Hierarchical DNS Anycast Routing for Edge Server Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing CDN routing systems face challenges in efficiently managing load balancing across geographically distributed edge servers, particularly when one server becomes overloaded, leading to bottlenecks and increased overhead in transmitting load information and reestablishing connections.
Innovation Solution
A routing system that utilizes DNS anycast IP addresses and domain names across multiple layers to dynamically offload traffic from overloaded edge servers by directing clients to less loaded servers within the same or lower layers, reducing the need for external load balancing servers and minimizing connection disruptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If edge servers are geographically distributed to reduce routing time, then message delivery speed is improved, but load balancing becomes difficult and bottlenecks occur when one server is overloaded
Solution Approach 1:
The system segments the CDN into multiple hierarchical layers (origin servers, regional edge servers, and local edge servers). Each layer handles specific routing responsibilities, with local edge servers serving nearby clients first, regional edge servers handling broader regions, and origin servers providing fallback capacity. This segmentation enables automatic load distribution without complex centralized control.
Solution Approach 2:
The patent introduces a hierarchical dimensional structure with multiple layers of edge servers at different geographical and functional levels. Instead of a flat single-layer architecture, the system creates vertical layers (origin → regional → local) and horizontal geographical distribution, adding dimensional complexity to simplify the load balancing problem through natural hierarchical routing.
2Productivity
If centralized load balancing servers are used to monitor and distribute traffic, then load distribution is improved, but system overhead and complexity increase
Solution Approach 1:
Each edge server autonomously monitors its own load conditions and makes independent decisions about request routing. When an edge server becomes overloaded, it automatically redirects requests to peer servers or upper-layer servers without needing to report to or receive instructions from a centralized load balancer. This self-service mechanism eliminates the need for complex centralized control infrastructure.
Solution Approach 2:
The system implements distributed feedback mechanisms where edge servers continuously monitor their own performance metrics (CPU utilization, memory usage, request queue length) and use this feedback to dynamically adjust their routing decisions. This decentralized feedback loop enables adaptive load balancing without centralized coordination overhead.
3Speed
If edge servers cache content locally, then response time is improved, but memory usage and cache management complexity increase
Solution Approach 1:
Different layers of the hierarchical CDN structure have different caching strategies tailored to their specific roles. Local edge servers cache frequently accessed content for immediate local delivery, regional edge servers cache broader content sets for their regions, and origin servers maintain complete content libraries. This localized quality approach optimizes memory usage at each level based on its specific performance requirements.
Data Source
AI summary
N nodes are assigned to a first layer of nodes having a first domain name server (DNS) anycast Internet Protocol (IP) address and a first fully qualified domain name (FQDN). M nodes are assigned to a second layer of nodes having a second DNS anycast IP address and a second FQDN. When a request to resolve the first FQDN for the first layer of nodes is received by a DNS of a node of the first layer and a load on the ES of the node is less than a threshold, the DNS handles the request by returning an edge server (ES) anycast IP address for the ES of the node. When the load on the ES is greater than the threshold, the DNS offloads the request by returning the second FQDN so that the second FQDN of the second layer is resolved to the second DNS anycast IP address.