When the Data Integration Service runs jobs in separate remote processes, stability increases because an unexpected interruption to one job does not affect all other jobs. In addition, you can better use the resources available on each node in the grid. When a node has the compute role only, the node does not have to run the service process. The machine uses all available processing power to run mappings.
The following image shows two of many nodes in a Data Integration Service grid. Node1 has the service role, and Node2 has the compute role. The Data Integration Service process on Node1 manages application deployments, logging, job requests, and job optimizations. The Service Manager on Node2 runs DTM instances in separate DTM processes started within containers.