QEScalor: Quantitative Elastic Scaling Framework in Distributed Streaming Processing.

Recently, researchers usually use the elastic scaling techniques as a powerful means of the distributed stream processing systems to deal with the high-speed data stream which arrives continuously and fluctuates constantly. The existing methods allocate the same amount of resources to the instances of the same operator, but they ignore the correlation between the operator performance and resource provision. It may lead to the waste of the resources caused by the over-provision or the huge overhead of the scheduling caused by the under-provision. To solve the above problems, we present a quantitative elastic scaling framework, named QEScalor, to allocate resources for the operator instances quantitatively based on the actual performance requirements. The experimental results show that compared with the existing works, the QEScalor can not only achieve resource-efficient elastic scaling with lower cost, but also it can enhance the total performance of the DSPAs.