Implement and evaluate an optional tuple level cache in front of the B+Tree to support faster point tests for frequently accessed tuples. For example, this could be used to buffer resolution of term identifiers to RDF Values on the ID2TERM index.
There is already a termCache in the LexiconRelation which plays this role. The questions are (a) whether the term cache can be usefully extracted from the LexiconRelation and made a part of the B+Tree interface itself (which might require indexing by a byte key rather than a Long) and (b) whether distributed hash tables (DHTs) might be approach all around for such data. The drawbacks of DHTs are: (1) they are more difficult to rebalance when the size of the cluster changes; (2) this would be an entirely new persistent data structure for bigdata; and (3) cache fabrics already specialize in DHTs.