From: Vikram Dadlaney (vdadlaney@gmail.com)
Date: Mon Aug 21 2006 - 12:56:48 ART
>Am I close to being accurate in this description?
Per my understanding you are correct on all counts in your last post.
We are talking frame mode over here. In frame mode a router keeps all the
labels it receives from all its neighbors whether upstream or downstream in
the LIB (Label Information Base). However it will only use the label towards
its downstream neighbor which will be in the LFIB(Label Forwarding
Information Base) table. This let's it converge faster in case its
downstream neighbor decides to take a break (it fails :-) ).
Label 10 will never be used if the next hop per rtr-2 remains rtr-3. However
should rtr-3 disappear and you have another link via rtr-1 connected to
network d than any packet generated by a host on the subnet connected to
rtr-2 will have to go via rtr-1 and that's when rtr-2 would use that label
10. In frame mode labels are generated per platform hence the label remains
the same.
>>IOW, why doesn't rtr-2 just advertise the label it learned from its
>>downstream neighbor, rtr-3 to its upstream neighbor, rtr-1? If it did
that,
>>then wouldn't traffic flow faster since rtr-2 wouldn't have to actually
swap
>>labels for packets going to net-D?
I don't believe that it has anything to do with traffic flowing faster. MPLS
is just another forwarding technology. In traditional Layer 3 forwarding a
Router has to look up the Layer 3 header information to forward the packet
at each hop (extract the information from the header) but in MPLS it has to
do that only once. Than that header is mapped to a label and the packet is
forwarded based on that label.
Hope that clears some doubts and as always if there are some errors than
appreciate it if others could correct me. Thanks
On 8/21/06, Tim <ccie2be@nyc.rr.com> wrote:
>
> Hey Chris,
>
>
>
> Thanks for bearing with me as I try to absorb what's going on here.
>
>
>
> I need to think about the first part of what you wrote some more but let
> me
> confirm I understand the 2nd part regarding the 3 labels for net-D on
> rtr-2.
>
>
>
> To stay in sync, I'll repeat the network diagram here:
>
>
>
> net-A rtr-1 net-B rtr-2 net-C rtr-3 net-D
>
>
>
>
>
> rtr-1 generates a label, let's say its = 10, for net-D and advertises it
> to
> rtr-2.
>
>
>
> rtr-2 generates a label, = 20, for net-D and advertises it to rtr-1 and
> rtr-3.
>
>
>
> rtr-3 generates a label, = 30, for net-D and advertises it to rtr-2.
>
>
>
> So, now rtr-2 knows of 3 labels for net-D.
>
>
>
> When a packet for net-D arrives on rtr-1, rtr-1 will use the label 20 it
> learned from rtr-2 to send the packet to rtr-2, correct?
>
>
>
> And, once the labeled packet gets to rtr-2, rtr-2 will then use the label
> it
> learned from rtr-3 (30) to send the labeled packet to rtr-3, correct?
>
>
>
> So, in this example, label 10 will never be used, right?
>
>
>
> However, if rtr-2 for some reason ever did need to send traffic for net-D
> to
> rtr-1 ( assuming a different topology), it would then use label=10 to send
> traffic to rtr-1.
>
>
>
> Am I close to being accurate in this description?
>
>
>
> Thanks again, Tim
>
>
>
>
>
> _____
>
> From: Chris Broadway [mailto:midatlanticnet@gmail.com]
> Sent: Monday, August 21, 2006 9:52 AM
> To: Tim
> Subject: Re: MPLS newbie w/ a couple basic Q's
>
>
>
> To answer the first portion, you have to look at the whole picture and the
> current method of MPLS deployment. You would have your LSRs running MPLS
> in
> the MPLS domain and either OSPF or ISIS be the IGP for this domain. This
> is
> on your diagram as rtr1,rtr2, and rtr3. Network "D" is also important
> and
> contributes to the operation and how things get labeled. Consider rtr3
> the
> "edge" of the MPLS domain (LER). It would have a BGP relationship to the
> gateway router for Network "D". This would make rtr 3 aware of the BGP
> table from Network D. YOU DO NOT WANT THE CORE LSRs TO HAVE THIS
> KNOWLEDGE.
> Since your diagram is only three routers, rtr2 would be the "CORE" LSR,
> simply because it is not a LER. All LERs would have an IBGP peering to
> each
> other. This would make all LERs aware of Network D. This means rtr1
> would
> now know how to get to Network D, is through rtr3. But how does he get to
> rtr3? This is where the IGP is used. The IGP is used only as the
> transport
> for the BGP. Even if there were 100 routes in network D, rtr2 would only
> see the IGP to rtr1 and 3 in its routing table. The BGP table of the LERs
> is what would show all the routes. With all this said, I have not
> discussed
> layer2 or 3 VPN and VRFs used with MPLS. But it would generally still
> follow this principle.
>
>
>
> The second portion:
>
> the router knows of the label coming from rtr1 and rtr3. Also the router
> knows of the label it gives for traffic going to rtr1 and 3. The
> operation
> from rtr2 would sound something like this:
>
>
>
> "I see signaling coming in on label 30...I know anything from label 30 is
> from rtr1. The signaling is trying to get to rtr3...I know I have to swap
> label 30 with 35 for signaling going to rtr3"
>
>
>
> Then the opposite happens for the reverse traffic. The important thing
> here
> is that all that is set up before the actual traffic flow. The MPLS
> signaling (LDP or RSVP) is what determines the label paths and sets up the
> binding database. If the signaling does not happen, MPLS cannot work.
>
>
>
>
>
> -Broadway
>
> _______________________________________________________________________
> Subscription information may be found at:
> http://www.groupstudy.com/list/CCIELab.html
This archive was generated by hypermail 2.1.4 : Fri Sep 01 2006 - 15:41:58 ART