[G4] Broadband Speed Liar's Poker

[Timothy Pitts]

On 12/14/03 5:35 PM, "Robert A. Rosenberg" <hal9001 at panix.com> wrote:
> Even worse, given the way TCP/IP works, a 128Kbps Uplink may not be
> enough bandwidth to support a 4Mbps downlink. Each packet received
> must be acknowledged and due to the size of the packets, you can run
> out of available bandwidth at 128Kbps without playing games. Email is
> normally sent one line per packet (ie: an maximum of about 80
> charters per line/packet) and each line is supposed to get
> acknowledged (ACK) with a separate (I got it, send the next line)
> reply before the next one is sent. Thus you get a situation where you
> can not keep sending the "I got it" packets fast enough on a 128Kbs
> uplink to saturate the full 4Mbps downlink. There are ways to pack
> more than one line in a packet (so as to compensate for the
> packet-overhead) as well as keep the number of ACKs down by sending
> not sending one ACK per received packet but only periodic ones that
> say "I got all the packets up to Packet#x" (thus ACKing all the "in
> flight" packets at once instead of going Lock-Step
> (Send/ACK/Send/ACK/etc).
> 
> ----------

Most TCP/IP implementations use a sliding window protocol rather than the
"Stop and Wait" or "Alternating Bit" protocol that you describe above.

In a "Stop and Wait" protocol (ARQ is an example of this), the Utilization
(% of time actually sending data) results are less than extraordinary.

Luckily for both protocols the size of the ACK is much smaller than the size
of the packets you can receive.

The sliding window protocol allows the sender to send larger and larger
amounts of data up to a limit (the window max size) without receiving an
ACK. In this situation, you definitely have enough upload (128 Kbps) vs. dl
(4Mbps). The Utilization of your link speed also increases dramatically.

-- 
Timothy Pitts
tpitts at mac.com