Quartz

Abstract

Geo-distributed systems ranging from databases to cyber-physical applications increasingly rely on a shared and precise notion of time to achieve coordination. This is especially true for cyber-physical applications ranging from local-scale robotic-coordination and city-scale traffic management to regional/planetary-scale smart grids. Each of these applications utilizes event orderings and timing offsets to make real-time decisions, so as to perform coordinated action at their distributed endpoints. The emergence of edge computing, specifically to facilitate low-latency decision-making, is leveraging the trend where multiple cyber-physical and software applications with different timing requirements will coexist in both the cloud and at the edge. To enable such fault-tolerant time-based coordinated applications running on multi-tenant geo-scale infrastructure, we introduce the Quartz framework, which exposes Time-as-a-Service. Quartz allows geo-distributed application components to each specify its timing requirements, while it autonomously orchestrates the underlying infrastructure to meet them. Centered around a shared virtualized notion of time, based on the time-line abstraction [1], Quartz provides an API which makes it easy to develop time-based geo-distributed applications. Using this API, Quartz feeds back the timing uncertainty, i.e., the delivered Quality of Time (QoT) [1] back to each application, enabling it to be fault-tolerant in the face of clock-synchronization failure. Quartz is designed for containerized applications, features a distributed architecture and is implemented using containerized micro-services. Experimental evaluations on real-world embedded, edge and cloud platforms highlight the performance and scalability of our architecture.