Re: [P2PSIP] Broken Stop and Wait Algorithm?

["Bless, Roland (TM)" <[e-mail]>]

Hi Michael,

Am 19.10.2013 00:43, schrieb Michael Chen:
> Your fragment re-assembler must tolerate duplicate fragments, because it
> is the nature of the UDP transport. By comparison, ACK is never repeated.

Thanks for clarification. I guessed that it's up to the upper layers
to deal with it, but IMHO this actually an unnecessary potential
additional source of duplicated fragments besides IP packet duplication
and if you don't care about duplicates you can even leave out sequence
numbers in the stop and wait protocol...

Detecting duplicated fragments may not always help:
If you have a single fragment only, you'll immediately pass it
up to the message transport for instance. So in this case you need
to check for duplicate transaction IDs or rely on the implementation
(sec. 6.2.1.):
   Receiving nodes MUST generate semantically equivalent
   responses to retransmissions of the same request (this can be
   determined by transaction ID) if the request is received within the
   maximum request lifetime (15 seconds).

> The sequence number must always incremented so that the receiver can
> build the FramedMessage.received bitmask for the last 32 fragments.

Yes, I read that part, so that you can estimate loss rates
or RTTs.

>     --------- Original Message ---------
>     Subject: [P2PSIP] Broken Stop and Wait Algorithm?
>     From: "Roland Bless" <[e-mail]>
>     Date: 10/18/13 12:46 pm
>     To: "P2PSIP WG" <[e-mail]>
> 
>     Hi all,
> 
>     while implementing more of the RELOAD protocol we came across
>     the following issue:
> 
>     In sec. 6.6.3.1. Stop and Wait Sender Algorithm
>     http://tools.ietf.org/html/draft-ietf-p2psip-base-26#section-6.6.3.1
> 
>     the spec says:
>     For each retransmission, the sequence number MUST be incremented.
> 
>     Retransmissions continue until a response is received, or until a
>     total of 5 requests have been sent or there has been a hard ICMP
>     error [RFC1122] or a TLS alert. The sender knows a response was
>     received when it receives an ACK with a sequence number that
>     indicates it is a response to one of the transmissions of this
>     messages.
> 
> 
>     This would cause potential duplicates at the receiving side:
> 
>     A B
>     ---------> FramedMsg #1 -------> ---> deliver FramedMsg #1
> 
>     |*drop* <-- Ack #1 ><----------
> 
>     (A: timeout
>     + retransmit)
> 
>     ---------> FramedMsg #2 -------> ---> deliver FramedMsg #2
>     (same as above, i.e.
>     *duplicate*).
>     <----------- Ack #2 ><-----------
> 
>     (A: ack received,
>     transmit next frame)
> 
>     ---------> FramedMsg #3 -------> ---> deliver FramedMsg #3
> 
>     Given the current scheme, the receiver cannot
>     distinguish between the retransmitted frame (#2) and
>     a new one (#3). What was the reason to use _new_ sequence numbers
>     for retransmissions? Unless you increment the
>     sequence number by more than 5 for each _new_
>     frame, the receiving side could not filter out
>     duplicates.
> 
>     Did I miss something?