Re: [PWE3] PWE3 question

[Eric Rosen <erosen@cisco.com>]

Neil> I also don't buy the N^2/2 reduction argument either.....if A wants to
Neil> talk to B 'something' has to be configured p2p to allow this

I guess you mean "if A wants to  talk to B over a pseudowire ...".  If layer
3 connectivity is all that  is required, then there's this newfangled notion
of connectionless  networking that eliminates  the p2p stuff.   Nah, that'll
never catch on ;-)

Assuming  the context  here is  pseudowires, then  for a  full mesh  among N
endpoints  of course  one  needs N**2  PWs.   However, there  is no  O(N**2)
scaling problem in the network because  (a) any given PW endpoint only has N
PWs, and (b)  in the core of the  network one only needs state  for one mp2p
LSP per endpoint.  So the endpoints scale O(N) and the core scales O(N).

It is  true on  the other hand,  that the  PW provisioning system  scales as
O(N**2).  Not  much to do about that,  layer 2 just doesn't  scale nearly as
well as layer  3; that's an inherent  problem of layer 2 which  can never be
eliminated by NMS systems or even by VPLS-like schemes. 

Neil> mp2p LSPs simply push the problem up a layer, they don't get rid of it
Neil> (and the BGP4 LSPs in rfc2457 do exactly this function). 

The  BGP-distributed  labels of  RFC2547  do  not  create p2p  anything.  In
RFC2547, mp2p LSPs based on  BGP-distributed labels are carried in mp2p LSPs
based on LDP-distributed labels.  (That's why it's a layer 3 VPN service.)











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