Re: [CDNi] Request for WG Adoption: Capacity Insights

[Andrew Ryan <[e-mail]>]

Hi Kevin,

   I wanted to take a moment before diving in to thank you for the 
fantastic feedback and engagement!  I am excited to have more and more 
folks involved with this.  As an FYI I will be unable to make the IETF 
CDNI session coming up, so I am hoping to drive a lot of progress on 
this through the email list.

Please see answers inline

On 7/9/2022 10:37 AM, Kevin Ma wrote:
> Hi Andrew,
>
>   Thanks for the responses.  Some additional comments:
>
> > > - What is the scope of uniqueness for the id? If it needs to be 
> global, then it needs a registry.  If it needs to be dCDN unique, then 
> it is just on the dCDN to ensure?
> > @Ben keep me honest here, but the intention is that the ID be unique 
> within the scope of the uCDN and dCDN, i.e. the namespace is specific 
> to this CDNi delegation relationship.In this case, it would be up to 
> the dCDN to ensure uniqueness in the payload that it is returning to 
> the uCDN.
>
> This could probably just be clarified and made more explicit in the text.

ack - I agree we can clarify this


>
> > > - Is there a valid range for time-granularity, data-percentile, 
> and/or latency, e.g., 0-31622400, 0-100, and/or 0-2678400, 
> respectively?  And what happens if values are out of range?
> > In terms of valid ranges, I think the data-percentile would have a 
> natural range of 0-100.  The others, I believe would not necessarily 
> have a bound (outside of integer limits).  The usefulness of very 
> large time-granularity and latency values somewhat diminish the 
> usefulness of this data for near real time traffic steering decisions.
>
> I guess if you are only specifying a transport encoding, and not 
> providing any guidance on how to interpret a given value, I guess I 
> can see how you would want to limit restrictions on those fields.  
> Percentages outside of 0-100 are obvious ones, but I also don't know 
> what a negative latency or negative time-granularity means, and a 
> latency/time-granularity beyond a certain point (1yr?) is pretty 
> meaningless.  My concern is that, from an interoperability standpoint, 
> if the interpretation is different, then it could cause problems.  If 
> the interpretation is going to be defined in a different spec or 
> customized by individual CDNs, then we could make a stronger statement 
> about that, but it feels like we could still be a little bit tighter 
> on the bounds.

You raise a very valid point in that at a bare minimum, we should be 
restricting to non negative values, as you point out, there is no 
situation that I can think of where an advertisement of negative 
capacity makes any sense.  Do you feel that clarifying the range to be 
positive integers and leaving the limits "unbounded" would be sufficient 
here?


>
> > > - What happens if there are duplicate names in a source object?
> > I think the intention behind declaring that the name needs to be 
> unique circumvented any discussion on how to handle duplicate names
>
> With respect to error handling, I think it might be worth going a step 
> beyond just specifying what a correctly formatted object looks like 
> and provide some guidance on how to handle foreseeable error 
> conditions (e.g., non-unique or duplicated ids, out-of-range values, 
> etc.).  I don't think we've done this as much in the past, but the 
> previous capabilities were very binary (i.e., they contain a list of 
> tokens and either you understand the token or not); but the complexity 
> of information in the telemetry objects feels to me like there is more 
> room for ambiguity and so could benefit from more discussion of how to 
> handle unexpected data.  It may be as simple as: If the data doesn't 
> make sense, ignore/discard it.  I could be off here, but that's my gut 
> feel.

I agree that specifying what to do if the payload doesn't make sense is 
a valuable addition.  I would propose that we state that invalid 
payloads can be ignored and even state that any previously valid 
payloads can be honored until such a time as a new valid payload can be 
acquired.  This will prevent situations where uCDNs who fail to get a 
valid payload from a dCDN don't stop all delegation due to a lack of 
updated limits, the uCDN will continue to use previously negotiated 
limits.  Using stale information might be considered a risk if the dCDN 
is truly having an outage, but in this case, this payload is more so 
about capacity guidance rather than any guarantee of service 
reliability, so I feel that the "use stale" case is probably more 
productive.

>
> > > - Should there be a registry for telemetry source types?  Will 
> these be globally unique and well specified (in an RFC?) for 
> interoperability?
> > @Ben I do think we will want a registry for telemetry source objects.
> > > - Should there be a registry for capacity limit types?  Will these 
> be globally unique and well specified (in an RFC?) for interoperability?
> > I do believe this would make sense
>
> I agree that these should be registries.
>
ack.  We may need some guidance on how to best define this.

> > > - I think we also need a privacy section that discusses the 
> potential issues of divulging internal information about the CDN.
> > part of the data requirements are such that the data for capacity is 
> scoped specifically to the uCDN,->dCDN delegation relationship..i.e. 
> the data is very narrowly scoped.  The data being presented is also 
> data about the dCDN by the dCDN.  Would there be any suggestions on 
> how we might want to frame a statement for this?
>
> Though we rely on the underlying FCI protocol to handle auth and 
> encryption (and the rest is out of the hands of the FCI object 
> designer), I think it is worth discussing what happens if there is a 
> breach, if there are specific consequences of specific data loss; or a 
> statement that the data is so generic that it doesn't matter.  I think 
> the question is: is there anything a competitor/adversary could gain 
> from this data?  Since it only describes the source and doesn't 
> contain the actual telemetry, the value of units and limits is minimal 
> without hacking the actual source stream? If so, a statement to that 
> effect would probably suffice.

In terms of the FCI.CapacityLimits payload, part of the design of the 
limits provided are what the dCDN is willing to offer to a specific uCDN 
for capacity.  If we look at this another way, it's not a full and 
complete mapping of all available capacity a dCDN may have.  We decided 
on this approach for several reasons, but at least one potential benefit 
of this is that it would limit the exposure to a dCDN of competitors 
being able to map out a full capacity footprint of the dCDN.


In terms of someone gaining access to Telemetry endpoints which are 
referenced in FCI.Telemetry, this could be considered a issue for both 
the uCDN and the dCDN as this exposes actual usage information which is 
problematic for many reasons.  This particular data set though is 
provided outside the scope of the FCI payload, and relies on external 
APIs/datrasources/etc..

Would the above two statements be useful if included into the document?

>
> > > - This is subdividing a footprint?  Would it make more sense (or 
> at least be simpler) to just define more granular footprints.  It 
> seems confusing to make essentially another way specify footprints.
> > This was a very lengthy part of our discussion in this draft.  There 
> are use cases where "all traffic is not created equal" i.e Low Latency 
> HLS will have much different request characteristics than Bulk game 
> downloads or even Video On Demand.  The CDNi Footprint object doesn't 
> allow for granularity on a published host basis.  We were working 
> under the fundamental premise that we were trying to cause as little 
> disruption to the fundamental components of CDNi and trying to 
> encapsulate as much as we could related to capacity management without 
> major overhauls to the original spec.  We would greatly welcome 
> discussion on this topic.
>
> I think this is the big one, and I too would welcome some discussion 
> on the topic.  My initial reaction is: maybe we should try to define a 
> new footprint type, rather than create a sub-footprint for just this 
> one object; it sets a bad precedent.  But, it needs more thought.

:-) this one may open up a large can of worms to be sure.  Part of our 
thought process with this draft was to keep it as isolated as we could 
to help with the adoption process, but perhaps that was not the right 
approach to take.

As we mentioned, the purpose of providing the "scope" within the 
FCI.CapacityLimits object was to allow finer granularity within a 
Footprint, such that if an uCDN is delegating lots of different types of 
traffic to a dCDN, that the dCDN had a way to differentiate capacity 
advertisements for identifiers which might correlate to the type of 
delivery.  Another way of saying that would be that in a lot of cases, 
uCDNs would have different published hosts for different forms of video 
delivery (live.example.com, vod.example.com, etc..). The dCDN would be 
aware of these published hosts due to the configuration metadata 
associated with it, and would be able to be aware of "traffic type" and 
could decide to offer different rps or other such limits based on that.

Published host is not the only differentiating factor though, different 
delivery types could be handled on the same domain, but with different 
paths (www.example.com/live, www.example.com/vod). In this case, using a 
published host in a scope within the FCI.CapacityLimits object would not 
be sufficient, but we did attempt to incorporate the ability to 
reference service-id or property-id which maps into groupings of 
configurations that the dCDN and uCDN are aware of.

The summary of the issue is that we have a potential gap in the 
definition of footprint, the specificity that is provided by the 
configuration spec, and the potential use case for allowing capacity 
payloads to share the same specificity as the configuration spec.

A counter point to providing this level of granularity is that in 
practice, it may be very cumbersome for dCDNs to provide very granular 
advertisements to a uCDN, and may simply opt to provide more of a 
blanket capacity advertisement, which the uCDN is free to use as it will 
(I feel like this will be the case for the most part).  There may be 
instances though where a dCDN does want to get very granular though, 
which is why we attempted to bring this functionality into the payload.

I am wondering if it make sense to try and schedule a breakout session 
to discuss this further, esspecially with several of the folks who have 
been heavily involved with new draft for footprints.



>
> thanx!
>
> --  Kevin J. Ma
>
> On Fri, Jul 8, 2022 at 12:30 PM Andrew Ryan <[e-mail]> 
> wrote:
>
>
>     On 7/4/2022 12:35 PM, Kevin Ma wrote:
>>     Hi Andrew,
>>
>>       (As a chair) Given the reduced scope, i.e., just two FCI
>>     objects (though see my comments below), I think adoption would be
>>     within scope of the current charter if other folks feel this is
>>     good work for the WG to take on.  I don't have a problem with
>>     issuing a formal call for adoption.  I've included some comments
>>     on the current draft below.
>>
>>     thanx.
>>
>>     --  Kevin J. Ma
>>
>>     section 1.3:
>>       - In the new paragraph about the uCDN calling a dCDN API
>>     interface, what interface is that?  Is that an interface outside
>>     of CDNI.  It's not clear if that relates to anything in this
>>     particular draft?  It would be good to clarify this.
>
>
>     The intention here was that the uCDN would interact with an
>     interface which provides FCI payloads.  To remove ambiguity I can
>     remove the the reference all together:
>
>     Original:
>
>     In normal operation a uCDN will communicate with a dCDN, via an
>     interface, to collect and understand any limits that a dCDN has
>     set forth for traffic delegation from a uCDN
>
>     Proposed:
>
>     In normal operation a uCDN will communicate with a dCDN to collect
>     and understand any limits that a dCDN has set forth for traffic
>     delegation from a uCDN
>
>
>>
>>     section 2.1.1:
>>       - What is the scope of uniqueness for the id?  If it needs to
>>     be global, then it needs a registry.  If it needs to be dCDN
>>     unique, then it is just on the dCDN to ensure?
>
>     @Ben keep me honest here, but the intention is that the ID be
>     unique within the scope of the uCDN and dCDN, i.e. the namespace
>     is specific to this CDNi delegation relationship.In this case, it
>     would be up to the dCDN to ensure uniqueness in the payload that
>     it is returning to the uCDN.
>
>>       - the configuration descriptions references a yet to be defined
>>     Telemetry Interface?  Is that in the scope of this document?  If
>>     not, it would be good to clarify (or remove) this.
>
>     The Telemetry interface is NOT in scope of this document, we can
>     add clarifying language to explicitly state this. In Section
>     2.1.1.1 an additonal sentence can be added:
>     Proposed:  The definition of a CDNI Telemetry interface is outside
>     the scope of this document.
>
>
>>
>>     section 2.1.1.1 <http://2.1.1.1>:
>>       - Should there be a registry for telemetry source types?  Will
>>     these be globally unique and well specified (in an RFC?) for
>>     interoperability?
>
>     @Ben I do think we will want a registry for telemetry source
>     objects.  The declaration of a "generic" type was meant as a place
>     holder until a formal Telemetry Interface was defined.  The
>     Generic source type also allows CDNs which already have an
>     existing relationship and have already established data channels
>     for providing and consuming Telemetry to be used "as is" without
>     requiring additional integration steps.
>
>
>>
>>     section 2.1.1.2 <http://2.1.1.2>:
>>       - What happens if there are duplicate names in a source object?
>
>     I think the intention behind declaring that the name needs to be
>     unique circumvented any discussion on how to handle duplicate names
>
>
>>       - Is there a valid range for time-granularity, data-percentile,
>>     and/or latency, e.g., 0-31622400, 0-100, and/or 0-2678400,
>>     respectively?  And what happens if values are out of range?
>
>     The intention is that these are meant to provide guidance to the
>     uCDN in terms of how accurate and how far behind real time the
>     data set is, to help inform how to react from a traffic steering
>     perspective.
>
>     In terms of valid ranges, I think the data-percentile would have a
>     natural range of 0-100.  The others, I believe would not
>     necessarily have a bound (outside of integer limits).  The
>     usefulness of very large time-granularity and latency values
>     somewhat diminish the usefulness of this data for near real time
>     traffic steering decisions.
>
>
>
>>
>>     section 2.2.1:
>>       - "CAN" is not an RFC2119 term.  "CAN" -> "MAY"?
>
>     Understood, we will work to change this
>
>
>>       - What is the scope of uniqueness for the id?  If it needs to
>>     be global, then it needs a registry.  If it needs to be dCDN
>>     unique, then it is just on the dCDN to ensure?
>
>     This should be unique between the uCDN and dCDN
>
>
>>       - Are there limits related to maximum-hard, maximum-soft, and
>>     current?  Should those be specified along with the capacity limit
>>     types?  And what happens if values are out of range?
>
>     I am not sure we want to introduce an artificial cap outside of
>     the definition of the Integer itself as this will allow for the
>     most flexibility and hopefully prevent having to revisit the draft
>     to increment values in the future.
>
>
>>       - section 2.2.1 already defines a "Telemetry Source Object";
>>     can we come up with a different name to make it less confusing?
>
>     @ben ^^
>
>
>>
>>     section 2.2.1.1 <http://2.2.1.1>:
>>       - Should there be a registry for capacity limit types?  Will
>>     these be globally unique and well specified (in an RFC?) for
>>     interoperability?
>     I do believe this would make sense and part of the intention of
>     defining it in this draft was an attempt to address that.  If
>     there is a more appropriate way to accomplish this, I would
>     welcome guidance.
>>
>>     section 2.2.1.2 <http://2.2.1.2>:
>>       - so, these are pointers back into a "Telemetry Source
>>     Object"?  What happens if the reference is not found?
>
>     @Ben we will need to discuss this case.  It might be enough to
>     require that dCDNs ensure consistency in this manner, and if not,
>     the payload could be rejected by the uCDN.
>
>
>>
>>     section 2.2.1.3 <http://2.2.1.3>:
>>       - This is subdividing a footprint?  Would it make more sense
>>     (or at least be simpler) to just define more granular
>>     footprints.  It seems confusing to make essentially another way
>>     specify footprints.
>
>     This was a very lengthy part of our discussion in this draft. 
>     There are use cases where "all traffic is not created equal" i.e
>     Low Latency HLS will have much different request characteristics
>     than Bulk game downloads or even Video On Demand.  The CDNi
>     Footprint object doesn't allow for granularity on a published host
>     basis. We were working under the fundamental premise that we were
>     trying to cause as little disruption to the fundamental components
>     of CDNi and trying to encapsulate as much as we could related to
>     capacity management without major overhauls to the original spec. 
>     We would greatly welcome discussion on this topic.
>
>
>>
>>     section 4:
>>       - I think there should probably be discussions of the potential
>>     attacks if the data is intercepted/spoofed.
>
>     Would there be any additional considerations outside of what is
>     already established for FCI in general?
>
>
>>       - I think we also need a privacy section that discusses the
>>     potential issues of divulging internal information about the CDN.
>
>     part of the data requirements are such that the data for capacity
>     is scoped specifically to the uCDN,->dCDN delegation
>     relationship..i.e. the data is very narrowly scoped.  The data
>     being presented is also data about the dCDN by the dCDN.  Would
>     there be any suggestions on how we might want to frame a statement
>     for this?
>
>>
>>     nits:
>>
>>     section 1.3:
>>       "it's current usage" -> "its current usage"
>>       "uCDN of limits of" -> "uCDN limit on"
>>       ", so that the uCDN can use to" -> " so that the uCDN can use
>>     it to"
>>       " i.e. " -> ", i.e., "
>>       "In the event that a" -> "In the event a"
>
>     ACK
>
>
>>
>>     section 2.1:
>>       "CDNi" -> "CDNI"
>>       "reference, allows" -> "reference allows"
>>       "non ambiguous metric use use" -> "non ambiguous metric use be
>>     used"
>>       "current usage and how" -> "current usage.  How"
>>       "to it's dCDN" -> "to its dCDN"
>>
>     ack
>
>
>>     section 2.1.1.2 <http://2.1.1.2>:
>>       " .  I.e. is" -> ", e.g., is"
>>       "5 minutes" -> "300 seconds (i.e., 5 minutes)"
>>       "one hour" -> "3600 seconds (i.e., one hour)"
>>
>>
>
>     ack
>
>>
>>
>>     On Tue, Apr 5, 2022 at 9:08 AM Andrew Ryan
>>     <[e-mail]> wrote:
>>
>>         Greetings,
>>
>>            Similar to the recent solicitation for adoption of the
>>         CDNI Metadata
>>         Model Extensions, I would like to submit a draft for
>>         consideration of
>>         adoption:
>>
>>         https://datatracker.ietf.org/doc/draft-ryan-cdni-capacity-insights-extensions/
>>
>>         The draft outlines extensions to the FCI payload registry,
>>         which will
>>         enable communication of Capacity/Delegation limits between a
>>         uCDN and
>>         dCDN.  This draft has been presented at IETF 111, IETF 112
>>         and most
>>         recently at IETF 113  from which we have incorporated some
>>         great feed
>>         back into several revisions of the draft.
>>
>>         I look forward to any feedback and/or support for the
>>         adoption of this
>>         draft.  Thank you very much for your time.
>>
>>         Andrew Ryan
>>
>>         _______________________________________________
>>         CDNi mailing list
>>         [e-mail]
>>         https://www.ietf.org/mailman/listinfo/cdni
>>