Re: [Spud] Fwd: New Version Notification for draft-herbert-transports-over-udp-00.txt

[Joe Touch <[e-mail]>]

On 5/19/2016 5:20 PM, Toerless Eckert wrote:
> On Thu, May 19, 2016 at 04:55:56PM -0700, Joe Touch wrote:
>> On 5/19/2016 4:48 PM, Toerless Eckert wrote:
>>>> Please don't *#*!-up the Internet to fix an OS implementation issue.
>>> Still haven't made the point why this is *#*!-up the Internet.
>> I did - in summary: you're adding a new protocol that we don't otherwise
>> need, can't seamlessly transition to kernel-based transports without
>> requiring the additional encapsulation overhead, and only end up making
>> DPI look one more header in.
> - Ideally, you would just design optimized new transports that use
>   UDP as demux layer. Think about it. 

That's why I said you were reinventing the Internet at the UDP layer. 
If you expect the end game is that we keep UDP as an additional demuxing
layer, I think this is not just a bad idea, it's a VERY bad idea.

> - IMHO there are only minute benefits of in-kernel transport protocol. I'd be happy
>   to have them die out. OS vendors can happily implement TCP wih UDP ports as
>   a library. 

Some do. I recall that there were ways to update old versions of Windows
with new transport stacks (Trumpet, I think?).

> That immediately makes it part of my docker container and gives
>   me predictable behavior when i distribute my container.
> - No deeper look with DPI needed.

I agree. So go fix that, IMO, and leave the Internet alone.

Joe