Re: my broken TCP is faster on broken networks

Albert D. Cahalan (acahalan@cs.uml.edu)
Fri, 11 Sep 1998 17:18:49 -0400 (EDT)


> 2 minute delays are fine in bulk traffic. You whole thinking is broken
> however.
>
> If n users * x packets with 2 second rtt simply dont fit down the pipe in
> question you lose.
>
> Chunks of the net are congested. That congestion however you will
> note doesnt cause a congestion explosion

You are completely ignoring several important things.

1. It's not all TCP/IP. There are many protocols without backoff.
If the TCP/IP connections do the backoff, other protocols will
have an advantage and will consume all bandwidth. TCP/IP won't
get a fair share by being polite.

2. The exponential backoff can not be enforced, since users can
restart the connection. That is exactly what happens.

3. The last time I heard, http was the #1 protocol. Since the
connections are short-lived, backoff works poorly anyway.

Look, this is dumb:

1. user clicks on a hyperlink
2. backoff takes effect
3. user clicks "Stop" and tries the link again, sending MORE packets
4. congestion explosion, as every user does the above many times

That is what really happens. Maybe you'd like to get rid of that
damn human. Humans mess up the math. Proofs that don't consider
human behavior are WRONG.

It may seem evil to reduce the backoff. It is far worse to have
behavior that makes those awful humans restart the connection.
I've not seen anyone really address real-world human behavior.
Networks do not exist all alone without humans.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/faq.html