Multi-Cluster Orchestration: Performance Benchmarking Of Cross-Cluster Service Meshes In High-Traffic Retail Environments

Abstract

Modern retail enterprises operate distributed microservices architectures spanning multiple Kubernetes clusters across geographic regions to ensure high availability, regulatory compliance, and optimal customer experience. Service meshes provide critical capabilities for inter-service communication, observability, and security, yet their performance characteristics in multi-cluster configurations under high-traffic retail workloads remain insufficiently understood. This research conducts comprehensive performance benchmarking of leading service mesh solutions—Istio, Linkerd, and Consul—in multi-cluster configurations representative of large-scale retail environments. Through systematic evaluation across three major retail organizations during peak traffic periods including Black Friday and holiday shopping seasons, we measure latency, throughput, resource consumption, and failure recovery characteristics under realistic workloads reaching 50,000 requests per second. Our findings reveal significant performance variations across service mesh implementations, with Linkerd demonstrating 23% lower p99 latency and 31% reduced CPU overhead compared to Istio in multi-cluster scenarios, while Consul exhibited superior cross-cluster service discovery performance but higher memory consumption. The research identifies critical trade-offs between feature richness and performance overhead, revealing that cross-cluster traffic routing introduces 8-15ms additional latency depending on mesh architecture. We contribute empirically validated performance profiles enabling architects to select appropriate service mesh solutions based on specific retail workload characteristics, traffic patterns, and operational constraints.