[e2e] Transparent subnets with Proxy Arp

[Vernon Schryver]

> From: Bob Braden <braden at ISI.EDU>

> ...
> describe.  The mechanism is simple, but TCP's reaction to such
> perturabations are not.  See PILC.  (Of course, in 1989 we did not
> understand TCP performance as completely as we do in 2001).

I wonder if that wouldn't be better stated "understand what we
don't know about TCP performance as we do now."

Other than some advances in using brute force (e.g. much faster CPUs,
vastly fatter pipes, interrupt reductions, page flipping, and hardware
checksums) and some interesting and promising theory (e.g. RED and
ECN), what concrete (i.e. deployed) evidence of better understanding
is there?  I guess tweaks such as (officially) increasing the initial
window and fixing some TSP "featuers" in the BSD code that were uncovered
by early versions of HTTP servers would count, but only barely.

Part of my point is that 1989 was rather recent in the history of TCP
performance improvements.  What in the PILC documents was not pretty
well known in 1989, even if not written down or not known to such
people as those who would invent WAP?


Vernon Schryver    vjs at rhyolite.com