The big_pipe used on the homepage is quite cool. The initial request just returns <script> blocks as each partial is rendered. I wonder how they work it on the

The big_pipe used on the homepage is quite cool. The initial request just returns <script> blocks as each partial is rendered. I wonder how they work it on the server side - they use PHP so they're not using any kind of threading. Perhaps the initial PHP request passes off most of the heavy lifting to their backend services over asynchronous Thrift calls. If that's the case then their PHP layer wouldn't be doing too much work, which doesn't seem to really tie in with them releasing HPHP.