Re: PN2AC transition clarification in n+k redundancy model

Pascal,

The intention is that the pending-to-active transition happens in a similar manner to the inactive-to-active transition.
So the "4.3.4.3. ASP Active Procedures" section, that you quote below, is applicable also to the pending-to-active
transition.

You are correct that in some cases "Allowing a single Active ASP to move AS-ACTIVE would help avoid traffic failure
conditions". However it's equally possible that allowing a single (underpowered) ASP to go active, in a situation where
2 or more ASPs are needed to handle a high traffic load, could just prolong the period of instability before equilibrium
is reached. So it's probably best to make the values for "n" and "k" configurable, and let the user decide how the ASP
should behave.

Barry Nagelberg

> -----Original Message-----
> From: [email protected] [mailto:[email protected]]On
> Behalf Of Pascal Bouveron (Netbricks)
> Sent: Monday, April 07, 2008 1:14 PM
> To: [email protected]
> Cc: Pascal Bouveron (Netbricks)
> Subject: [Sigtran] PN2AC transition clarification in n+k redundancy
> model
>
>
> Hello,
>
> I need some clarifiction for the PN2AC transition in n+k redundancy model.
> What condition should trigger the move from AS-PENDING to AS-ACTIVE state ?
>
> Is 1 ASP-ACTIVE enough to move AS-ACTIVE again, or should the AS wait for "n" Active ASPs again, as during the startup
> phase ?
>
> In an n+k redundancy model, an AS moves AS-PENDING when it has no more Active ASP/IPSP to serve its traffic.
>
> In rfc4666, the "Figure 4: AS State Transition Diagram" describes the "PN2AC" transition (Pending-to-Active) as "One
ASP
> moves to
> ASP-ACTIVE".
> Meaning that a single ASP/IPSP moving ASP_ACTIVE is enough to move AS_ACTIVE again.
>
> However, there is no text dealing explicitely with this particular case in the procedures sections.
>
> The only section dealing with the n+k redundancy case is "4.3.4.3.  ASP Active Procedures":
>
> "   For the n+k redundancy case, ASPs that are in that AS should
>   coordinate among themselves the number of active ASPs in the AS and
>   should start sending traffic only after n ASPs are active."
>
> and:
>
> "    At startup or restart phases, an SGP or IPSP, upon receipt of an ASP
>   Active message for the first ASP in a Loadshare AS, SHOULD NOT direct
>   traffic to a newly active ASP until it determines that there are
>   sufficient resources to handle the expected load (e.g., until there
>   are "n" ASPs in state ASP-ACTIVE in the AS). "
>
>
> So, the question is:
> is 1 ASP-ACTIVE enough to move AS-ACTIVE again, or should the AS wait for "n" Active ASPs again ?
>
> Allowing a single Active ASP to move AS-ACTIVE would help avoid traffic failure conditions, by allowing to run in
> "degraded-capacity" condition, rather than having to wait for the "full-capacity" to be reached again.
>
> Any thoughts about that ?
>
> Thanks.
>
> Pascal Bouveron
> www.netbricks.com
>
>
> _______________________________________________
> Sigtran mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/sigtran