A CDN-based Domain Name System

Abstract

To enhance end-user experiences, Content Distribution Networks (CDNs) effectively exploit the Domain Name System (DNS) to redirect end-users to close-by content replicas over short time scales. While the use of DNS has brought a significant advantage to CDNs, in this paper we confirm that reliance on DNS also poses a fundamental threat to large-scale CDNs’ content distribution model. In particular, we demonstrate that a considerable penetration of public DNS resolving services (e.g., OpenDNS and GoogleDNS) effectively corrupts the CDN approach, equally the large-scale server distribution and quick DNS redirections. Our contributions are threefold. First, we systematically evaluate and quantify how the use of public DNS resolving services impacts Akamai’s content distribution model, the corresponding end-user service performance, and ISPs that host CDN edge servers. Second, we show that a CDN-based DNS architecture, which can effectively function as both the current authoritative nameservers and resolvers, coexist with the current DNS infrastructure, and directly response end-users with authoritative DNS records. Third, we implement a prototype CDN-based DNS system by using popular Web 2.0 websites as a vehicle to publish DNS records onto CDN edge servers. We demonstrate that in an extreme setting, such an approach can achieve one order of magnitude faster lookup times relative to existing DNS resolving systems. Most importantly, we argue that the proposed design provides a realistic way to change DNS and address a number of long-lasting and emerging problems associated with this basic Internet service.