关键词: 交通监控视频,HDFS,交通事件,数据放置,吞吐量

Abstract: Utilizing HDFS to store and process large scale traffic surveillance video is a reliable,highly efficient and scalable solution.However,the default rack awareness data placement strategy in HDFS may cause hotspots when placing data.To address this problem,this paper presented a traffic surveillance video data placement strategy based on traffic event density.The characteristic of traffic surveillance videos allows us to classify them according to traffic event types.When placing new data,the proposed strategy predicts the load of each datanode which is influenced by various of traffic events the datanode stores,then combines the instant load and disk capacity to evaluate each datanode,and chooses the most suitable datanode to store new data.Experiments show that the proposed strategy alleviates the hotspot problem and effectively improves the load balancing and throughput in comparison with the default strategy.

Key words: Traffic surveillance video,HDFS,Traffic event,Data placement,Throughput