Figure 3

Event sequence during cluster and cloud execution. First, ExperAgent uses the Partition Manager to calculate which of M samples go to which of N Slave nodes. It then invokes the first sample on the first Slave node and continues until all Slaves are occupied. When any one of the Slaves finishes (asynchronously), it writes the run data for that sample and then signals the ExperAgent, which shuts down that sample and spawns a new sample on that Slave. This continues until all M samples are finished, at which point the ExperAgent uses the Statistics Module to aggregate the data and calculate the derived observations. ExperAgent then writes that data to disk via the Data Management Module.