Re: [Anima-signaling] Concern about GRASP Flood message

[Brian E Carpenter <[e-mail]>]

On 22/07/2016 00:03, Toerless Eckert wrote:
> On Thu, Jul 21, 2016 at 10:31:32PM +1200, Brian E Carpenter wrote:
>> The loop count is already a mandatory field in every objective.
>> We can add original-loop-count as a specific field in this
>> objective. No problem.
> 
> Ok. Any guidelines when you consider some attrbite to be part of
> an objective as opposed to eg: a GRASP header ?
> 
>>>  Refine flood pruning as: 
>>>  If the distance of a received announcement is smaller than the
>>>  distsance of the cached announcement, flood the new packet (and put
>>>  into cache).
>>
>> I have a problem with that but it's connected with another change
>> we have to make: caching multiple versions of a flooded objective
>> iff they come from different sources. We need to do that so that
>> in your use case (and probably others), the user ASA can make its
>> own choice among several alternatives. So in the flood cache we
>> will overwrite the cached value from address A if we receive a
>> new value from A, but the cached value from address B will not be
>> overwritten. As long as we do that, the distance comparison isn't
>> in the cache, it's in the user ASA.
> 
> So there are two comparisons:
> 
> The first comparison is in GRASP when a new message from A is received.
> It's compared with the exxisting cache of A, and when it's better
> (longer remaining cache time or lower distance), then it does get into
> the cache and is flooded. Else it's dropped, and existing cache is better.
> 
> The second comparison is as you said in the ASA between the A cache and B
> cache.
> 
>> So, to be clear, the objective syntax is:
>>
>> objective = ["ANI_Registrar", objective-flags, loop-count, original-loop-count, stack-list]
>>
>> and in the flood cache, it is tagged with its sending address (which
>> I noticed is already in the Flood message anyway, so that part is solved).
> 
> Ok.
> 
> And the stack list will have the locators ?

Certainly they will be available somehow, but let me pass on that question for this
week. I need to think through all the implications of what we've been discussing
when I'm not in a hotel with 1200 geeks.

   Brian

> 
> Cheers
>     Toerless
> 
>>     Brian
>>
>>> Cheers
>>>     Toerless
>>>
>>> On Thu, Jul 21, 2016 at 09:06:02PM +1200, Brian E Carpenter wrote:
>>>> On 21/07/2016 20:45, Carsten Bormann wrote:
>>>>> Brian E Carpenter wrote:
>>>>>> There is already one use case emerging: using Flood
>>>>>> to announce bootstrap Registrars to all bootstrap Proxies. A proxy needs to
>>>>>> be able to choose which registrar to use, so needs to
>>>>>
>>>>> Well, the content of the message then appears to be "x is offering to be
>>>>> bootstrap registrar".  This can of course alternatively be represented
>>>>> by "x says: 'I am offering to be bootstrap registrar'".
>>>>
>>>> Exactly, but the Flood mechanism loses track of x when the message is relayed.
>>>> That would need a protocol change.
>>>>
>>>>    Brian
>>>
>