[c-nsp] Congestion avoidance for unresponsive flows

[Aaron Daubman]

Greetings,

In all of the documentation I've been able to find, for congestion
avoidance Cisco routers use (W)RED  if configured and otherwise
default to tail-drop.

e.g. info at:
http://www.cisco.com/univercd/cc/td/doc/product/software/ios120/12cgcr/qos_c/qcpart3/qcconavd.htm

I'm wondering if there are any thoughts on how to better accommodate
unresponsive flows (UDP, RTP, etc...)?

A scenario might be similar to the following:
You have a router connected to an atypically high-delay and
periodically unavailable link (~1 second delay, occasional flap, down
for seconds to minutes).  This router will also be servicing a high
percentage of real-time traffic (video, voice...).  Is there any
better way to manage the queues for such traffic than tail-drop (which
would queue the first few seconds on outage and then start dropping
packets, such that upon link-restoration there would be a burst of old
packets followed by the current stream of packets) or WRED (which
would quickly turn into tail-drop with unresponsive flows upon
link-failure)?

It would seem that head-drop (Drop from Front) would be a better
mechanism in such cases (start dropping the oldest packets first such
that upon restoration the latest information will flow).  Is there any
way to implement this on Cisco gear?  Any other recommendations?

Also, are there any known implementations of queue-management
strategies that track packet age (in terms of how long the packet has
been queued at the router) and would prioritize the decision to drop
older packets as opposed to younger packets?

Thanks,
     ~Aaron