Elastic Resource Provisioning for Batched Stream Processing System in Container Cloud

Batched stream processing systems achieve higher throughput than traditional stream processing systems while providing low latency guarantee. Recently, batched stream processing systems tend to be deployed in cloud due to their requirement of elasticity and cost efficiency. However, the performance of batched stream processing systems are hardly guaranteed in cloud because static resource provisioning for such systems does not fit for stream fluctuation and uneven workload distribution. In this paper, we propose EStream: an elastic batched stream processing system based on Spark Streaming, which transparently adjusts available resource to handle workload fluctuation and uneven distribution in container cloud. Specifically, EStream can automatically scale cluster when resource insufficiency or over-provisioning is detected under the situation of workload fluctuation. On the other hand, it conducts resource scheduling in cluster according to the workload distribution. Experimental results show that EStream is able to handle workload fluctuation and uneven distribution transparently and enhance resource efficiency, compared to original Spark Streaming.