Distance-Aware Cache Server Placement for Anycast CDNs via Genetic Algorithm

Abstract

In anycast content delivery networks (CDNs), client requests are naturally routed to the topologically closest cache server (CS) via the border gateway protocol (BGP). However, BGP routing policies frequently cause path inflation, directing clients to geographically distant servers and severely degrading delivery latency. While deploying a vast number of CSs is a common mitigation strategy, it inherently exacerbates BGP routing complexity and increases the risk of suboptimal server selection. To address this, we proposed operating multiple content-specific server sets with a limited number of CSs, leveraging the spatial locality of content demand. Yet, existing optimization models rely on coarse country-level demand coverage, failing to capture fine-grained, cross-border latency benefits. In this paper, a novel anycast CDN framework is proposed to optimize server placement using a genetic algorithm (GA) driven by a precise, geographical distance-based fitness function. The proposed approach is evaluated using realistic global datasets and empirical anycast routing distortion models. Simulation results demonstrate that the distance-based method significantly outperforms conventional coverage-based approaches, achieving lower average distance costs and highly stable optimization convergence, thereby enabling a more resilient and globally optimized CDN architecture.