2016/03/15 by Hugo Firth, Firth, Hugo, Paolo Missier +1
Computer Science · #Cloud Computing and Resource Management #Databases (cs.DB) #FOS: Computer and information sciences #Graph Theory and Algorithms #H.2.4 #Interconnection Networks and Systems #Parallel Computing and Optimization Techniques
paper · pdf · doi:10.48550/arxiv.1603.04626
openalex publication_date 2016/03/15 · openalex created_date 2022/08/23 · openalex updated_date 2026/07/28
Graph partitioning has long been seen as a viable approach to address Graph\nDBMS scalability. A partitioning, however, may introduce extra query processing\nlatency unless it is sensitive to a specific query workload, and optimised to\nminimise inter-partition traversals for that workload. Additionally, it should\nalso be possible to incrementally adjust the partitioning in reaction to\nchanges in the graph topology, the query workload, or both. Because of their\ncomplexity, current partitioning algorithms fall short of one or both of these\nrequirements, as they are designed for offline use and as one-off operations.\nThe TAPER system aims to address both requirements, whilst leveraging existing\npartitioning algorithms. TAPER takes any given initial partitioning as a\nstarting point, and iteratively adjusts it by swapping chosen vertices across\npartitions, heuristically reducing the probability of inter-partition\ntraversals for a given pattern matching queries workload. Iterations are\ninexpensive thanks to time and space optimisations in the underlying support\ndata structures. We evaluate TAPER on two different large test graphs and over\nrealistic query workloads. Our results indicate that, given a hash-based\npartitioning, TAPER reduces the number of inter-partition traversals by around\n80%; given an unweighted METIS partitioning, by around 30%. These reductions\nare achieved within 8 iterations and with the additional advantage of being\nworkload-aware and usable online.\n