[Cryptography] How to De-Bollocks Cryptography?

[iang]

On 13/08/2024 07:02, Phillip Hallam-Baker wrote:
>
> I don't think TLSsimple is what we need.


I think a huge mistake that is made is that people think they want a 
connection-oriented protocol.  Almost all applications are datagrams - 
discrete packets of some length that are more or less independent of any 
other packets.

Obviously, web is datagrams - each request, each page, each recursive 
insert, etc. Even live voice/video is datagrams as the robust protocols 
are lossy, and the normal arrangement is a subset of all the packets. 
Songs and movies are just big datagrams, so write a jumbo-datagram 
protocol over small datagrams, it's a month's work by a good intern.

What happens however is that people assume TCP (or TLS) for 
"reliability" and don't get it. So they are forced down the path of 
adding a reliable datagram protocol over the top of an unreliable 
connection protocol. The end result is a mess as two complex layers 
fight each other.

Nicely, nowadays, we can say - look at SPDY. (It almost exactly mirrored 
the work I'd done & aired on this group a decade back called SDP1. And 
has the advantage of brand and a big adoption win.

iang


ps; an old datagram on bad connections 
https://iang.org/ssl/reliable_connections_are_not.html
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.metzdowd.com/pipermail/cryptography/attachments/20240814/0fee5fc1/attachment.htm>