Review all serializable classes used in the bigdata core and in the bigdata-rdf module for forward versioning mechanisms when they are persisted against the database.
Write unit tests designed to exercise all persistable data structures by writing them onto a store and then reading them back. We should have a main() routine for that class which saves a copy of the journal. At each release, we can save a version of the journal with the data into SVN for backward compatibility read tests.
Since we know that we would like to have compatibility in the future with a bigdata C/C++ library for GPU processing , I would also suggest that we address this compatibility issue at the same time and avoid any possible future pain in this area since C compatibility might well break some of the existing serialization mechanisms.