If you are interested in testing Brotli on the web the CloudFlare test server for HTTP/2 and IPv6 also does Brotli compression.<p><a href="https://http2.cloudflare.com/" rel="nofollow">https://http2.cloudflare.com/</a><p>And the NGINX module that does it has been open sourced: <a href="https://github.com/cloudflare/ngx_brotli_module" rel="nofollow">https://github.com/cloudflare/ngx_brotli_module</a>
Random note about accept-encoding that might have been corrupted by my memory: at some point in the past wikipedia was blocking Googlebot due to it hammering their servers. Some debugging revealed that Googlebot's accept-encoding header was different than browsers and so it was missing the wikipedia cache. (Wikipedia heavily relies (relied?) upon a caching layer because their pages can take like two full seconds of CPU time to render otherwise.)<p>The kinda awful fix was to add support for the compression variant nobody uses (deflate?) to Googlebot so that its accept-encoding header matched the common one.<p>It would be ironic if adding this header to Firefox wrecked their page load performance for wikipedia!
Wondering if this will fare better than gzip for my use case of compressing a Windows 95 disk image (<a href="http://win95.ajf.me/" rel="nofollow">http://win95.ajf.me/</a>). Will have to try it out.
I'm not a compression expert at all, but I've read about ANS a few times lately, and I wonder whether an ANS variant of Brotli might be an even better choice.