* Gateways can be smarter than they are. Gateways have access to
congestion data which is not available to lower level clients.
* Early Random Drop - if we are exceeding the queue size, then drop arriving
packets based on a fixed drop probability. Targets aggressive users.
Arguably fair since each arriving packet has the same chance of being
dropped.
* IP Source Quench - sets bit in header if above queue size. Source checks
for bit and if it's prevalent, reduces window size exponentially.
* RED - min/max thresholds. If below min, no packets dropped. If above max,
every packet is marked. Inbetween, packets are marked based on average
queue size. Chances of a connections packet being marked is proportional to
the it's share of bandwidth.
* When RED drops packets above max, it adequately controls average queue
size. When it simply marks the packets, it relies on clients to control
queue size. Further, it takes a whole roundtrip for each connection to see
a reduction in queue size.
* RED does well in indentifying misbehaving users, but so does Early Random
Drop. RED must attain better performance by controlling source output.