Analysis Efficieny of Graph Computation Based on Partitioning Techniques

Graph computation system performance is generally affected by the quality of graph partition. The balance factor and small edge cut ratio is used to measure the quality of graph partition. A balance graph partition with small edge cut ratio is generally preferred since it cut down the expensive network communication cost. Even so, according to the study of the Giraph, the execution over well partitioned graph strength is even two times worse than simple random partitions. This system only optimize for simple partition strategies and cannot efficiently handle the increasing workload of local message processing when high quality graph partition with the encryption process is used. In this paper, we provide a new processor and dynamic concurrency control models. The new message processor at the same time processes local and remote message in a unified way, the dynamic model adaptively adjust the concurrency of the processor based on the online statics. Keywords-Graph computation, graph partition, message processing.