[7975] in www-talk@info.cern.ch
Re: Session tracking
daemon@ATHENA.MIT.EDU (Thomas Maslen)
Wed Apr 19 19:25:25 1995
Date: Wed, 19 Apr 1995 18:48:28 +0500
Errors-To: procmaster@www19.w3.org
Reply-To: tmaslen@verity.com
From: Thomas Maslen <tmaslen@verity.com>
To: Multiple recipients of list <www-talk@www10.w3.org>
> This can be implemented in about 500 lines of C code so it's not
> really as complex as it might sound.
This is not one of the more reassuring comments I've seen in WWW design
discussions.
We're not just implementing a client or a server; we're doing protocol
design. Protocol design is hard. (Well, good protocol design is hard).
The best example I know of this is a message that Van Jacobson sent (in
1988?) describing modifications to TCP to deal with congestion (slow-start
and congestion detection/backoff). It's a pretty long and dense E-mail
message. Almost all of the message is an analysis of the problem and a
solution; the code to implement it is something like two variables and half
a dozen lines of code (and, I think, no changes to the bits in the protocol
spec, just a change in the way that implementations use it). It's a small
change, but it makes all the difference to how the net runs (a small matter
of avoiding meltdown).
So... "it's only 500 lines of C" may provide a reasonable comfort level
for, say, a discussion about a whizzy new GUI feature, but when it's about
a protocol it's pretty scary.
Thomas Maslen
tmaslen@verity.com My opinions, not Verity's
