Prompt: Dynamic Data-Partitioning for Distributed Micro-batch Stream Processing Systems

Advances in real-world applications require high-throughput processing over large data streams. Micro-batching has been proposed to support the needs of these applications. In micro-batching, the processing and batching of the data are interleaved, where the incoming data tuples are first buffered as data blocks, and then are processed collectively using parallel function constructs (e.g., Map-Reduce). The size of a micro-batch is set to guarantee a certain response-time latency that is to conform to the application's service-level agreement. In contrast to tuple-at-a-time data stream processing, micro-batching has the potential to sustain higher data rates. However, existing micro-batch stream processing systems use basic data-partitioning techniques that do not account for data skew and variable data rates. Load-awareness is necessary to maintain performance and to enhance resource utilization. A new data partitioning scheme termed Prompt is presented that leverages the characteristics of the micro-batch processing model. In the batching phase, a frequency-aware buffering mechanism is introduced that progressively maintains run-time statistics, and provides online key-based sorting as data tuples arrive. Because achieving optimal data partitioning is NP-Hard in this context, a workload-aware greedy algorithm is introduced that partitions the buffered data tuples efficiently for the Map stage. In the processing phase, a load-aware distribution mechanism is presented that balances the size of the input to the Reduce stage without incurring inter-task communication overhead. Moreover, Prompt elastically adapts resource consumption according to workload changes. Experimental results using real and synthetic data sets demonstrate that Prompt is robust against fluctuations in data distribution and arrival rates. Furthermore, Prompt achieves up to 200% improvement in system throughput over state-of-the-art techniques without degradation in latency.