ECCO: Leveraging Cross-Camera Correlations for Efficient Live Video Continuous Learning

Recent advances in video analytics address real-time data drift by continuously retraining specialized, lightweight DNN models for individual cameras. However, the current practice of retraining a separate model for each camera suffers from high compute and communication costs, making it unscalable. We present ECCO, a new video analytics framework designed for resource-efficient continuous learning. The key insight is that the data drift, which necessitates model retraining, often shows temporal and spatial correlations across nearby cameras. By identifying cameras that experience similar drift and retraining a shared model for them, ECCO can substantially reduce the associated compute and communication costs. Specifically, ECCO introduces: (i) a lightweight grouping algorithm that dynamically forms and updates camera groups; (ii) a GPU allocator that dynamically assigns GPU resources across different groups to improve retraining accuracy and ensure fairness; and (iii) a transmission controller at each camera that configures frame sampling and coordinates bandwidth sharing with other cameras based on its assigned GPU resources. We conducted extensive evaluations on three distinctive datasets for two vision tasks. Compared to leading baselines, ECCO improves retraining accuracy by 6.7%-18.1% using the same compute and communication resources, or supports 3.3 times more concurrent cameras at the same accuracy.

CC-Fuzz: Genetic algorithm-based fuzzing for stress testing congestion control algorithms

Congestion control research has experienced a significant increase in interest in the past few years, with many purpose-built algorithms being designed with the needs of specific applications in mind. These algorithms undergo limited testing before being deployed on the Internet, where they interact with other congestion control algorithms and run across a variety of network conditions. This often results in unforeseen performance issues in the wild due to algorithmic inadequacies or implementation bugs, and these issues are often hard to identify since packet traces are not available. In this paper, we present CC-Fuzz, an automated congestion control testing framework that uses a genetic search algorithm in order to stress test congestion control algorithms by generating adversarial network traces and traffic patterns. Initial results using this approach are promising - CC-Fuzz automatically found a bug in BBR that causes it to stall permanently, and is able to automatically discover the well-known low-rate TCP attack, among other things.