RE: Lossless fail-over

Jan,

Please see below for some comments/questions.

   Thanks,
   Tolga

> -----Original Message-----
> From: Kolomaznik Jan [mailto:[email protected]]
> Sent: Tuesday, June 20, 2006 6:37 AM
> To: [email protected]; Tolga Asveren
> Cc: [email protected]
> Subject: RE: [Sigtran] Lossless fail-over
>
>
> I thing there was a lot of misunderstandings yesterday. I used incorrect
> terminology. Words like fail-over or change-back belong to SS7 when
> changing the signalling link within a signalling link set. This was very
> misleading. I'm sorry for it.
>
> There may exist two levels of ASP1 backup. ASP1 may be backuped on SCTP
> and/or M3UA/ SUA level.
>
> a) When the SCTP level is used than the ASP1 and ASP2 (which backups
> ASP1, uses different IP connection and may be geographically far from
> ASP1) operate over the same SCTP level. Although ASP1 and ASP2 may be
> realized on different computers their SCTP levels looks like one SCTP
> level on multihomed AS from ASP3 point of view. This type of backup is
> analogy to the Tolga's mentioned SS7 backup.
[TOLGA]This would be similar to having a distributed MTP2, which is not
common -actually I never heard of a real-life implementation-. A distributed
SCTP would be hard and more importantly inefficient to implement. Again I
never heard of something like this. So, this type of backup seem not to
exist in real-life (and is not something I tried to imply before, sorry for
the confusion ).

There are different issues redundancy tries to address:
i)Host Redundancy
ii)Network Interface Redundancy
iii)Network Redundancy

For i), in SS7 people usually have multiple hosts acting as a single point
code. In SUA/M3UA this is addressed by using multiple ASPs for the same AS.
When there is a host failure neither SS7 not M3UA/SUA provides lossless
failover.

For ii), SS7 addresses network interface redundancy by utilizing multiple
links, possibly using multiple SS7 cards. In M3UA/SUA this is addressed with
SCTP multihoming. Please note, when there is a failover in an SCTP
association, this happens with no loss/missequencing.

For iii), SIGTRAN relies on reasonable IP network design and on SCTP
retransmisions. SCTP multihoming plays a role as well by allowing a host to
coexist in multiple subnets. Actually even with a singlehomed system,
network redundancy is not hard to achieve. For QoS, there are generic QoS
mechanisms defined for IP, e.g. RSVP, Differentiated Services, MPLS.
Actually also in SS7 between two nodes directly connected with a SS7 link,
there is an underlying topology, which needs to be engineered.

>
> b) There may exist also different type of ASP1 backup. Lets to imagine
> the M3UA/SUA only is synchronized between ASP1 and ASP2. In this case
> the SCTP level on ASP1 is not the same entity like SCTP level on ASP2
> from ASP3 point of view - it looks like two single-homed SCTP entities.
> I think this situation has no analogy in SS7.
[TOLGA]To me, this is similar to having multiple hosts acting as a single
signaling point. All redundancy issues could be addressed as specified
above.
>
> The difference between a) and b) is in network failure handling. In case
> a) the SCTP handles it. In this case there is not necessary something
> new. There exist everything what SCTP need for it. Different situation
> is in case b). In this case M3UA/SUA must handle any network failure.
> And (in my opinion) the CORID mechanism is very helpful in such
> situations. The right question is if it is necessary to make ASP backup
> according to case b) and not according to case a)? When somebody is able
> to create synchronization on M3UA/SUA level what may be a reason for
> non-creation synchronization on SCTP level also?
[TOLGA]The rationale is network/network interface errors are handled by SCTP
+ IP network design and for host failures, some loss of messages is
tolerated like in SS7.
>
> The difference is also in reaction time on network failure. Both
> detection and reaction time is much shorter in case a), but case a) is
> more difficult for realization.
[TOLGA]For a), you would be dependent on detecting loss of a single path,
for b) of multiple paths. Is there really a huge difference? It could be the
case that SCTP parameters need to be choosen a bit loose if IP network
quality is really low. In that case I believe it could be helpful for an ASP
to help SGP to detect failure of another ASP.
>
> So my question is if there is something helpful (like CORID) for network
> failure recovery in situations like b)?
>
> The benefit of CORID I see also in possibility in usage when network
> failure is repaired and the original entity (ASP1 in our example) tries
> to takeover traffic back - in case of b) of course.
>
> Jan
>
> -----Original Message-----
> From: Brian F. G. Bidulock [mailto:[email protected]]
> Sent: Monday, June 19, 2006 10:02 PM
> To: Tolga Asveren
> Cc: [email protected]
> Subject: Re: [Sigtran] Lossless fail-over
>
> Tolga,
>
> Tolga Asveren wrote:                            (Mon, 19 Jun 2006
> 12:37:03)
> >
> > a) Relying on a single router for IP connectivity is really very very
> bad
>
> You still miss the point: it doesn't require one router: all that it
> requires is one back-hoe.
>
> --brian
>
> --
> Brian F. G. Bidulock
> [email protected]
> http://www.openss7.org/
>
> _______________________________________________
> Sigtran mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/sigtran