Description Miroslav Kandic 2012-11-01 18:18:48 EDT

One of the major reasons we partition and replicate data is to improve scalability (and availability) applying parallel processing against partitions/replicas (kind of map-reduce).
EclipseLink should run in parallel at least the queries in case of "replication-partitioning" and "union-partitioning" policies. Ideally, it should do also manipulative operations in parallel.

With the current implementation a query against partitioned data will be slower than the same query when all data are in the same DB instance because of latency, context switching, etc.
If EclipseLink runs the same query in parallel, in general, it should be faster than traditional RDB which has all data in one DB instance.