Re: cheap endpoints

Tom Anderson (tom@boris.cs.washington.edu)
Fri, 13 Aug 1999 17:44:11 -0700 (PDT)

yes, this is a problem, but I think the implication is much
more positive.

With the small change to TCP to add nonces, *any* single point
in the network that sees both forward and backward packets
can enforce TCP semantics on all flows through that point, even
if both endpoints are stupid/evil.

One of the points of the original DARPA proposal was to enable
stupid endpoints by moving intelligence into the network.
This work helps not hurts.

tom
-----
X-Authentication-Warning: sake.cs.washington.edu: cardwell owned process doing -bs
Date: Fri, 13 Aug 1999 13:38:46 -0700 (PDT)
From: Neal Cardwell <cardwell@cs.washington.edu>
To: syn@cs.washington.edu
Subject: cheap endpoints
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII

a discussion on the end-to-end list has brought up the point (that i think
we've tossed around) that the Internet is going to be at the mercy of the
random el-cheapo minimalist TCP/IP stacks in upcoming networked toasters.
hmm. a poor TCP implementation could be bad enough, but a
selfish/malicious one (a la our TCP Daytona) could easily cause real
problems.

---------- Forwarded message ----------
Date: Fri, 13 Aug 1999 14:26:03 -0400
From: Craig Metz <cmetz@inner.net>
To: Frank Kastenholz <kasten@argon.com>
Cc: end2end-interest@ISI.EDU
Subject: Re: why it's noteworthy (was RE: wee tee cee pee ...

In message <3.0.3.32.19990813135415.009a84d0@shultz.argon.com>, you write:
>At 10:28 AM 8/13/99 -0600, Vernon Schryver wrote:
>
>>A more interesting issue is why anyone cares? Today MByte EPROM, EEPROM,
>>NVRAM, and SRAM are readily available...
>
>because it opens up the possibilities of putting tcp/ip into places
>where it's never been before because the cost would be too high.

And because some mangler wants to save a few cents a part, we'll get stacks
that are too minimal and are ill behaved on the network. We'll pay for it in
the impact on our network infrastructure.

-Craig