Most of the current research work on timely streaming data processing focuses on minimizing average tuple latency instead of strict individual tuple latency upper-bound, that is, deadline. In this paper, we propose a novel deadline-scheduling strategy, namely tick scheduling (TS), dealing with applications with specified deadline constraints over high volume, possibly bursting, and continuous data streams. We demonstrate that TS policy, which combines precise batch scheduling plan construction and adaptive batch maintenance mechanism can significantly improve system performance by greatly reducing system overheads and adapting gracefully to the time-varying data arrival-rate. Experimental results show the significant improvements provided by our proposed policy.
展开▼