Edge Multilevel Edge Server Co-operation in Content Delivery Network using Hierarchical Classification

Debabrata Sarddar, Subhankar Roy, Priyajit SenInternational Journal of Grid and Distributed Computing 2017 · DOI

Abstract

A Content Delivery Network (CDN) is a system of distributed servers that deliver web pages and other web contents to a user from a suitable Edge server. The purpose of a CDN is to provide web contents faster and with a higher availability. This service is effective in speeding up the delivery of content of websites with high traffic and websites that have global reach. The edge servers are cache servers that are deployed over the geography. These servers are used to store same contents from main server i.e. the content replicas are cached in these servers. User's requests for a web content may be served directly from any of these edge servers instead of the main server or origin server. This approach overcomes the limitations of network infrastructure in the internet. In this paper we propose a way to select appropriate edge servers when there is cache miss in the primary Edge Server up to the Origin Server. We have developed a technique for CDN provider to route user's request from the requesting machine to edge servers or to the origin in a Content Delivery Network (CDN) by applying hierarchical classification of edge servers to cluster them based on their response delay i.e. effectively based on the distance among the edge server.

Introduction: Addressing the CDN "Cliff"

As global internet traffic continues to grow, Content Delivery Networks (CDNs) have become essential infrastructure, placing "edge servers" geographically close to users to minimize latency. However, traditional CDN architectures often suffer from a "cliff" problem: when a local edge server experiences a cache miss, the request is typically routed directly back to the central "origin" server.

Global distribution of Origin and Edge servers in a typical CDN.

This standard approach introduces two significant challenges. First, it places an immense burden on the origin server, which can become a performance bottleneck during high-traffic events. Second, it ignores the high probability that a neighboring edge server—potentially just a few milliseconds away—might already have the requested data. This paper proposes a hierarchical cooperation strategy to bridge this gap.

Methodology: Agglomerative Hierarchical Clustering

The core of the proposed system is the application of Agglomerative Hierarchical Clustering to organize the CDN nodes into a structured, cooperative network.

1. Latency-Based Distance Metric

Instead of relying on physical GPS coordinates, the authors utilize network latency (measured via 'ping' response times) as the primary "distance" metric. This approach accounts for real-world network congestion and routing complexities that physical distance often ignores.

2. Bottom-Up Clustering Process

The algorithm begins by treating every edge server and the origin server as individual clusters. It then iteratively merges the two "closest" clusters (those with the lowest mutual latency) until a single, hierarchical tree—or dendrogram—is formed.

Hierarchical clustering of edge servers and the origin server based on network latency.

3. Hierarchical Routing Logic

When a user request arrives:

  • The request is first handled by the primary (nearest) edge server.
  • In the event of a cache miss, the server consults a local "content table"—an index of its neighbors' cached data—and forwards the request to the next server up the hierarchical tree.
  • The request only reaches the origin server if it cannot be satisfied by any of the cooperating peers within the cluster hierarchy.

Key Insights and Qualitative Results

The proposed architecture treats the Origin Server as a peer in the clustering process, ensuring it naturally sits at the top of the hierarchy as the ultimate fail-safe. By maintaining local content tables, edge servers can instantly determine if a neighbor has the required data, avoiding unnecessary network hops.

While the paper focuses on architectural logic rather than quantitative benchmarks, the authors conclude that this multilevel approach offers several key benefits:

  • Reduced Origin Load: By satisfying more requests within edge clusters, the central infrastructure is protected from traffic spikes.
  • Improved Reliability: The hierarchical tree provides pre-calculated alternative paths, making the system more robust against individual server failures.
  • Minimized Response Delay: Users receive content faster as the "search radius" expands incrementally through low-latency neighbors rather than jumping immediately to a distant origin.

Limitations and Future Directions

A primary concern identified by the authors is the "Deviant Request" Problem. If a user's requests are highly unique, the system might perform multiple successive hops through the hierarchy only to find a cache miss at every level, potentially increasing total latency compared to a direct origin request.

Future research should focus on empirical validation in simulated environments (such as NS-3) to quantify the trade-off between reduced origin load and hop-count overhead. Additionally, mechanisms for dynamic re-clustering will be necessary to adapt to real-time changes in network topology and congestion.

Summary updated

Abstract

A Content Delivery Network (CDN) is a system of distributed servers that deliver web pages and other web contents to a user from a suitable Edge server. The purpose of a CDN is to provide web contents faster and with a higher availability. This service is effective in speeding up the delivery of content of websites with high traffic and websites that have global reach. The edge servers are cache servers that are deployed over the geography. These servers are used to store same contents from main server i.e. the content replicas are cached in these servers. User's requests for a web content may be served directly from any of these edge servers instead of the main server or origin server. This approach overcomes the limitations of network infrastructure in the internet. In this paper we propose a way to select appropriate edge servers when there is cache miss in the primary Edge Server up to the Origin Server. We have developed a technique for CDN provider to route user's request from the requesting machine to edge servers or to the origin in a Content Delivery Network (CDN) by applying hierarchical classification of edge servers to cluster them based on their response delay i.e. effectively based on the distance among the edge server.
BibTeX
@article{sarddar2017edge,
  author = {Debabrata Sarddar and Subhankar Roy and Priyajit Sen},
  title = {{Edge Multilevel Edge Server Co-operation in Content Delivery Network using Hierarchical Classification}},
  journal = {International Journal of Grid and Distributed Computing},
  year = {2017},
  doi = {10.14257/ijgdc.2017.10.3.04},
}
Figure