Exploring HTTP/2 Header Compression

Abstract

For every HTTP/1.1 request and response, almost the same set of headers is transferred. This wastes bandwidth, the most expensive resource of the browser-server communication. To solve this issue, header compression for HTTP/2 was standardized. During the standardization, we found that one element of the compression technology, the so called reference set, contributes little to the compression ratio while its mechanism is complicated. With our proposal, the reference set was removed and the specification and implementations were drastically simplified. For high performance implementation of header compression, we devised token-based reverse indices, length guessing for Huffman encoding, and pre-calculated state transitions for Huffman decoding.