Re: [Dime] OVLI: comments to 4.1

[Ben Campbell <[e-mail]>]

On Dec 12, 2013, at 10:02 AM, Wiehe, Ulrich (NSN - DE/Munich) <[e-mail]> wrote:

> Steve,
> ok, but then its only fair to let individual implementers (of reacting nodes) make the decision whether or not to spend the effort of include a sequence number in the OC-Supported-Features (i.e. make the sequence number optional). 

I don't think that's the sort of fairness the requirements contemplated :-)

Reacting Nodes would only need to calculate a new sequence number if it's capabilities change. That won't happen often. Also, it's trivially easy to calculate a new sequence number. If the reacting node does not include one, the reporting node is forced to re-check the capabilities on every single Diameter request that includes one. (Which will likely be "all requests".)

>  
> Isn’t the situation similar to the proposal to introduce Ongoing-Throttling-Information mandatorily in requests and let implementers of reporting nodes decide whether they want to make use of this information?

No, I don't think it is. That only matters in overload situations. This applies to every single request all the time.

>  
> Ulrich
>  
> See also inline
>  
>  
> From: ext Steve Donovan [mailto:[e-mail]] 
> Sent: Thursday, December 12, 2013 4:34 PM
> To: Wiehe, Ulrich (NSN - DE/Munich); [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
> Ulrich,
> 
> Thanks for the correction -- obviously typed pre coffee this morning.
> 
> The tradeoff is requiring processing on every request versus maintaining state to allow avoiding that processing on 99.99% of requests.
> 
> [Wiehe, Ulrich (NSN - DE/Munich)] The tradeoff is between a) checking the Supported_Features in every request and know what to do, and b) checking the sequence number in every request, looking up a database and then (in 99.99% of cases) know what to do.
> 
> 
> 
> We should let individual implementers make the decision on which approach they prefer.
> 
> Steve
> 
> On 12/12/13 9:22 AM, Wiehe, Ulrich (NSN - DE/Munich) wrote:
> Steve,
>  
> you probably mean „…can’t  be optional…“ rather than “…can’t be mandatory…”
>  
> If so, I understand your logic but this means that  reacting nodes are forced to  implement sequence numbers in OC-Supported-Features just because some reporting nodes prefer to do things complicated rather than simple and I’m not convinced that this would be the way to go.
>  
> Ulrich
>  
>  
> From: ext Steve Donovan [mailto:[e-mail]] 
> Sent: Thursday, December 12, 2013 2:55 PM
> To: Wiehe, Ulrich (NSN - DE/Munich); [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
> Ulrich,
> 
> No, the sequence number can't be mandatory.  It always needs to be sent to allow for end-points that what to use the logic I proposed.
> 
> An endpoint that receives the sequence number can choose to ignore it, but endpoints must always send sequence numbers.
> 
> Steve
> 
> On 12/12/13 3:30 AM, Wiehe, Ulrich (NSN - DE/Munich) wrote:
> Steve,
>  
> do you mean „ keep sequence number in OC-Supported-Features as an optional AVP that may be ignored by the receiver and need notbe included by the sender”?
>  
> Ulrich
>  
> From: DiME [mailto:[e-mail]] On Behalf Of ext Steve Donovan
> Sent: Wednesday, December 11, 2013 2:34 PM
> To: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
> Ulrich,
> 
> My email showed how a reporting node could avoid the work of recalculating capability information on a message by message basis using the sequence number.  This approach does require maintaining state.
> 
> As Ben pointed out, it is also possible to not follow my logic and do as you propose by ignoring the sequence number and going through the work of calculating the response every time.  
> 
> Which approach to take is clearly an implementation decision.  We should keep the sequence number to allow for the stateful approach for implementations that are willing to take advantage of maintaining state to avoid work.
> 
> Steve
> 
> On 12/11/13 1:34 AM, Wiehe, Ulrich (NSN - DE/Munich) wrote:
> Jouni,
>  
> I do not agree.
>  
> My statement was general: reporting node may be server or SFA; supported features may be defined features (default algo) or future extensions.
>  
> Ulrich
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Tuesday, December 10, 2013 10:46 PM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
> Ulrich,
>  
> On Dec 10, 2013, at 2:18 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
>  
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Tuesday, December 10, 2013 12:32 PM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
>  
> On Dec 10, 2013, at 12:41 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> Hi Jouni,
>  
> my understanding from Steve's clarification was that the reporting node would setup and maintain a database of "states", one per reacting node where a state of a reacting node is identified by a sequence number and the database entry would contain the pre-calculated OLR. 
> Hard to see how this is done without consuming resources.
>  
> It is mostly static information. Still I do not see how
> you would get away without any state.
> [Wiehe, Ulrich (NSN - DE/Munich)] There is no need for the reporting node to store and maintain information per reacting node because the reacting node actually tells within the request message all information the reporting node needs to know. The reporting node simply fetches the pre-calculated OLR that matches the supported features of the reacting node as indicated in the request and appends it to the answer.
>  
> This could be true for the default algorithm in this spec. However,
> given the extension mechanism we have in place it is quite possible
> that the assumption does not hold for new features.
>  
> Also, if I were to implement a server front end agent that would
> definitely need state and still ought to comply the protocol spec.
>  
> - Jouni
>  
>  
>  
>  
> Furthermore,
> Why would the reacting node be interested in config changes of the reporting node?
> Isn't it so that the reacting node is only interested in OLR changes?
>  
> Sigh.. say the traffic abatement algorithm changes or new features
> get added. Those have more impact than just OLR change.
>  
> - Jouni
>  
>  
> Ulrich
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Tuesday, December 10, 2013 9:41 AM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
>  
> On Dec 9, 2013, at 4:43 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> But maintaining a specific state per reacting node is very resource consuming for the (overloaded) reporting node. It is simpler and more efficient to base any processing logic on actual information received in the request rather than on information stored from previous communication.
>  
> The "state" in a reporting node is merely the current configuration
> and a counter for sequence number. Hard to me see that as resource
> consuming function.
>  
> And the earlier comment of mine is done from reacting node point
> of view, since it is more interested in the possible config changes
> in the reporting node (e.g. S6a is stateless application, configuration
> can change at any time).
>  
> - Jouni
>  
>  
>  
> Ulrich
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Monday, December 09, 2013 2:25 PM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
>  
> On Dec 9, 2013, at 3:13 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> There is a fundamental difference:
> OLRs need to be stored, Feature-Vectors not.
>  
> How come feature vector does not need to be stored? I do not
> get that? I would set my implementation to a specific state
> or mode based on the feature vector. When that changes I'd
> like to know that. And then keep functioning based on that.
>  
> - Jouni
>  
> When receiving an OLR you may want to know whether its worth the effort to replace an already stored OLR with the received OLR.
> When receiving a Feature-Vector you just act on it.
>  
> Ulrich 
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Monday, December 09, 2013 1:55 PM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
>  
> In the same vein as folks wanted send OLRs with the new or old information,
> the feature vector should behave in a same way IMHO. That implies there are
> situations when a reception of the feature vector does not change anything
> in an endpoint current behavior.
>  
> - Jouni
>  
> On Dec 9, 2013, at 2:47 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> Isn't it so that the Feature-Vector (if present) always contains something that an implementation needs to act upon?
>  
> Ulrich
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Monday, December 09, 2013 1:12 PM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
> Ulrich,
>  
> On Dec 6, 2013, at 3:03 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> Hi Jouni,
>  
> thank you for your response.
>  
> With regard to 3) 
> I still fail to see the usecase for Sequence-Number or TimeStamp within OC-Feature-Vector. Please clarify.
>  
> Since we also allow extending the OC-Feature-Vector beyond recognition, 
> it has good chances become a rather complex grouped AVP. In that context
> the Sequence-Number allows an easy and quick way to check if the feature
> vector contains something that an implementation needs to act upon.
>  
> With regard to 4)
> This was not obvious to me. (The obvious typo is the missing "of" between "one" and "the").
>  
> Ack. Fixed the missing 'of'.
>  
> - Jouni
>  
>  
> Best regards
> Ulrich
>  
>  
> -----Original Message-----
> From: ext Jouni Korhonen [mailto:[e-mail]] 
> Sent: Friday, December 06, 2013 11:17 AM
> To: Wiehe, Ulrich (NSN - DE/Munich)
> Cc: [e-mail]
> Subject: Re: [Dime] OVLI: comments to 4.1
>  
>  
> On Dec 5, 2013, at 3:23 PM, "Wiehe, Ulrich (NSN - DE/Munich)" <[e-mail]> wrote:
>  
> Dear all,
>  
> here are comments to clause 4.1:
>  
> 1. The OC-Feature-Vector AVP is no longer a vector; the name of the AVP may be misleading. Proposal is to rename it to "OC-Supported-Features AVP"
>  
> OK with me.
>  
> 2. The OC-Feature AVP is a vector of features. Proposal is to rename it to "OC-Feature-Vector AVP"
>  
> OK with me.
>  
> 3. The OC-Sequence-Number within OC-Feature-Vector only makes sense if the receiving reporting endpoint can determine the identity of the reacting endpoint (which is not necessarily the origin host (client),
>  
> My original proposal was to have seqnr as a timestamp. Some folks argued
> it is no good and suggested seqnr. I still think time makes more sense than
> seqnr.
>  
> it may be an agent and it may not always be the same agent), and if the reporting endpoint is required to store the OC-Feature-Vector / reacting-endpoint-identity pair (which I think both is not required). The reporting endpoint can base its processing logic on the actually received OC-Feature-Vector value, no matter whether it is brand-new or old but stil valid. Proposal is to delete OC-Sequence-Number AVP from OC-Feature-Vector.
>  
> Do not agree removing it.
>  
> 4. The text
>  
> The reporting node that sends the answer also includes the OC-
> Feature-Vector AVP that describe the capabilities it supports.  The
> set of capabilities advertised by the reporting node depends on local
> policies.  At least one the announced capabilities MUST match
> mutually.  If there is no single matching capability the reacting
> node MUST act as if it does not implement DOIC and cease inserting
> any DOIC related AVPs into any Diameter messages with this specific
> reacting node.
>  
> is not clear.  Would the reporting node include the OC-Feature-Vector AVP in the answer only if there is at least one matching capability? 
>  
> Because then they have found a way to exchange something that both ends
> know how to handle it.
>  
> Mandating the reacting node to cease for all time inserting OC-Feature-Vector AVPs if it did not get back 
>  
> There is an obvious typo. It should say:
>  
> policies.  At least one the announced capabilities MUST match
> mutually.  If there is no single matching capability the reporting
> node MUST act as if it does not implement DOIC and cease inserting
> any DOIC related AVPs into any Diameter messages with this specific
> reacting node.
>  
> - JOuni
>  
>  
> at least one match is also not ok. The request might have been a realm-type request (i.e. without Destination Host) and the reacting node cannot control whether subsequent requests will take the same path to the same reporting node.
> Even if the request contains a destination host the reacting node cannot know wether the reacting node's capabilities have been modified by the time a subsequent request is sent. 
> Proposal is to keep only the first sentence and delete the rest.
>  
> Ulrich
>  
>  
> _______________________________________________
> DiME mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dime
>  
>  
>  
>  
>  
>  
>  
> _______________________________________________
> DiME mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dime
>  
>  
>  
>  
> _______________________________________________
> DiME mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/dime