RE: [PWE3] Proposed new 5.4.3 for arch draft

["Yaakov Stein" <yaakov_s at rad.com>]

Ben, 
 
> I maybe missing something but I don't see how using a 
> best-effort flavour of MPLS (i.e. any non-P2P non-co-ps mode 
> of MPLS) can really solve the IP equal-cost problem when it 
> introduces it's own equal-cost problem (because it tracks the 
> IP IGP).  It's just shifting the problem to somewhere else in 
> the network stack.

What I meant was that one could define the FEC based on more
than just the destination IP address, e.g. using the source IP address
as well. Then the LER could assign a label and solve "fish diagram"
overload problems for best-effort IP.

> Under what circumstances does a cl-ps best-effort flavour of 
> MPLS add value above and beyond what could be achieved using 
> cl-ps best-effort IP (I can't think of any)?

Continuing my previous remark, the performance of such a network
is better than an IP ECMP scheme that sends packets along different
routes on a packet-by-packet basis, based only on IP destination
address.

Of course the MPLS network is being used in "CO mode",
even though there is no TE. 
I really don't know what "CL MPLS" is,
unless you mean that MPLS switches can also forward unlabeled IP
packets, 
but then they are functioning as an IP routers.

> > Admittedly peaking into client layer headers is a hack, but  it is
such 
> > a useful one that it may qualify as one of those rare exceptions
where 
> > the rules can be broken
 
> Useful to whom?  I would argue that if you consider its full 
> implications (including OAM, fault finding/diagnosis, SLA 
> tracking and NMS/OSS implications) then it isn't that useful 
> to service providers, it's just a complicated pain in the backside.

Once again, I am not forcing you to use ECMP
for well-engineered operator networks with TE. 
But this mechanism IS useful for best-effort traffic,
and there is an awful lot of best-effort traffic out there.
 
> IMO it's a case of treating the symptoms and not the cause, 
> this is further compounded by the additional OAM etc. hacks 
> required to fault manage the original ECMP hack.  IMO time 
> would be better spent deprecating the hack and concentrating 
> on how to migrate away from it than producing new hacks that 
> try and fix the old hacks.

The question is whether MPLS can still be changed,
e.g. whether it is still possible to add type ECN fields.
However, the PWE list is not the best place to ask that question.


Y(J)S

_______________________________________________
pwe3 mailing list
pwe3@ietf.org
https://www1.ietf.org/mailman/listinfo/pwe3