On Mon, Jul 14, 2014 at 12:32 PM, Cory Benfield <cory@lukasa.co.uk> wrote: > I'd also love it if we could broaden our corpus of sample headers (as > used by the HTTP2JP > group's HPACK test cases) to reflect what you believe to be a more > reasonable sample of web traffic. Having that large corpus of data > will be useful for evaluating future compression proposals (assuming > there will be some). > Agreed, it would be. There are major privacy concerns baked into any collection of this kind, though, that I simply don't know how to reasonably solve. So instead, we're instrumenting the browser itself to aggregate the metric of interest, and definitely not collecting any raw headers. The core metric of interest seems to be the compression performance (aggregated byte-length ratio) of a delta encoder + Huffman. We can provide separate ratios for an encoder with and without the reference set. This should hopefully inform the question before us, but is of little utility for future experimentation with compressors. That's unfortunate, but I think it's just the trade-off for getting good coverage.
Received on Monday, 14 July 2014 20:19:31 UTC