IMHO, SPDY should be implemented on the frontend servers like varnish, HAProxy or nginx. If you do SPDY on the backend server, you cause a lot of additional problems:
1) your backend servers should use their hardware for actually running the web application, not for encrypting responses.
2) usually, your backend servers only serve the dynamically generated content. SPDY's biggest advantage is that it would allow to handle multiple requests in one connection, but this really shines for static assets (images, stylesheets) which in turn are rarely served by the backend.
3) Even if you want to do SPDY on the backend, your frontend proxies still have to support it, because that's what the clients are talking to.
So as it stands now, having SPDY on the backend server (or even in your application like this project here) doesn't really give you any of SPDY's advantages unless you begin serving all of your site directly from your application servers, which we learned over the last years, just doesn't scale as well.
The author here. I can understand that it's not clear from the README, but I only chose a Rack implementation for the backend in order to make developing those 'Adapters' easier for others, since Rack is pretty standard in the Ruby world.
One of the implemented adapters, Proxy, acts a proxy to another HTTP server.
In the performance section:
"[The results] show the power that lies in the simple addition of a SPDY server to an existing nginx/Thin configuration (and probably also Unicorn/nginx.)"
I am totally for the separation of front- and backend.
Yes, SPDY is a cross-platform protocol. This submission is a Ruby implementation of a SPDY server for Rack applications like Ruby on Rails. The news.ycombinator.com server is written in the arc language. SPDY is awesome.
Ok. It's my fault for not being clear in my original message. I wasn't talking about the linked SPDY implementation but about SPDY in general. Sorry, I should have marked it as OT.
No, HN is currently using the regular HTTP protocol. In my opinion, the costs of implementing SPDY support, at the moment, outweigh the benefits of that support.
I don't know why, but I find HN pretty slow for a page that only shows 30 links. Engadget takes about half the time to load. Granted this is with Adblock and WidgetBlock, which I use to block all 3rd party connections besides Disqus, but HN doesn't use any of that anyway, so it's still strange.
HN does no output buffering, causing rendering time to often be closer to 5 seconds than 100ms in Europe. This is trivial to fix, but I guess the SFO crowd don't feel the pain so it probably never will.
Edit: nope, that's not it. Using the fastest UK machine with clean net available to me, tcpdump never shows more than 1 unacked packet on the wire at any time, gzip is turned off, and keepalive is turned off. Gzipping the 24k home page drops it to 4.3k.
None of this really accounts for the crazy jitter we see in Europe while rendering the main table.
1) your backend servers should use their hardware for actually running the web application, not for encrypting responses.
2) usually, your backend servers only serve the dynamically generated content. SPDY's biggest advantage is that it would allow to handle multiple requests in one connection, but this really shines for static assets (images, stylesheets) which in turn are rarely served by the backend.
3) Even if you want to do SPDY on the backend, your frontend proxies still have to support it, because that's what the clients are talking to.
So as it stands now, having SPDY on the backend server (or even in your application like this project here) doesn't really give you any of SPDY's advantages unless you begin serving all of your site directly from your application servers, which we learned over the last years, just doesn't scale as well.