Re: [Anima] Review comments on draft-ietf-anima-grasp-api

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

Hi Guangpeng,

I will try to explain it in much more detail in the new draft. You are quite correct
that the ASA doesn't need to know any details of the session ID, but the problem
comes when the same function is used as the callback for multiple sessions.

Regards
   Brian Carpenter

On 19-Sep-19 18:34, Liguangpeng (Roc, Network Technology Laboratory) wrote:
>> Sorry to insist, but I really disagree. I'm thinking about how I would code the
>> callback function for a resource allocator, and I'm sure it would need some sort
>> of identifier for the particular resource request, since it might be handling many
>> resource requests in parallel. In GRASP terms that is a session identifier.
> 
> I agree there must be a session identifier in GRASP. However I think this draft is about to define the GRASP API for ASA. The ASA invokes functions through these APIs, and a grasp library implement APIs which includes encapsulating GRASP messages, send out them and wait for receiving response messages. So for call-back approach, all the magic is in the library, the library is responsible for mapping session identifier in messages to correct call-back threads. The ASA just need to tell library which call-back function would be executed when invoke a asynchronous API, the ASA knows nothing about the session identifier used in the messages, thus donot need 'session-nonce' parameter in the API.
> 
> Best regards,
> Guangpeng
> 
>> -----Original Message-----
>> From: Brian E Carpenter [mailto:[e-mail]]
>> Sent: Thursday, September 19, 2019 10:13 AM
>> To: Liguangpeng (Roc, Network Technology Laboratory)
>> <[e-mail]>; [e-mail]
>> Subject: Re: Review comments on draft-ietf-anima-grasp-api
>>
>> On 19-Sep-19 13:23, Liguangpeng (Roc, Network Technology Laboratory) wrote:
>>> Please see inline.
>>>
>>>> -----Original Message-----
>>>> From: Brian E Carpenter [mailto:[e-mail]]
>>>> Sent: Thursday, September 19, 2019 4:39 AM
>>>> To: Liguangpeng (Roc, Network Technology Laboratory)
>>>> <[e-mail]>; [e-mail]
>>>> Subject: Re: Review comments on draft-ietf-anima-grasp-api
>>>>
>>>> On 18-Sep-19 18:22, Liguangpeng (Roc, Network Technology Laboratory)
>> wrote:
>>>>> Hi authors,
>>>>>
>>>>> I read the draft. I think the API list for grasp is complete and
>>>>> good. But I have
>>>> some comments on section 2.2.
>>>>>
>>>>> 2.2. Asynchronous Operations
>>>>>
>>>>> '... there are two main techniques for such parallel operations:
>>>> multi-threading, or a polling or 'event loop' structure.'
>>>>>
>>>>> According to my knowledge, 'multi-threading' is a different thing
>>>>> with
>>>> asynchronous programming mechanism. So I prefer to 'asynchronous
>>>> programming mechanisms with multi-threading' here.
>>>>
>>>> OK, will rephrase.
>>>>
>>>>> The main techniques for asynchronous programming discussed in draft
>>>>> should
>>>> be 'polling', however, polling is not so efficient. The more
>>>> efficient one is 'call back' mechanism.
>>>>
>>>> Yes, we certainly need to discuss callbacks. I'm not sure it is
>>>> always more efficient, because that depends on the underlying
>> implementation.
>>>>
>>> Agree that all performance rely on the implementation. Generally saying,
>> polling will consume CPU resource to query response again and again until
>> timeout occurs. In other hand, every implement of agent based on polling
>> mechanism library will couple with the library. That means most time different
>> programmers will write the code respectively for the agent and library. It's
>> more hard to guarantee achieve the best performance. Compared to polling,
>> call back mechanism decouples the whole workflow. Only the library
>> programmer need to care about the asynchronous part performance. The
>> agent programmers just need to implement logics according to 'call back'
>> mechanism then everyone can get the same performance provided by the
>> same library.
>>>
>>>>> 'session_nonce' parameter may be useful for polling, but 'call back
>>>>> function
>>>> pointer( or reference in some programming language)' is normally used
>>>> as parameter of invoked functions. All logics should be executed in
>>>> call back functions which will be called when some events occur, for
>>>> example receiving response message for the request, or message timeout.
>>>>
>>>> Yes. However, if you have two simultaneous negotiations, you will
>>>> still need the session_nonce in the callback, I believe.
>>>
>>> Yes or not, I think. If the library is implemented based on call back mechanism,
>> the agent programmer don't need to know anything about 'session-nonce'. This
>> means there needn't be this parameter in this kind of API. Indeed, the
>> implementer of library need to consider 'session_nonce' carefully if its name is
>> still 'session_nonce'. So, my opinion is 'session_nonce' is necessary parameter
>> for polling-based API, but not for call-back-based API.
>>
>> Sorry to insist, but I really disagree. I'm thinking about how I would code the
>> callback function for a resource allocator, and I'm sure it would need some sort
>> of identifier for the particular resource request, since it might be handling many
>> resource requests in parallel. In GRASP terms that is a session identifier.
>>
>> Regards
>>     Brian
>>
>>>
>>>> Thanks very much,
>>>>
>>>>     Brian
>>>>
>>>>>
>>>>> Best regards,
>>>>>
>>>>> Guangpeng Li
>>>>
>>> Best regards,
>>> Guangpeng
>>>