RE: RE: Inconsistencies in QDDIM -07 Draft

Let's just put a stake in the ground and say that you must have the
CalculationService if you have a REDDropper (Bob's 1..n cardinality).
Specificity is clearer and simpler than having multiple degrees of freedom.

Andea
  -----Original Message-----
  From: [email protected] [mailto:[email protected]]On Behalf Of
Andrew Smith
  Sent: Wednesday, May 29, 2002 1:55 PM
  To: [email protected]; Andrea Westerinen
  Cc: Policy@Ietf. Org
  Subject: RE: [Policy] RE: Inconsistencies in QDDIM -07 Draft


  Bob, Andrew,

  Rather than inventing a special case for this, why not just use a
CalculationService instance that does a very simple calculation? Or is this
issue really just because people feel a need to be backwards-compatible with
an earlier draft?

  Andrew Smith
    -----Original Message-----
    From: [email protected] [mailto:[email protected]]On Behalf Of
[email protected]
    Sent: Wednesday, May 29, 2002 12:49 PM
    To: Andrea Westerinen
    Cc: Policy@Ietf. Org; [email protected]
    Subject: RE: [Policy] RE: Inconsistencies in QDDIM -07 Draft


    Andrea,

    Just to be clear: you're asking whether to allow room in the model for a
"watches" linkage between a RED dropper and a queue that does not traverse a
CalculationService instance. As best I can figure out, in such a case:

    - only one queue could be watched;
    - the watched queue would have to be the queue from which packets are
dropped;
    - there would be no smoothing of the current queue depth for the watched
queue, since this is what the CalculationService does.

    I'm not saying yes or no - just asking whether this is in fact what
you're proposing.

    Regards,
    Bob

    Bob Moore
    Advanced Design and Technology
    Application Integration Middleware Division
    IBM Software Group
    +1-919-254-4436
    [email protected]

    "Andrea Westerinen" <[email protected]>





                  "Andrea Westerinen" <[email protected]>
                  Sent by: [email protected]
                  05/29/02 02:17 PM


          To: Robert Moore/Raleigh/IBM@IBMUS
          cc: "Policy@Ietf. Org" <[email protected]>
          Subject: RE: [Policy] RE: Inconsistencies in QDDIM -07 Draft



    In the original incarnations of QDDIM, we had NextService only, for a
REDDropper. The problem was that the Dropper did not have to base its
calculations only on the queue from which it dropped. So, CalculationService
was defined to describe the calculations separate from the queue to drop. We
did not mandate the existence of the calculation definitions before, so I
did not want to do that now. However, I am not religious about this one -
mandating the association (based on cardinality) does force specificity.

    Do others have an opinion on this?
    Andrea
        -----Original Message-----
        From: [email protected] [mailto:[email protected]]On Behalf
Of [email protected]
        Sent: Wednesday, May 29, 2002 11:02 AM
        To: Andrea Westerinen
        Cc: Policy@Ietf. Org
        Subject: [Policy] RE: Inconsistencies in QDDIM -07 Draft

        OK, but then how does a RED dropper get to a queue to watch if it
doesn't go via a CalculationService? I didn't think we were using
NextService for this.

        Regards,
        Bob

        Bob Moore
        Advanced Design and Technology
        Application Integration Middleware Division
        IBM Software Group
        +1-919-254-4436
        [email protected]

        "Andrea Westerinen" <[email protected]>


                              "Andrea Westerinen" <[email protected]>
                              05/29/02 01:34 PM


              To: Robert Moore/Raleigh/IBM@IBMUS
              cc: "Policy@Ietf. Org" <[email protected]>
              Subject: RE: Inconsistencies in QDDIM -07 Draft


        Bob, I think that the operative words in my email are "IF you define
a CalculationService, then ..." Since CalculationService is a new concept
and may not be visibly instrumented, I did not want to make it mandatory.
Hence, the 0..n on the CalculationService, as related to a REDDropper. I
think that the "one or more" wording in the Description is a cut and paste
error.

        Andrea
                -----Original Message-----
                From: [email protected] [mailto:[email protected]]
                Sent: Wednesday, May 29, 2002 10:36 AM
                To: Andrea Westerinen
                Cc: Policy@Ietf. Org
                Subject: Re: Inconsistencies in QDDIM -07 Draft

                Hi Andrea,

                Busy editing QDDIM ....

                From CR 795:

                //
==================================================================
                // CalculationServiceForDropper
                //
==================================================================
                [Association, Experimental, Version ("2.7.0"),
                Description (
                "This association is a subclass of ServiceServiceDependency,
"
                "and represents the reliance of a REDDropperService on one
or "
                "more DropThresholdCalculationServices. The latter calculate
"
                "average queue depth, based on the observed depths of a "
                "queue. The specific queue examined by each
CalculationService "
                "is defined using the CalculationBasedOnQueue
association.") ]
                class CIM_CalculationServiceForDropper :
CIM_ServiceServiceDependency {
                [Override ("Antecedent"), Description (
                "A calculation service for the dropper.") ]
                CIM_DropThresholdCalculationService REF Antecedent;
                [Override ("Dependent"), Description (
                "The RED dropper which is dependent on average queue depth "
                "calculations by the Antecedent Service.") ]
                CIM_REDDropperService REF Dependent;
                };

                For the calculation service, the words say "one or more",
but the cardinality says 0..n. You also propose 0..n for QDDIM, but I think
that the words are correct here - the cardinality should be 1..n, since the
only path to get from a RED dropper to a queue for it to examine is via a
calculation service instance.

                Assuming that you'll say "Yes, of course," I've tentatively
gone with 1..n in the QDDIM update. You can still talk me out of this is
you're quick.

                Otherwise, I've edited QDDIM as you've proposed in this
note.

                Regards,
                Bob

                Bob Moore
                Advanced Design and Technology
                Application Integration Middleware Division
                IBM Software Group
                +1-919-254-4436
                [email protected]

                "Andrea Westerinen" <[email protected]>

                                "Andrea Westerinen" <[email protected]>
                                05/14/02 06:28 PM


                      To: "Policy@Ietf. Org" <[email protected]>, Robert
Moore/Raleigh/IBM@IBMUS
                      cc:
                      Subject: Inconsistencies in QDDIM -07 Draft

                In reading QDDIM very closely :-), I noticed some
inconsistencies in the text in Section 3, in the formal class definitions in
Section 4, and in the intent. Here are my observations and recommendations
...

                1. For RED droppers, I think that we wanted the following
...
                A REDDropperService can be related to many
DropThresholdCalculationServices (many to many), but each CalculationService
examines a single queue (many to one).

                2. This means that ...
                IF you define a DropThresholdCalculationService, you really
MUST specify the queue that is examined. So, CalculationBasedOnQueue should
have a cardinality of 1..1 on the QueuingService side (currently it has
0..1).
                AND
                A REDDropper MAY be associated with one or more
CalculationServices (each looking at a specific queue). So,
CalculationServiceForDropper should have a cardinality of 0..n on the
DropThresholdCalculationService side (currently it is a mandatory 1).

                3. For a HeadTailDropperService, its determinations can also
be based on several queues (as discussed at the bottom of page 23). So,
HeadTailDropQueueBinding should have a cardinality of 1..n for the
QueuingService (currently, it has mandatory 1). I agree that at least queue
is mandatory - but more than one can be examined.

                Andrea