Cell Relay Archive

[Mrityunjay Kumar <mrityu@galaxy.npi.stpn.soft.net>]

> My question is what is the significant problem with retransmission and why it
> is not performed with other PDUs also.

Consider following case: 

SSCOP A	-----BGN---------->SSCOP B(S1)
	Lost	<----BGAK-- (S1->S2)	
	-----BGN----------> 
	<-----------------  

A and B both boot up. Initially SQ(retransmission id) is 0.
A increments its SQ to 1, sends a BGN to B and goes into
outgoing_connection_pending state. B receives BGN, assigns its SQ 1, and
goes into incoming_connection_pending state.B (After getting response
from the sscf-user) sends BGAK and goes into data_transfer_ready state.
Now suppose this PDU is lost. 
A times out and sends a BGN again, **WITH SAME SQ**. B detects
retransmission, and sends a BGREJ again, maintaining its state as
data_transfer_ready.
If retransmission check is not there(as in Q.SAAL), this BGN will be
identified a reestablishment request(which it is not!) and the sscf-user
will be reset, thus incurring unnecessary overhead.

Similarly, in case of RS and ER pdu, where actions involve significant
overhead(like informing sscf, initializing variables, buffers, lists),
we want to avoid additional overhead due to retransmissions, and this is
done by detecting a retransmission by this method. Though this method of
retransmission check has a problem in case of BGN, because if now A
reboots and attempts legitimately(no retransmission!), B still ignores
him! ATM Forum proposed solution is to accept the BGN in this
scenario.Anyway..

For other PDUs, retransmission check is not performed because:
(1) SD pdu is anyway protected by the sequence numbers.
(2) Other PDUs are not retransmitted at all.


Hope this helps,


With regards,
Mrityunjay
Network Programs, India.