Add binary exchange RFC 0009

[dan13bauer]

No description provided.

[dan13bauer]

@aditi-pandit @majetideepak This is the RFC for the binary exchange protocol. Can you please take a look before I move it out of the draft state? Thank you!

[tdcmeehan]

Does BinX provide security and does it encrypt the exchange?

[dan13bauer]

Does BinX provide security and does it encrypt the exchange?

It supports TLS encryption. Whenever SSL is configured for the HTTP server (http-server.https.enabled), it uses the same configuration and establishes TLS connections for the exchange.

[tdcmeehan]

I understand why, in theory, using TCP sockets should be faster than HTTP. But can you go into more depth on where we are seeing the speedup from: is it the lack of headers on the page data? How the data is chunked in HTTP?

[dan13bauer]

I understand why, in theory, using TCP sockets should be faster than HTTP. But can you go into more depth on where we are seeing the speedup from: is it the lack of headers on the page data? How the data is chunked in HTTP?

I don't have a good answer to that. The initial starting point was an observation that the CPU load of the HTTP stack is relatively high. @yingsu00, you have observed that recently, too, right? Our assumption was that a simpler stack should be able to provide some improvement but we didn't conduct a detail analysis yet.

[tdcmeehan]

If we haven't gotten to the bottom of that, IMO it would be worthwhile to do that, because I'm wondering if a simpler change could be done by optimizing how we are using HTTP. We don't know that until we root cause where in the HTTP stack we're spending the cycles.

[majetideepak]

Some nits and questions.
Like Tim, I am also curious to know where the wins are coming from given both Wangle and Proxygen use Folly under the hood.