The Data Grid Spark connector creates by default two partitions per each Data Grid node, each partition specifies a subset of the data in that particular node.
Those partitions are then sent to the Spark workers that will process them in parallel. If the number of Spark workers is less than the number of Data Grid nodes, some delay can occur since each worker has a maximum capacity of executing tasks in parallel. For this reason it is recommended to have at least the same number of Spark workers as Data Grid nodes to take advantage of the parallelism.
In addition, if a Spark worker is co-located in the same node as the Data Grid node, the connector will distribute tasks so that each worker only processes data found in the local Data Grid node.