Load balancing, operator instance collocations and horizontal scaling are critical issues in Parallel Stream Processing Engines to achieve low data processing latency, optimized cluster utilization and minimized communication cost respectively. In previous work, these issues are typically tackled separately and independently. We argue that these problems are tightly coupled in the sense that they all need to determine the allocations of workloads and migrate computational states at runtime. Optimizing them independently would result in suboptimal solutions. Therefore, in this paper, we investigate how these three issues can be modeled as one integrated optimization problem. In particular, we first consider jobs, where workload allocations have little effect on the communication cost, and model the problem of load balance as a Mixed-Integer Linear Program. Afterwards, we present an extended solution called ALBIC, which supports general jobs. We implement the proposed techniques on top of Apache Storm, an open-source Parallel Stream Processing Engine. The extensive experimental results over both synthetic and real datasets show that our techniques clearly outperform existing approaches.

The emerging interest in Massively Parallel Stream Processing Engines (MPSPEs), which are able to process long-standing computations over data streams with ever-growing velocity at a large-scale cluster, calls for efficient dynamic resource management techniques to avoid any waste of resources and/or excessive processing latency. In this paper, we propose an approach to integrate dynamic resource management with passive fault-tolerance mechanisms in a MPSPE so that we can harvest the checkpoints prepared for failure recovery to enhance the efficiency of dynamic load migrations. To maximize the opportunity of reusing checkpoints for fast load migration, we formally define a checkpoint allocation problem and provide a pragmatic algorithm to solve it. We implement all the proposed techniques on top of Apache Storm, an open-source MPSPE, and conduct extensive experiments using a real dataset to examine various aspects of our techniques. The results show that our techniques can greatly improve the efficiency of dynamic resource reconfiguration without imposing significant overhead or latency to the normal job execution.

Recently there has been an increasing interest in building distributed platforms for processing of fast data streams. In this demonstration, we highlight the need for elasticity in distributed data stream processing systems and present Enorm, a data stream processing platform with focus on elasticity, i.e. the ability to dynamically scale resource usage according to the runtime workload fluctuations. In order to achieve dynamic scaling with minimal overhead and latency, we use an integrated approach for both fault-tolerance and elasticity. The idea is that both fault-tolerance and elasticity essentially require replicating or migrating computation states among different nodes. Integrating and sharing the state management operations between the two modules can not only provide abundant opportunities to reduce the system's runtime overhead but also simplify the system's architecture.