Parallel personalized pagerank on dynamic graphs

Abstract

Personalized PageRank (PPR) is a well-known proximity measure in graphs. To meet the need for dynamic PPR maintenance, recent works have proposed a local update scheme to support incremental computation. Nevertheless, sequential execution of the scheme is still too slow for highspeed stream processing. Therefore, we are motivated to design a parallel approach for dynamic PPR computation. First, as updates always come in batches, we devise a batch processing method to reduce synchronization cost among every single update and enable more parallelism for iterative parallel execution. Our theoretical analysis shows that the parallel approach has the same asymptotic complexity as the sequential approach. Second, we devise novel optimization techniques to effectively reduce runtime overheads for parallel processes. Experimental evaluation shows that our parallel algorithm can achieve orders of magnitude speedups on GPUs and multi-core CPUs compared with the state-of-the-art sequential algorithm.

Efficient Representative Subset Selection over Sliding Windows

“…In many cases, data is generated rapidly and only available as a stream [11,18,19,30,39]. To address the requirement for summarizing such datasets in real-time, RSS over data streams [2,10,29,32] has been extensively studied in recent years.…”

Parallel processing of approximate single-source personalized PageRank queries using shared-memory

“…To the best of our knowledge, most research works of PPR computation on shared-memory, e.g, [7,13,[16][17][18], focus on sequential algorithms and do not consider the computing capacity of multicore systems. There are few research works [19,20] on parallel PPR computation. Guo et al [19] propose a parallel solution for PPR computing on dynamic graphs, focusing on updating PPRs when new edges are added into the graph.…”

“…There are few research works [19,20] on parallel PPR computation. Guo et al [19] propose a parallel solution for PPR computing on dynamic graphs, focusing on updating PPRs when new edges are added into the graph. However, their solution assumes that all the forward push [7] results are available, and is impractical to support approximate SSPPR query answering.…”

“…However, the solutions in this category typically incur thousands of seconds to handle large graphs, e.g., on Twitter and are too slow to support real-world applications. There is another line of local update based approach [7,16,19,41] which is also based on the matrix-based definition. However, such solutions either provide no approximation guarantee to the SSPPR queries or cannot be directly applied to answer SSPPR queries.…”

“…The first is the state of the art parallel solution for PPR as we mentioned in Section 2.7.1. The solution proposed by Guo et al [19], however, provides no approximation guarantee to the SSPPR query, and the space consumption to pre-store all the forward push result is prohibitive for large graphs, e.g., Twitter.…”