Re: Processor Outage and Resynchronizing Sequence Numbers

Hecht,

Hecht Martin wrote:                           (Tue, 24 Feb 2009 15:41:31)
> Hi Brian,
> 
> Thanks for this and the other responses.
> 
> It is my understanding that any message received during the local
> processor outage and received between the processor recovered and the
> READY handshake, will not be sent to L3.  This means that they are lost
> messages.

They may or may not be sent to L3 depending on the
duration of the outage.

If the processor outage was of "long-term", the messages
are too old and MTP L3 will flush buffers (discarding
these messages) before resuming.

However, when the outage is short, L3 will simply resume
without flushing buffers and these messages in the RB
will be delivered to L3 in the normal way upon resume.

> 
> I understand from your emails that the sync will be as if they were
> ack'd, even though they are lost, i.e. not sent to L3.
> 

If you are used to thinking of an L2 that only acks
messages when they are delivered to L3, consider that
these messages _have_ been presented to L3 as a group,
and so are all acknowledged: L3 simply decides whether
to discard the group or not after they are presented.

Therefore it is not a resynchronization of sequence
numbers.  

ITU-T Rec. Q.704 (07/96) 5.6.2:

  "In the case that processor outage is of long-term,
   the remote side completes the time-controlled
   changeover procedure.  In order to (sic) avoid
   sending out old messages (see clause 8/Q.703) the
   level 2 buffers on both sides of the link should be
   flushed immediately, when the local/remote processor
   outage state terminates.  The decision whether
   processor outage is of long-term is a local thing.
   At the remote side long-term processor outage occurs
   when the time-controlled changeover timer T1 expires.
   At the local side an equivalent timer is used in
   quite the same way."

MTP L3 timer T1 represents the maximum amount of time
that traffic will be withheld (buffered) on time-
controlled changeover.  Any message that is older (has
been buffered for longer) than the duration of this
timers during processor outage is "too old" to be
placed on the network and is flushed.  This is because
messages in the same flow (SLS) may have already been
diverted under the time-controlled changeover procedure
to an alternate link and message loss is preferrable to
message missequencing:

ITU-T Rec. Q.706 (03/93) 1.2:

  b) Loss of messages

     Not more than one in 10^7 messages will be lost
     due to failure in the MTP.

  c) Message out-of-sequence

     Not more than one in 10^10 messages will be
     delivered out-of-sequence to the User Parts due to
     failure in the MTP.  This value also includes
     duplication of messages.

> Is this understanding correct?
> 
> Thanks again,
> Martin

-- 
Brian F. G. Bidulock
[email protected]
http://www.openss7.org/