Emerging deep learning-based video analytics tasks demand computation-intensive neural networks and powerful computing resources on the cloud to achieve high accuracy. Due to the latency requirement and limited network bandwidth, edge-cloud systems adaptively compress the data to strike a balance between overall analytics accuracy and bandwidth consumption. However, the degraded data leads to another issue of poortail accuracy, which means the extremely low accuracy of a few semantic classes and video frames. Autonomous robotics applications especially value the tail accuracy performance but suffer using the prior edge-cloud systems. We present Runespoor, an edge-cloud video analytics system to manage the tail accuracy and enable emerging robotics applications. We train and deploy a super-resolution model tailored for the tail accuracy of analytics tasks on the server to significantly improves the performance on hard-to-detect classes and sophisticated frames. During online operation, we use an adaptive data rate controller to further improve the tail performance by instantly adjusting the data rate policy according to the video content. Our evaluation shows that Runespoor improves class-wise tail accuracy by up to 300%, frame-wise 90%/99% tail accuracy by up to 22%/54%, and greatly improves the overall accuracy and bandwidth trade-off.

Emerging deep learning-based video analytics tasks demand computation-intensive neural networks and powerful computing resources on the cloud to achieve high accuracy. Due to the latency requirement and limited network bandwidth, edge-cloud systems adaptively compress the data to strike a balance between overall analytics accuracy and bandwidth consumption. However, the degraded data leads to another issue of poor tail accuracy, which means the extremely low accuracy of a few semantic classes and video frames. Autonomous robotics applications especially value the tail accuracy performance but suffer using the prior edge-cloud systems.We present Runespoor, an edge-cloud video analytics system to manage the tail accuracy and enable emerging robotics applications. We train and deploy a super-resolution model tailored for the tail accuracy of analytics tasks on the server to significantly improves the performance on hard-to-detect classes and sophisticated frames. During online operation, we use an adaptive data rate controller to further improve the tail performance by instantly adjusting the data rate policy according to the video content. Our evaluation shows that Runespoor improves class-wise tail accuracy by up to 300%, frame-wise 90%/99% tail accuracy by up to 22%/54%, and greatly improves the overall accuracy and bandwidth trade-off.

Low latency stream processing on large clusters consisting of hundreds to thousands of servers is an increasingly important challenge. A crucial barrier to tackling this challenge is stragglers, i.e., tasks that are significantly straggling behind others in processing the stream data. However, prior straggler mitigation solutions have significant limitations. They balance streaming workloads among tasks but may incur imbalanced backlogs when the workloads exhibit variance, causing stragglers as well. Fortunately, we observe that carefully scheduling the outgoing tuples of different tasks can yield benefits for balancing backlogs, and thus avoids stragglers. To this end, we present Hone, a tuple scheduler that aims to minimize the maximum queue backlog of all tasks over time. Hone leverages an online Largest-Backlog-First (LBF) algorithm with a provable good competitive ratio to perform efficient tuple scheduling. We have implemented Hone based on Apache Storm and evaluated it extensively via both simulations and testbed experiments. Our results show that under the same workload balancing strategy-shuffle grouping, Hone outperforms the original Storm significantly, with the end-to-end tuple processing latency reduced by 78.7 percent on average.

Indoor localization has been an active research field for decades, where received signal strength (RSS) fingerprinting based methodology is widely adopted and induces many important localization techniques, such as the recently proposed one building fingerprints database with crowdsourcing. While efforts have been dedicated to improve accuracy and efficiency of localization, performance of the RSS fingerprinting based methodology itself is still unknown in a theoretical perspective. In this paper, we present a general probabilistic model to shed light on a fundamental issue: how good the RSS fingerprinting based indoor localization can achieve? Concretely, we present the probability that a user can be localized in a region with certain size. We reveal the interaction among accuracy, reliability, and the number of measurements in the localization process. Moreover, we present the optimal fingerprints reporting strategy that can achieve the best localization accuracy with given reliability and the number of measurements, which provides a design guideline for the RSS fingerprinting based indoor localization system. Further, we analyze the influence of imperfect database information on the reliability of localization, and find that the impact of imperfect information is still under control with reasonable number of samplings when building the database.

Recent study on Wi-Fi RSS fingerprinting based indoor localization reveals that reporting fingerprints with respect to different set of access points (APs) results in location estimations in different levels of accuracy; however, how to find the best strategy for fingerprints reporting with reasonable computational cost, and how to exploit the finding to streamline the design of the localization system are still unknown. In this paper, we revisit the design principles of the localization system with the opportunity provided by the theory of best fingerprints reporting strategy. We first present the localization reliability bounds under the best strategy, and develop algorithms to find the best strategy in practice. We then demonstrate how the best strategy theory can be utilized to improve accuracy of location estimation by resolving the issue of similar fingerprints for both faraway and close-by locations. An iterative algorithm is developed to cross check fingerprints sampled in different locations, in order to derive the best possible result of localization. Moreover, we reveal the relationship between accuracy of location estimation and coverage of Wi-Fi signals in a region, when planning deployment of APs. Experiment results are presented to validate our analysis and design.