Throughput Aware Replica Selection for Multi Region Distributed Systems

Modern Multi region distributed systems rely on data replication to ensure availability, fault tolerance, and low access latency. Replica selection plays a critical role in determining system performance, particularly under varying workloads and heterogeneous regional capacities. As a result, replicas with limited capacity may become overloaded, while higher capacity replicas remain underutilized, leading to inefficient request handling and reduced system throughput. In multi region environments, replicas operate under diverse network conditions, hardware configurations, and workload characteristics. Static routing decisions amplify these disparities by treating all replicas as equivalent, regardless of their ability to sustain incoming request rates. Under high load, this imbalance causes request queues to grow at weaker replicas, increasing processing delays and limiting overall throughput. At the same time, stronger replicas are unable to compensate due to rigid routing policies. As the number of regions increases, these inefficiencies become more pronounced, restricting scalability despite the availability of additional replicas. This paper addresses the problem of throughput degradation in multi region distributed systems by focusing on replica selection behavior. The study emphasizes the importance of aligning request distribution with the runtime throughput characteristics of replicas. By examining throughput trends across different region counts, the work highlights how baseline selection strategies constrain performance when replica capacities vary. The paper concentrates on throughput as the primary performance metric and discusses the need for selection mechanisms that reflect actual processing capability. Addressing replica selection from a throughput perspective enables more effective utilization of distributed resources and supports scalable operation in geographically distributed systems.