Search for: All records

Modern developers rely on container-orchestration frameworks like Kubernetes to deploy and manage hybrid workloads that span the edge and cloud. When network conditions between the edge and cloud change unexpectedly, a workload must adapt its internal behavior. Unfortunately, container-orchestration frameworks do not offer an easy way to express, deploy, and manage adaptation strategies. As a result, fine-tuning or modifying a workload's adaptive behavior can require modifying containers built from large, complex codebases that may be maintained by separate development teams. This paper presents BumbleBee, a lightweight extension for container-orchestration frameworks that separates the concerns of application logic and adaptation logic. BumbleBee provides a simple in-network programming abstraction for making decisions about network data using application semantics. Experiments with a BumbleBee prototype show that edge ML-workloads can adapt to network variability and survive disconnections, edge stream-processing workloads can improve benchmark results between 37.8% and 23x , and HLS video-streaming can reduce stalled playback by 77%. 

Increasingly, vehicles sold today are connected cars: they offer vehicle-to-infrastructure connectivity through built-in WiFi and cellular interfaces, and they act as mobile hotspots for devices in the vehicle. We study the connection quality available to connected cars today, focusing on user-facing, latency-sensitive applications. We find that network latency varies significantly and unpredictably at short time scales and that high tail latency substantially degrades user experience. We also find an increase in coverage options available due to commercial WiFi offerings and that variations in latency across network options are not well-correlated. Based on these findings, we develop RAVEN, an in-kernel MPTCP scheduler that mitigates tail latency and network unpredictability by using redundant transmission when confidence about network latency predictions is low. RAVEN has several novel design features. It operates transparently, without application modification or hints, to improve interactive latency. It seamlessly supports three or more wireless networks. Its in-kernel implementation allows proactive cancellation of transmissions made unnecessary through redundancy. Finally, it explicitly considers how the age of measurements affects confidence in predictions, allowing better handling of interactive applications that transmit infrequently and networks that exhibit periods of temporary poor performance. Results from speech, music, and recommender applications in both emulated and live vehicle experiments show substantial improvement in application response time 

Vehicular applications must not demand too much of a driver's attention. They often run in the background and initiate interactions with the driver to deliver important information. We argue that the vehicular computing system must schedule interactions by considering their priority, the attention they will demand, and how much attention the driver currently has to spare. Based on these considerations, it should either allow a given interaction or defer it. We describe a prototype called Gremlin that leverages edge computing infrastructure to help schedule interactions initiated by vehicular applications. It continuously performs four tasks: (1) monitoring driving conditions to estimate the driver's available attention, (2) recording interactions for analysis, (3) generating a user-specific quantitative model of the attention required for each distinct interaction, and (4) scheduling new interactions based on the above data. Gremlin performs the third task on edge computing infrastructure. Offload is attractive because the analysis is too computationally demanding to run on vehicular platforms. Since recording size for each interaction can be large, it is preferable to perform the offloaded computation at the edge of the network rather than in the cloud, and thereby conserve wide-area network bandwidth. We evaluate Gremlin by comparing its decisions to those recommended by a vehicular UI expert. Gremlin's decisions agree with the expert's over 90% of the time, much more frequently than the coarse-grained scheduling policies used by current vehicle systems. Further, we find that offloading of analysis to edge platforms reduces use of wide-area networks by an average of 15MB per analyzed interaction.