When comparing comparison between the Parallel Computing Toolbox (PCT) and MATLAB Distributed Computing Server (MDCS), please keep in mind the following:
- Contention for resources: network and computing. If the network of worker machines has low bandwidth or high latency, computation could be slowed.
- Hardware and/or platform difference may be a large contributor to the performance difference.
- Measure only the time spent executing the computating routing will show the performance discrepancy due to hardware.
- A worker by default runs single-threaded, whether the worker is hosted locally or on a cluster, even if the MATLAB functions used support multi-threading. Therefore it may not be always fair to compare the performance of a FOR loop with a PARFOR if the computating routine that uses largely functions that supports multithreading. This is because the overhead involved with using a new process (worker) is much larger than a new thread.
- Passing parameters. Parameters are automatically passed to worker machines during the execution of parallel computations. If there are a large number of parameters, or they take a large amount of memory, passing them may slow the execution of your computation.
You may also be interested in the following resources on profiling and optimizing your parallel computing performance:
