The MPLS WG Archive

[neil.2.harrison@bt.com]

Shahram Davari wrote 31 May 2002 19:40
> To: 'Yakov Rekhter'
> Cc: Giles Heron; 'mpls@uu.net'; 'ppvpn@ppvpn.francetelecom.com'
> Subject: RE: Basic LDP Question 
> 
> 
>  
> > > in IP also you can do ACL.
> > 
> > yes, but it has its own cost/complexity.
> 
> Sure, but using LDP you are forced to run some kind of OAM, to detect
> mis-config of LSRs. So nothing is gained. The cost/complexity 
> of filtering is
> replaced with cost/complexity of OAM.
> 
> -Shahram
NH=> I am not sure you are capturing all the issues here Shahram.  Some P
LSRs do load-balancing based on IP address hashing as a sort of TE function
because LDP just gives SPF and so leads to networks that are
traffic/topology sensitive.  This gives rise to the additional points:
-	might this not also be considered a security issue itself from a
customer perspective, ie VPN customer IP pkts are looked at for sorting into
different streams?
-	what about the added processing load this creates on the P routers
involved?
-	does it not break the simple 'label forwarding' paradigm that seems
to be the raison detre for MPLS?
-	and don't we effectively get back to IP fowarding now.....albeit
with an inability to do IP forwarding because the P routers don't have the
IP routes (of the VPNs)

To which you can now add your OAM issue because yes we have to fault-manage
these entities since they do fail...and this now includes the load-balancing
functions.

Further:
-	given we can't give any solid SLAs to such VPNs when based on LDP
(because of the VPN population pkt merging issue);
-	so to compensate we have to over-engineer and hope;
-	VPN topologies are invariably long-holding....so their LSPs really
need decoupling from the IGP in case it flexes routes or misbehaves;

In view of all these factors perhaps it might have been wiser to make
ER/signalled LSPs the interworking default for VPNs rather than LDP in
rfc2547.

regards, Neil