On Message Sequences for Incremental Network Computing

In a network computing platform, tasks compete with others for shared resources to communicate messages. Incremental computing masks communication latency by overlapping computation with communication. However, a sequence of messages with a large latency variance still makes computations proceed intermittently. In this paper, the impact of the message sequence on computation efficiency is studied and a framework which employs a well organized message sequence to maximize the efficiency of computations is introduced. Firstly, a network computing model for performing incremental computations is proposed. Based on the model, theorems are developed as the groundwork based on which algorithms for finding a well organized message sequence are derived. Finally, algorithms which find a well organized message sequence in 1 (( ) ) k r k O + and ( ) ! ( !) k r k O comparison steps are given for sending r input data items using r k mes-