DDSS: Federated Fog-Based Microservice Scheduling for IoT-Enabled ICUs

Abstract

The continuous expansion of connected medical devices in the Intensive Care Unit (ICU) environment has led to the emergence of certain classes of computational problems which cannot be addressed effectively using classical scheduling mechanisms on servers. The static allocation schemes do not work well with the heterogeneity and burstiness of modern IoT ICU environments. This paper introduces the Dynamic Dedicated Server Scheduling (DDSS) framework a microservice-oriented resource orchestration architecture that unifies Fog Computing and Federated Learning (FL) to deliver low-latency, privacy-preserving, and adaptive scheduling for ICU-grade IoT environments. Within the DDSS model, patient monitoring workloads are decomposed into five independently deployable microservice categories and mapped dynamically onto a tiered fog infrastructure using a context-aware, multi-objective scheduling engine. A federated learning layer enables the eight edge nodes to collaboratively update the scheduling model while keeping the underlying patient data at the local nodes. This design supports the low-latency requirements of ICU operations while also addressing healthcare data governance constraints. To evaluate the proposed DDSS framework, a 72-hour ICU workload simulation was performed and compared with three conventional scheduling approaches: First-Come-First-Served (FCFS), Round-Robin (RR), and Priority Queue (PQ). Under heavy-load conditions, DDSS recorded a mean completion latency of 61.3 ms for Priority-Critical microservices, which was 69.1% lower than that of the best-performing baseline scheduler. Resource utilization improved to 82.4% with a deadline satisfaction rate of 97.3% for life-critical tasks, and scheduling throughput reached 5,460 tasks per hour a 40.4% gain over the PQ baseline. The federated gradient overhead was always below 38.7 KB per round per node, validating that the privacy-enabled intelligence layer does not pose any excessive network costs. This research offers an interesting perspective into achieving computationally intelligent, clinically adaptive, and regulatory compliant IoT healthcare architecture.