to load the massive stress data, it got 2:40min, and to process the bundles, it got 14min (including the load time) using Xmx/Xms of 500M. When using more than one thread on bundle processing, it will became faster.

to best create this performance test i think its ideal create a simple java application that reads the same amout of artifacts that we have to load, read one by one and store its content in memory. it this application we have a clear notion about pure JVM handling to benchmark with our federation engine.