Re: [Dime] Mirja Kühlewind's Discuss on draft-ietf-dime-drmp-05: (with DISCUSS and COMMENT)

[ken carlberg <[e-mail]>]

Hi Mirja,

> On May 10, 2016, at 10:46 AM, Mirja Kuehlewind (IETF) <[e-mail]> wrote:
> 
> 
> I don’t think it is a good idea to assign a default priority to non-priority-defined requests at all. If you have high priority traffic that does not support this extension (yet) this traffic could be starved by lower priority traffic when assigning a middle range priority. I don’t think that is what you want to achieve.

I think your concern is quite valid, and my apologies if I missed something in the thread, but I think one thing that hasn’t been brought up is the notion of local policy and its separation from labels or priority values.  A number of years ago, the IETF community had some significant back and forth on the topic in the context of Emergency Telecommunications Service.  Specifically, in RFC-3689 we stated the following:

   3) Policy

      Policy MUST be kept separate from label(s).  This topic has
      generated a fair amount of debate, and so we provide additional
      guidance from the following:

      A set of labels may be defined as being related to each other.
      Characteristics (e.g., drop precedence) may also be attributed to
      these labels.  [11] is an example of a related set of labels based
      on a specific characteristic.

      However, the mechanisms used to achieve a stated characteristic
      MUST NOT be stated in the definition of a label.  Local policy
      determines mechanism(s) used to achieve or support a specific
      characteristic.  This allows for the possibility of different
      mechanisms to achieve the same stated characteristic.

      The interaction between unrelated labels MUST NOT be embedded
      within the definition of a label.  Local policy states the actions
      (if any) to be taken if unrelated labeled traffic merges at a
      node.

      Finally, labels may have additional characteristics added to them
      as a result of local policy.

Hence, the local policy associated with priority(s) *may* state that a weighted round robin in the servicing of queues *must* be used to ensure that no queue is ever starved.  Or, a more draconian position could be taken that does allow starvation.  It would be up to the operator/manager of the service to decide on the what is the local policy.  This is the way its been implemented and deployed in a few systems that I’m aware of that support prioritization.

-ken

(qualifier: just expressing my own personal opinion.  YMMV, consult doctor before use :-)