Re: [Tsvwg] SCTP applicability statement

["Brian F. G. Bidulock" <[e-mail]>]

Allison,

A couble of remarks on the watchdog timer from AAA and its
(lack) of applicability to SS7 UAs:

1)  The timer is 30 to 60 seconds.  SS7 UAs must normally
    failover with 0.1-0.5 seconds.

2)  The watchdog timer is an application transaction timer.
    SS7 UAs have no implicit acknowledgement or transaction
    structure within the message flow that is usable by M2PA,
    M2UA, M3UA or SUA in determining when to cancel the
    watchdog timer.  Data in these UAs is unstructured (or
    at least its structure is not known to the UA).  A sequence
    number would need to be added to messages and acknolwedged
    separately to cancel the timer.

Because there is no structure know by the UA (no separate
transaction and response as there is in AAA), sequence numbers
and acknowledgements would be required for each message.
Then the watchdog timer reduces to a retransmission timer.

So, it appears that you are saying that SIGTRAN protocols must
implement sequence numbers, acknowledgements, queueing and
retransmission after all.

Are you sure this is what you wish to recommend rather than
fixing several problems in SCTP?

--brian


Allison Mankin wrote:                                     (Sun, 04 Nov 2001 21:41:38)
> Brian,
> 
> > So, is your answer no?
> 
> Yes, my answer is no :) - SCTP is not intended to fail over between
> different associations - the protocol reliability is within
> the association.  With the endpoints being on different boxes,
> that is, in arbitrary network locations, it would be a different
> protocol to do what you request.
> 
> But multiple SCTP associations can still be used as failovers
> for each other, through use of an application layer watchdog and
> a short list of duplicate transactions that might need to be sent.
> The section on aaa transport that I recommended (3.2 in
> draft-ietf-aaa-transport-04.txt) is about a robust way to do this.
> 
> > Is it fair to say that SCTP does not support failover between
> > multi-homed hosts, between associations, your second case?
> > 
> Yes, that's fair.  But to repeat, those different associations have
> arbitrary endpoints.   How do SIGTRAN applications know what to make
> of a set of message that start arriving suddenly because failover has
> occurred - there must be application support of backup server use.
> 
> > Are you saying that applications such as SIGTRAN, RSERPOOL, AAA,
> > must implement their own sequence numbers and acknowledgements
> > and retransmission queues between associations?
> > 
> No, I think SIGTRAN and RSERPOOL apps (and aaa, by way of the draft I
> recommended you look at) can use multiple associations with an application
> layer algorithm that is different (and simpler) than sequence numbers,
> acks and transport retransmission queues.  
> 
> There needs to be duplicate detection of transactions, but failing over
> within the SCTP association can result in duplicates too.  
> 
> I think you should look at the algorithm I've recommended and compare it
> with the specific requirements.  
> 
> Allison
> 
> > 
> > Allison Mankin wrote:                                     (Sun, 04 Nov 2001 16:
> > 22:14)
> > >
> > > > Does SCTP support failover between multi-homed hosts?
> > > > 
> > > > 
> > > > Is failover between multi-homed hosts within the scope of SCTP?
> > > 
> > > Brian,
> > > 
> > > SCTP's multi-homing has no relation to anything between*
> > > hosts, only between interfaces of one host.  There might be
> > > some kind of cluster that appears to its SCTP peer be one host 
> > > with multiple interfaces, but it would not be within the scope
> > > of SCTP to know about the cluster. 
> > > 
> > > In draft-ietf-aaa-transport-04.txt, section 3.2. there's a 
> > > good discussion of failover/failback, which distinguishes
> > > among:
> > > 
> > > - what a cluster does in its cluster OS or cluster communication
> > >   system 
> > > - what application-level watchdogs do, where the failover is among
> > >   multiple potential peers (here AAA servers)
> > > - what SCTP does
> > > 
> > > If all interfaces of an SCTP host are down, that's the second case.
> > > 
> 
> 
> ---
> You are currently subscribed to sigtran as: [e-mail]
> To unsubscribe send a blank email to [e-mail]

-- 
Brian F. G. Bidulock
[e-mail]
http://www.openss7.org/

_______________________________________________
tsvwg mailing list
[e-mail]
http://www1.ietf.org/mailman/listinfo/tsvwg