[Cryptography] TLS/DTLS Use Cases

[Nico Williams]

On Tue, Apr 08, 2014 at 12:12:54PM -0700, Bear wrote:
> On Sat, 2014-04-05 at 18:23 -0500, Nico Williams wrote:
> > Also, HTTP is just about the worst datagram protocol ever.  There's no
> > XID, so responses have to be sent in the same order as requests over
> > any one keptalive TCP connection.  Yuck.  (When I've brought this up
> > in the context of HTTPbis I've been told to go away.)
> 
> To be fair, keep-alive was not part of the design.  Http was initially 
> a completely stateless protocol, and actually a fairly well designed 
> one.  The reason keep-alive is not well supported is because it's got 
> nothing to do with the original design and was bolted on as an
> afterthought. 

It was added as a new minor version of the protocol.  That would tend to
indicate (to me anyways) that it wasn't an afterthought.

> Is there a take-home lesson there?  Only that if we engage in elegant 
> design we should not trust those who come after us not to screw it up.

Are you saying that HTTP/1.0 was elegant?  Well, I suppose it was, if we
ignore all the complexity of text, line-oriented headers (the two ways
to express multiple header values, continuation lines, verbosity).  The
elegant part is the REST/CRUD aspects, IMO.

Anyways, the part that interests me here is that there's still no
interest in fixing this, particularly when HTTP/2.0 is so much about
performance.  Perhaps the lesson is that we don't learn from our
lessons.

(I haven't checked recently, so it's possible that this has been
addressed since.  I sure hope so!)

Nico
--