - High concurrency/ high scalability (developed by a developer at Mixi, Japanese equivalent of Facebook) - More detail at http://tokyocabinet.sourceforge.net/index.html How dm-tokyo-cabinet-adapter stores data into Tokyo Cabinet. - Each object is stored as "ObjName.bdb" file: object id as key and entire data marshalled as value - Each attribute is stored as "ObjNameAttribute.bdb" file: attribute name as key and reference object id as value - The above architecture can also be considerd "ObjeName.bdb" as table and "ObjNameAttribute.bdb" as indexes for each attributes. - Currently implements basic CRUD, association, and eql finder. Motivation behind the development. - To experiment what you can do with basic hash based database. Interesting post at http://groups.google.com/group/merb/browse_thread/thread/a8c6b154576c6270 - To learn internal of DataMapper and how to implement ORM/adapter How to install 1. Install Tokyo Cabinet http://tokyocabinet.sourceforge.net 2. Install Ruby Binding http://tokyocabinet.sourceforge.net/rubydoc/ 3. Install dm-tokyo-cabinet-adapter 3.1 download dm-tokyo-cabinet-adapter 3.2 cd to the dir 3.3 gem build dm-tokyo-cabinet-adapter.gemspec 3.4 gem install dm-tokyo-cabinet-adapter-0.0.2.gem 4. Create data dir 5. Setup database.yml like below :development: :adapter: tokyo_cabinet :data_path: <%= Pathname(__FILE__).dirname.expand_path + 'data' %> 6. The rest is usual way to setup datamapper on Merb. Benchmarking results http://gist.github.com/25946 My current implementation is a lot slower than MySQL and sqlite, but changing some data storage strategies speed up the performance dramatically, so there are lots of potential for optimization. Further research topics/TODO - Implement outstanding tasks such as multi conditions, other finder conditions (<, <=, like), Data Types - At this moment, it's only uses B tree, as it covers the wide range of functionality (range search, duplicate values, transaction support, and so on). Consider replacing part to HDB or FDB for compact storage and speed. - Are there any ways to retrieve first/last key on FDB/HDB? - Can TC support "not equal" operation?