Benchmark based performance characterization and scalable design of sensing-as-a-service platform

Advancements and proliferation of sensing and actuation technologies in diverse computing and physical devices call for a flexible Sensing-as-a-Service Platform (SeaaS-P) that can enable massive-scale next generation applications. A key aspect of such a platform, which has been largely unexplored in terms of applicability at a massive scale, is the notification system (e.g. city incident notification) to send event alerts to subscribers. Conventional event-based systems (e.g. publish-subscribe) typically trigger notifications with every event arrival. Observations show resources (e.g. CPU) can become bottleneck and get overused when event rates and number of subscriptions go beyond certain thresholds. Thus, it may be necessary to opportunistically delay the notification, i.e. not send the notifications with every event update (e.g., time-triggered periodic notifications), depending on the delay-tolerance of the application. This paper proposes a: (i) benchmark based performance profiling of SeaaS-P in terms of resource utilizations and % notification drop; and (ii) design framework to determine the design boundaries of SeaaS-P notification system for both event-triggered and time-triggered notifications.