[e2e] What should e2e protocols know about lower layers?

Vernon Schryver vjs at calcite.rhyolite.com
Thu Oct 11 14:56:29 PDT 2001


> From: "Christian Huitema" <huitema at windows.microsoft.com>

> ...
> It should be pretty obvious that suppressing congestion control
> based on a naïve test such as "same subnet mask" is BAD. OTOH,
> developers do this for a reason: they are generally frustrated that
> TCP does not take full advantage of the capacity of a high speed
> network, especially in the case of short connections. Make no
> mistake, the pressure is going to increase with applications such
> as ISCSI and gigabit networks.

Yes.
There is a difference between the two proposals, "turn off TCP checksums
if local" and "turn off slow start if local," that I meant to point
out.  The first only hurts the applications or hosts that choose to
do it, but the second hurts innocents when "local" isn't really.
Thus, TCP code that is observed not doing slow start in public must
be treated the same, whether it's dumb junk that can't do slow
start at all or too smart by half junk that thinks all the world
is a bunch of CSMA/CD LANs (e.g. with broadcast) connected by Ciscos.

It's not that I don't sympathize with the major speed-up sending
a blast of MBytes of segments in response to the SYN-ACK on fast
local media, but that it is important to keep one's perversions really
private and not just presume, assume, and pretend.


Vernon Schryver    vjs at rhyolite.com



More information about the end2end-interest mailing list