The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2002-Mar> msg00189



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Y.1711 requirements from IETF

  • From: George Swallow <swallow@cisco.com>
  • Date: Thu, 21 Mar 2002 10:24:32 -0500
  • Cc: Shahram Davari <Shahram_Davari@pmc-sierra.com>, "'Thomas D. Nadeau'" <tnadeau@cisco.com>, "'mpls@uu.net'" <mpls@UU.NET>

At 09:21 AM 3/21/2002 -0500, David Allan wrote:

George:

Is suggesting that all LSRs have the full internet routing table and MPLS full connectivity a realistic example.

I got out of the business of telling operator how to run their networks long ago.  I would not suggest this, but
our implementation will support it.  The way most ISPs currently run, turning on LDP without doing anything
else will get you exactly the above. 

I thought eliminating BGP in the core simply meant default route to the BGP speakers, and full one-hop mesh of BGP speakers. Otherwise the internet COULD not scale as every LSR would require a full routing table, and metrics presumably would be leaked from BGP into the IGP. Could such a network **EVER** converge?

Yes you can use MPLS to eliminate BGP in the core.  Most ISPs today, however, are carrying the full internet routing table in *all* of their routers and they do converge.

...George

cheers
Dave

> -----Original Message-----
> From: George Swallow [mailto:swallow@cisco.com]
> Sent: Wednesday, March 20, 2002 5:59 PM
> To: Shahram Davari
> Cc: 'Thomas D. Nadeau'; 'mpls@uu.net'; swallow@cisco.com
> Subject: Re: Y.1711 requirements from IETF
>
>
> > Where did you get this " ALL LSPs MUST" statement from? May
> be you are
> > reading a different Y.1711 than I have. All I can find are
> the following
> > statement:
> >
> > " ... It is strongly recommended that CV OAM packets be
> generated on all LSPs
> > (in order to detect all defects and potentially provide
> protection against
> > traffic leakage both in and out of LSPs). "
>
> I have one customer providing an IP VPN service who has >1500 PE
> routers and probably well over a million LSPs.  Most of the LSPs
> represent VPN connectivity can carry very little traffic.
>
> Would you "strongly recommend" that CV OAM packets be generated on all
> their LSPs at a rate of 1 per second?  Can you point to *any*
> currently deployed network other than those using MPLS solely for
> traffic engineering where this would be a good idea?  If I'm carrying
> the full internet routing table and using LDP downstream unsolicited
> should every router generate 108,000 (the current number of routes) CV
> OAM packets per second?
>
> Perhaps you should revisit this "strongly recommended"
> recommendation.  Perhaps you should provide for generation of CV OAM
> packets at rates other than 1  second.
>
> ...George
>
> ==================================================================
> George Swallow       Cisco Systems                  (978) 497-8143
>                      250 Apollo Drive
>                      Chelmsford, Ma 01824
>
>