Re:HTTP 2.0
by wmf on Tuesday July 09, 2013 @10:36AM
(#6012866)
Yes, the reason is that starting a new stream multiplexed inside an already-open TCP connection is faster than starting a new TCP connection. Attempts to improve TCP itself are slowed down by slow uptake of new Windows versions.
[ Parent
| on HN ]
While there have been some tweaks around the edges in terms of congestion control and the like, the TCP wire format has not changed in over 25 years and it has very little to do with Windows. It has to do with the all the middle boxes and routers in the ossified backbone of the internet that choke if they see anything over IP that is not exactly what they expect.
That's not true. HTTP/2.0 (or at least the successor to HTTP/1.1) has been discussed for over a decade, there have been many OS releases since then!
Inventing a whole new protocol to overcome a TCP deficiency is IMO a terrible motive...
Re:HTTP 2.0
by pritambaral on Tuesday July 09, 2013 @11:40AM
(#6013200)
These guys have something to say about that: http://blog.chromium.org/2013/06/experimenting-with-quic.htm...
A new protocol that runs on UDP precisely to side-step slow TCP changes.
That's great, but it means it'll always be a userland library rather than a OS supported networking protocol... which means the API will never be a defacto standard akin to the BSD sockets API. The reference implementation is also in obtuse looking C++... which will hinder bindings to other languages.
In short, don't expect to see this outside of Android or Chrome.
On the plus side, this is exactly what UDP was for in the first place.
What efforts are these? That sounds much more interesting than mucking around with HTTP micro-optimizations.
See TCP Fast Open, MPTCP, ObsTCP, TCPCrypt, CurveCP, QUIC, MinimaLT, etc.
I also get the impression that TLS SNI/Snap Start/False Start/Ludicrous Start were also hampered by browsers that used the Windows SSL stack, although I never understood why they would do that.