On Oct 8, 2006, at 3:39 PM, YAMAMOTO Takashi wrote:

>> On Oct 8, 2006, at 2:49 PM, YAMAMOTO Takashi wrote:
>>
>>>>> "cwnd_inflation" sounds weird to me, given that what it does is
>>>>> ack handling for fast recovery.  otoh, "new_data_acked" inflates
>>>>> cwnd.
>>>>> isn't it better to unify these two callbacks?
>>>>
>>>> Well, "inflation" is probably not the best word, but I don't see  
>>>> why
>>>> we want to unify them. What do you have in mind?
>>>
>>> because both of them are called sequentially in most cases
>>> (unless rcvacktoomuch, for which we can't do much anyway),
>>> it isn't clear for me what's the benefit to have two callbacks
>>> rather than one.
>>
>> Right now, I don't see a way to have just one callback because of  
>> SACK.
>> We would have to call tcp_sack_newack() in every other
>> tcp_xxxx_cwnd_inflation().
>
> if you have sack, whatever xxxx is, tcp_xxxx_cwnd_inflation isn't
> needed to be called?
> (i'm not sure about the semantics of cwnd_inflation.)

Oh yes, you are right. I'll remove the new_data_acked and  
cwnd_inflation callbacks and add a new one called 'newack' (the name  
change is to use what we already have: tcp_reno_newack,  
tcp_sack_newack, etc.).

>
>> The point here is to introduce HSTCP and Westwood+ after these  
>> changes.
>
> these callbacks should be separated for them, you mean?

No, I was just saying what I would do next ;-)

--
Rui Paulo