On 3/14/14, 3:43 AM, Eggert, Lars wrote:
> Hi,
> 
> On 2014-3-14, at 10:36, joel jaeggli <[email protected]> wrote:
>> The expected response to congestion induced loss (or in fact any other
>> source of induced loss) across multiple windows is to fail. this is
>> entirely consistent with  how traditional tftp implementations address
>> persistent faults. If you treat acks of partial window sized blocks as
>> non-acks are in lockstep tftp implentations and you have imposed at
>> threshold after which you will fail then induced loss will cause you to
>> fail as it does deployed tftp implementations.
> 
> I don't quite follow. What is the behavior is some packets of a window are 
> lost, retransmission of the full window? Retransmission of the delta?
> 
> In any event, unless the window is shrunk and/or delays before retries are 
> being applied, I don't see anything here that actually reduces the sending 
> rate when loss is detected, which is required.

aborting the transfer is pretty effective at reducing reducing the rate.

> Lars
> 


Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to