A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs
SIAM Journal on Scientific Computing
Web++: a system for fast and reliable web service
ATEC '99 Proceedings of the annual conference on USENIX Annual Technical Conference
Dynamo: amazon's highly available key-value store
Proceedings of twenty-first ACM SIGOPS symposium on Operating systems principles
Dex: high-performance exploration on large graphs for information retrieval
Proceedings of the sixteenth ACM conference on Conference on information and knowledge management
Survey of graph database models
ACM Computing Surveys (CSUR)
Fast distributed graph partition and application
IPDPS'06 Proceedings of the 20th international conference on Parallel and distributed processing
Schism: a workload-driven approach to database replication and partitioning
Proceedings of the VLDB Endowment
Computer Science Review
Min-cut replication in partitioned networks
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
| Hi-index | 0.00 |
Executing walks in partitioned, distributed graphs with minimal latency requires reducing the number of network hops taken. This is especially important for graph databases that specialize on executing fast graph traversals. We present fast-forward-search, an algorithm that uses overlapping graph partitionings, i.e. replication, and parallel speculative execution to minimize the number of required network hops. We proof optimality of the algorithm, analyze storage, message, and computational overhead caused by the parallelism of fast-forward, and introduce escapicity, a metric for replica selection that helps reducing that parallelism at the price of lost optimality. Experiments for a set of smaller graphs indicate that fast-forward-search saves between 20-90% of network hops depending on graph and replication factor and that escapicity outperforms classic measures of network centrality as a metric for replica selection in our scheme.