Is Knime optimized for multiple sockets/ NUMA systems?


When running a parallel chunk loop in Knime on a NUMA system is there a chance the sub processes will be scheduled on multiple NUMA nodes and use memory from different nodes at the same time, thus creating inefficiencies?

If yes, would adding -XXUseNuma to .ini or setting up affinity in windows be a potential solution?



In Java you have almost no control over memory management. Everything is transparently done by the JVM. Setting this option may help in this case.

This topic was automatically closed 182 days after the last reply. New replies are no longer allowed.