The MPLS-OPS Archive

[Christopher Lewis <chrlewis@cisco.com>]

One nit-pick in-line.

At 12:02 PM 8/5/2002, Sachin Kalra wrote:
> Andy:
>
> >  
> > 1. If a packet is received by a PE and it does
> > not match any VRF, is the packet dropped? (my interpretation of the spec
> > is yes)
>
> This is implementation specific. There are two possibilities:
> A> The packet may be dropped if it does not match any VRF entry
> B> The packet may be destined to some server outside the VPN (let's
> say Yahoo server) then in this case, after exhausting all VRF entries,
> the longest match process can be done using regular IP routing table and
> route can be found by PE. 
>
> >  
> > 2. If a valid packet is received, the VRF
> > lookup is done at the PE, the label may be popped, is the longest match
> > process used to look up the IP address in forwarding to the ifc? (my
> > interpretation is yes)  If so, I am not clear on the application of
> > Penultimate Hop Popping in the VPN world.  Is this under
> > consideration in the working group?  Is there something already done
> > on this?
>
> What I understand is that you are talking about Egress PE.  No your
> interpretation is not correct.
>
> At Egress PE label popping, is similar to Penultimate Hop Popping, as
> while popping the last label from the stack Egress PE would know the
> outgoing interface and thus "longest match process used to look to
> the IP address..." is not required. 

Unless it is an aggregate label. Aggregate labels are used to identify
subnets, for example the subnet for the PE to CE link uses an aggregate
label and the PE needs to perform an IP lookup to determine which end of
the p2p link the packet is destined for. THis is not normally a big issue
as not much traffic is destined for that link. It is more of an issue if
the CE to PE link is a multi access network like ethernet, then the
additional lookup is necessary for all customer packets as the label to
get the packet to the right outgoing interface only identifies the
ethernet subnet, not each host on that ethernet.

Chris


> >  
> > 3. Can multiple VRF's be assigned to a single
> > ifc? (my interpretation is yes)
>
> Yes you are correct, it can be done. Consider the following case.
>
>                                 
> [Site A (VRF A)]
>                                
> /
> [PE]---------------[Switch]
>                                
> \
>                                 
> [Site B (VRF B)]
>
> In this case interface from PE to switch can be assigned two VRFs A and
> B.
>
> Regards,
> Sachin Kalra
>
>
>
> At 11:00 AM 8/5/02 -0500, Andrew Walding wrote:
> > MPLS Ops
> > Experts,
> >  
> > I have a couple of key questions:
> >  
> > 1. If a packet is received by a PE and it does
> > not match any VRF, is the packet dropped? (my interpretation of the spec
> > is yes)
> >  
> > 2. If a valid packet is received, the VRF
> > lookup is done at the PE, the label may be popped, is the longest match
> > process used to look up the IP address in forwarding to the ifc? (my
> > interpretation is yes)  If so, I am not clear on the application of
> > Penultimate Hop Popping in the VPN world.  Is this under
> > consideration in the working group?  Is there something already done
> > on this?
> >  
> > 3. Can multiple VRF's be assigned to a single
> > ifc? (my interpretation is yes)
> >  
> > Thanks in advance!
> >
> > Best Regards,
> >
> > Andy Walding
> > CellStream Inc. [
> > http://www.cellstream.com
> > ]