Edge Server Monitoring for Job Assignment

TL;DR

This work addresses real-time job assignment in edge computing under limited querying resources by modeling server availability as Markovian and formulating a Restless Multi-Armed Bandit problem. A Net-Gain Maximization policy is developed, combining a relaxed Lagrangian approach with per-dispatcher infinite-horizon MDPs to select queries that maximize the expected long-term job success rate, accounting for Age of Information. The approach leverages a dual decomposition and dynamic programming to compute relative action-values, enabling distributed, near-optimal scheduling across dispatchers. Experimental results show substantial gains over naive strategies, with up to 30% improvement over Round-Robin and over 105–107% improvement over Never-Query policies, demonstrating practical impact for bandwidth-constrained edge environments.

Abstract

In this paper, we study a goal-oriented communication problem for edge server monitoring, where compute jobs arrive intermittently at dispatchers and must be immediately assigned to distributed edge servers. Due to competing workloads and the dynamic nature of the edge environment, server availability fluctuates over time. To maintain accurate estimates of server availability states, each dispatcher updates its belief using two mechanisms: (i) active queries over shared communication channels and (ii) feedback from past job executions. We formulate a query scheduling problem that maximizes the job success rate under limited communication resources for queries. This problem is modeled as a Restless Multi-Armed Bandit (RMAB) with multiple actions and addressed using a Net-Gain Maximization (NGM) scheduling algorithm, which selects servers to query based on their expected improvement in execution performance. Simulation results show that the proposed NGM Policy significantly outperforms baseline strategies, achieving up to a 30% gain over the Round-Robin Policy and up to a 107% gain over the Never-Query Policy.

Figures (3)

• Figure 1: System model: Jobs arrive intermittently at each of the $N$ dispatchers. Upon arrival, a job is immediately assigned by the dispatcher to one of its $K$ servers. Each dispatcher estimates the availability of its $K$ servers using two sources of information: (i) active queries transmitted over $M$ shared communication channels and (ii) feedback obtained from past job executions.
• Figure 2: Average job success rate with increasing number of dispatchers ($N$), where the number of channels is $M=5$, the number of servers per dispatcher is $K=5$, and the job arrival probability is $\lambda_n=0.3$.
• Figure 3: Average job success rate with increasing number of channels ($M$), where the number of dispatchers is $N=15$, the number of servers per dispatcher is $K=5$, and the job arrival probability is $\lambda_n=0.3$.