LoGPC: Modeling network contention in message-passing programs

In many real applications, for example, those with frequent and irregular c ommunication patterns or those using large messages, network contention and contention for message processing resources can be a significant part of t he total execution time. This paper presents a new cost model, called LoGPC , that extends the LogP [9] and LogGP [4] models to account for the impact of network contention and network interface DMA behavior on the performance of message passing programs. We validate LoGPC by analyzing three applicat ions implemented with Active Messages [11], [19] on the MIT Alewife multipr ocessor. Our analysis shows that network contention accounts for up to 50 p ercent of the total execution time. In addition, we show that the impact of communication locality on the communication costs is at most a factor of t wo on Alewife. Finally. we use the model to identify trade-offs between syn chronous and asynchronous message passing styles.