Locality-Aware Routing in Stateful Streaming Applications

Abstract

International audienceDistributed stream processing engines continuously execute series of operators on data streams. Horizontal scaling is achieved by deploying multiple instances of each operator in order to process data tuples in parallel. As the application is distributed on an increasingly high number of servers, the likelihood that the stream is sent to a different server for each operator increases. This is particularly important in the case of stateful applications that rely on keys to deterministically route messages to a specific instance of an operator. Since network is a bottleneck for many stream applications, this behavior significantly degrades their performance. Our objective is to improve stream locality for stateful stream processing applications. We propose to analyse traces of the application to uncover correlations between the keys used in successive routing operations. By assigning correlated keys to instances hosted on the same server, we significantly reduce network consumption and increase performance while preserving load balance. Furthermore, this approach is executed online, so that the assignment can automatically adapt to changes in the characteristics of the data. Data migration is handled seamlessly with each routing configuration update. We implemented and evaluated our protocol using Apache Storm, with a real workload consisting of geo-tagged Flickr pictures as well as Twitter publications. Our results show a significant improvement in throughput

Runtime Adaptation of Data Stream Processing Systems: The State of the Art

“…Several research works investigate load distribution and routing strategies (e.g., [3,23,31,50,92,95,107,143]). For instance, Rivetti et al [143] present a solution to balance load among parallel instances of a stateless operator, accounting for variable tuple processing times.…”

“…Several solutions, especially those acting on data streams (e.g., load distribution, shedding), perform adaptation with iner granularity, at level of single tuples (e.g., [3,23,50,95,168,185]) or batches of tuples (e.g., [38,177]). Solutions acting at the infrastructure level usually work with the granularity of the computing node (e.g., [47,82,176]) or the network link [5].…”

Towards a High-Level Description for Generating Stream Processing Benchmark Applications

“…extraordinary events in sensor-networks [23] -and can follow a bimodal human behavior -e.g. day and night activities [24].…”

Dynamic Tuple Scheduling with Prediction for Data Stream Processing Systems

“…Amongst such systems, Apache Storm [4] and its successor Apache Heron [5] have enjoyed the most intensive attraction from both industries and academia since released. Existing works mainly focus on optimizing the placement, re-scaling, and migration of processing instances to improve the average response time of tuple processing [15] [21]- [24], overall throughput [14] [25] [26], resource utilization [27], and other performance metrics such as charges for communication [28] and revenue loss due to QoS violation [29] [30]. The focuses of such works are orthogonal to and can be well integrated with our devised scheme.…”

“…Update I cand (i) ← I cand (i)\{i * }. 15: Update instances' queue backlogs according to (7) - (9). an efficient and distributed scheme that solves problem (15) optimally and present its pseudocode in Algorithm 1.…”