Question: What are the basic primitives in routers that would be useful for
pushback and/or local ACC?

Detecting the aggregate:

* Primitive: Mechanisms for random sampling the traffic.
  Why: to get a sample of packets for detecting aggregates.
  Status: Some routers have access to the packet stream for random sampling.
  Alternate primitives:  An alternate primitive would be filtered
    sampling of some fraction of the traffic, sent to an off-line box.

* Primitive: Drop sampling, or mechanisms for saving headers of dropped 
  packets.
  Why: to get a sample of drops for detecting aggregates.
  Status: Current routers generally can't do this.

Determining which input interfaces the aggregate comes in on:

* Primitive: Tagging packets with the ingress interface.
  Why: to determine which upstream routers should be sent pushback requests.
  Status: Many routers do this.

* Primitive: Access to routing tables.
  Why: to narrow the congestion signature to send in the upstream request.
  Status: This is easy.  This is in software.  Routers can do this.

Rate-limiting the aggregate:

* Primitive: Rate-limiting an aggregate before the output queue.
  Why: to carry out rate-limiting of the aggregate.
  Status: Many routers can do this now, in some form.

* Primitive: Rate-limiting two different aggregates, with two different rates.
  Why: to carry out rate-limiting for multiple independent aggregates.
  Status: Many routers can do this also.

* Primitive: Rate-limiting the aggregate at the input interface?
  Why: This could be useful at a border router where the pushback
  request could not be propagated upstream.  
  Status:

- Sally Floyd, March 2001
  (From the pushback team.)