Statistically managing cloud operations for latency-tail-tolerance in IoT-enabled smart cities

Abstract

Smart City is typically large scale and IoT-enabled online services for huge amount of streaming data generated by sensors, and the services are often deployed in clouds, where infrastructure and services need to be maintained and optimised to achieve the best quality of service by using various cloud operations that are running in the background of Smart City business. However, on one hand the background operations inevitably press a negative impact on the latency of normal requests to the services, while on the other hand to maintain the short latency of requests usually results in unreasonably long latency of background operations. In this paper, on the architecture of IoT-enabled Smart City services deployed in clouds, our motivation is to find the best management policy of operations for normal request traffic, which is stable and stationary and to which operations are inserted. We focus on the system environment in which administrators are dealing with in most time other than discuss the special cases, such as outages or rush hours, and take the assumptions e.g. predictability and periodicity. The suitable management techniques for us are a class of online management policies, which are simple and applicable in practice. These policies harmonise the requests and the operations by giving higher priority to the normal requests other than forcing the background operations to wait unreasonably. We also propose a statistical policy, which is asymptotically optimal in the class. In order to compute the numeric solution, we employ a queueing model to analyse our policy. The evaluation indicates that our technique can effectively maintain the short latency of the requests, and meanwhile the latency of operations is not unreasonably long in comparison to the competitors.