Towards emulation of large scale complex network workloads on graph databases with XGDBench

Abstract

Graph database systems are getting a lot ofattention in recent times from the big data managementcommunity due to their efficiency in graph data storage andpowerful graph query specification abilities. In this paperwe present a methodology for modeling workload spikesin a graph database system using a scalable benchmarkingframework called XGDBench. We describe how two maintypes of workload spikes called data spikes and volume spikescan be implemented in the context of graph databases byconsidering realworld workload traces and empirical evidence.We implemented these features on XGDBench which wedeveloped using X10. We validated these features by runningworkloads on Titan which is a popular open source distributedgraph database server.We observed the ability of XGDBench ingenerating realistic workload spikes on Titan. The distributedarchitecture of XGDBench promotes implementation of suchtechniques efficiently through utilization of computing poweroffered by distributed memory compute clusters.