Motivated by the imperative for real-time responsiveness and data privacy preservation, large language models (LLMs) are increasingly deployed on resource-constrained edge devices to enable localized inference. To improve output quality, retrieval-augmented generation (RAG) is an efficient technique that seamlessly integrates local data into LLMs. However, existing edge computing paradigms primarily focus on single-node optimization, neglecting opportunities to holistically exploit distributed data and heterogeneous resources through cross-node collaboration. To bridge this gap, we propose CoEdge-RAG, a hierarchical scheduling framework for retrieval-augmented LLMs in collaborative edge computing. In general, privacy constraints preclude accurate a priori acquisition of heterogeneous data distributions across edge nodes, directly impeding RAG performance optimization. Thus, we first design an online query identification mechanism using proximal policy optimization (PPO), which autonomously infers query semantics and establishes cross-domain knowledge associations in an online manner. Second, we devise a dynamic inter-node scheduling strategy that balances workloads across heterogeneous edge nodes by synergizing historical performance analytics with real-time resource thresholds. Third, we develop an intra-node scheduler based on online convex optimization, adaptively allocating query processing ratios and memory resources to optimize the latency-quality trade-off under fluctuating assigned loads. Comprehensive evaluations across diverse QA benchmarks demonstrate that our proposed method significantly boosts the performance of collaborative retrieval-augmented LLMs, achieving performance gains of 4.23 % to 91.39% over baseline methods across all tasks.

Due to the penetration of edge computing, a wide variety of workloads are sunk down to the network edge to alleviate huge pressure of the cloud. With the presence of high input workload dynamics and intensive edge resource contention, it is highly non-trivial for an edge proxy to optimize the scheduling of heterogeneous services with diverse QoS requirements. In general, online services should be quickly completed in a quite stable running environment to meet their tight latency constraint, while offline services can be processed loosely for their elastic soft deadlines. To well coordinate such services at the resource-limited edge cluster, in this paper, we study an edge-centric resource provisioning optimization for dynamic online and offline services co-location, where the proxy seeks to maximize timely online service performances while maintaining satisfactory long-term offline service performances. However, intricate hybrid couplings for provisioning decisions arise due to heterogeneous constraints of the co-located services and their different time-scale performances. We hence first propose a reactive provisioning approach without requiring a prior knowledge of future system dynamics, which leverages a Lagrange relaxation for devising constraint-aware stochastic subgradient algorithm to deal with the challenge of hybrid couplings. To further boost the performance by integrating powerful machine learning techniques, we then advocate a predictive provisioning approach, where future request arrivals can be estimated accurately. To align with practical deployments, we incorporate a tunable prediction window mechanism, which well balances the potential improvement and degradation of online performance in imperfect prediction scenarios. With rigorous theoretical analysis and extensive trace-driven evaluations, we show the superior performance of our proposed algorithms for online and offline services co-location at the edge.