🤖 AI Summary
This paper addresses the information freshness optimization problem in multi-source single-server mobile edge computing systems, using Peak Age of Information (Peak AoI) as the performance metric. We jointly optimize the scheduling policy for stochastically arriving sources and the status sampling mechanism. Our key theoretical contribution is the rigorous proof that randomized schedulers are optimal under both preemptive and non-preemptive settings. Leveraging this insight, we decouple the joint optimization into an alternating optimization over scheduling frequency and sampling threshold, and propose a transmission-aware threshold-based sampling policy along with a general alternating optimization framework. The proposed algorithm effectively handles the inherent non-convexity of the problem. Numerical experiments demonstrate that it closely approaches the theoretical optimum across diverse system configurations, significantly reducing Peak AoI while achieving a balanced co-design of communication and computation resources.
📝 Abstract
Age of Information (AoI) is emerging as a novel metric for measuring information freshness in real-time monitoring systems. For computation-intensive status data, the information is not revealed until being processed. We consider a status update problem in a multi-source single-server system where the sources are scheduled to generate and transmit status data which are received and processed at the edge server. Generate-at-will sources with both random transmission time and process time are considered, introducing the joint optimization of source scheduling and status sampling on the basis of transmission-computation balancing. We show that a random scheduler is optimal for both non-preemptive and preemptive server settings, and the optimal sampler depends on the scheduling result and its structure remains consistent with the single-source system, i.e., threshold-based sampler for non-preemptive case and transmission-aware deterministic sampler for preemptive case. Then, the problem can be transformed to jointly optimizing the scheduling frequencies and the sampling thresholds/functions, which is non-convex. We proposed an alternation optimization algorithm to solve it. Numerical experiments show that the proposed algorithm can achieve the optimal in a wide range of settings.