Towards Dynamic Resource Provisioning for Traffic Mining Service Cloud

Real-time traffic data, especially floating car GPS data, has been collected in massive scale and is becoming increasingly rich, complex, and ubiquitous. Data mining approaches are necessary to design effective urban traffic patterns from massive historic traffic data sets. We have built RTIC-C system for traffic data mining based on cloud computing technique for its ability of ´´big data´´ processing and distributed map-reduce computing framework. However when more and more mining applications run on this platform, we need to dispatch enough resources but with minimum cost, like virtual machines, on demand to adapt to different mining requirements with budget or QoS constraints. In this paper, we firstly promoted a micro-kernel container for traffic mining services supporting light-weighted and measurable resource utilization, then we schemed a dynamic resource provisioning algorithm to predict resource utilization considering temporal and cost factors. Experiments on several metrics showed that our model achieved considerable performance and supported elastic computing with dynamic resource provisioning.