Caching static HTTP traffic in proxy-caches has reduced bandwidth consumption and download latency. However, web-caching performance is hard to increase further due to the growing number of non-cachable dynamic web-documents. Delta-encoding is a promising technique that exploits temporal correlation among different snapshots of a dynamic document, and renders dynamic traffic cachable. It achieves this by combining a cachable, previous snapshot of a document, called base-file, with small difference-file, called delta, to generate the current snapshot of the document. However, it has not yet been deployed due to the significant scalability concerns related to the storage requirements for base-files on the server-side.

In this paper we introduce class-based delta-encoding, a scalable scheme to perform delta-encoding on dynamic web-traffic. The idea is to group documents into classes, and store one document per class on the server-side. Thus, the proposed scheme exploits both temporal correlation in a dynamically evolving document, and spatial correlation among different documents. Finally, we present an architecture to deploy the scheme, that is transparent to clients, proxy-caches, and web-servers. Experimental results report that class-based delta-encoding combined with compression reduces the bandwidth consumption by a factor of 30, and the latency perceived by most users by a factor of 10 on average, without suffering from enormous storage requirements on the server-side.