From req/res to pub/sub: Exploring Media over QUIC Transport for DNS

TL;DR

This paper proposes a publish-subscribe extension to DNS using Media-over-QUIC to enable active distribution of resource record updates, improving update efficiency and reducing latency for certain use cases.

Contribution

It introduces a novel publish-subscribe DNS architecture based on Media-over-QUIC, including a prototype implementation and protocol design.

Findings

Reduces update traffic and latency for DNS record dissemination

Supports use cases like load balancing in content delivery networks

Introduces new challenges such as increased overhead and initial query latency

Abstract

The DNS is a key component of the Internet. Originally designed to facilitate the resolution of host names to IP addresses, its scope has continuously expanded over the years, today covering use cases such as load balancing or service discovery. While DNS was initially conceived as a rather static directory service in which resource records (RR) only change rarely, we have seen a number of use cases over the years where a DNS flavor that isn't purely based upon requesting and caching RRs, but rather on an active distribution of updates for all resolvers that showed interest in the respective records in the past, would be preferable. In this paper, we thus explore a publish-subscribe variant of DNS based on the Media-over-QUIC architecture, where we devise a strawman system and protocol proposal to enable pushing RR updates. We provide a prototype implementation, finding that DNS can benefit from a publish-subscribe variant: next to limiting update traffic, it can considerably reduce the time it takes for a resolver to receive the latest version of a record, thereby supporting use cases such as load balancing in content distribution networks. The publish-subscribe architecture also brings new challenges to the DNS, including a higher overhead for endpoints due to additional state management, and increased query latencies on first lookup, due to session establishment latencies.