Storing the AtomSpace in HyperTable

From OpenCog

Draft schema design for using a BigTable equivalent (HyperTable).

Part of the GSoC 2009: Distributed and Persistent AtomSpace project.


Current plans are to do development in hypertable, as opposed to Hadoop, mostly because Hadoop's Java implementation makes it harder to integrate with OpenCog.

Design considerations

Tables only have one primary index which is the key column.

This column also dictates how tablets (distributed chunks of the complete table) are stored together...

In order to support look-ups by type, outgoing set, name, etc. we'll need separate table for each index.

I found an email on the HyperTable users list that addresses this. This also refers the original poster on the list to the Star schema of database design.