Towards Quality-Assured Data Delivery in Cloud-Based IoT Platforms for Smart Cities

Abstract

Adaptive cloud elasticity is a key requirement and an open challenge for the cost-effective and performance-efficient allocation of resources for cloud-based applications. It becomes even more challenging for IoT applications due to the dynamic and unpredictable characteristics of IoT data including volume, velocity, variety, volatility, and heterogeneity. In this paper, we study the impact of varying velocity and volume of IoT data streams on (a) the performance of the data ingestion and storage process, and (b) the corresponding cloud resource utilization. We use the open-source OpenIoT platform as our test bed and run experiments under different data streaming settings. We study the throughput and error rate, as well as CPU and memory consumption during the ingestion process. Experimental results reveal some interesting insights into the performance of OpenIoT and allow us to draw more general observations for cloud-based IoT platforms, including (a) resource availability is not the only factor influencing performance of the data ingestion process, with the middleware and software components having significant impact, (b) allocating extra resources at the infrastructure layer of the IoT platform can potentially improve the performance of the data ingestion process, but may lead to resource wastage, and (c) adaptive scaling out/in using small sized compute units can significantly improve resource utilization while ensuring quality-assured data delivery. Based on these insights, we identify two key future research directions for quality-assured data delivery in cloud-based IoT platforms including adaptive resource optimization at the infrastructure layer and component configuration optimization and/or upgradation at the middleware layer.