The MPLS WG Archive

[Robert Raszuk <rraszuk@cisco.com>]

Jeff,

See below ..

> Jeff Bush wrote:
> 
> Hello:
> 
> My colleagues and I have some questions on on MPLS that are more of a
> generic nature - as we are getting more knowledgeable about the
> subject.  Your comments would be most appreciated. We apologize if some
> topics have already been discussed on the list.
> 
> 1.  Is there any issue with MTU conflicts as shims are added to the
> original packet?

Sure. You need to make sure that your other network devices for example
LAN switches can handle large frames when you add MPLS header to already
1500 byte packet. Otherwise you need to fragment.

> 2.  Is it true (as some ISP folks have stated) that MPLS VPNs are not
> scalable?

No. Those who say this simply don't like the concept of prefix based
MPLS. The facts are that almost significant ISPs & SPs and Europe are
deploying mpls-vpns, Asia & Australia following. In US the first focus
was given towards TE with only a few ISPs deploying MPLS-VPNs today.
 
> 3.  Is MPLS currently providing traffic engineering only - using TOS
> bits.  Are there deployments out there that provide QoS (voice over
> data) using MPLS (without ATM, say)?

DiffServ is orthogonal to MPLS. MPLS can enhance the qos capabilities
but not substitute for the lack of basic IP qos. Today most of the
production environments is using E-LSPs /coping or writing directly 3
bits in EXP field in the MPLS header/ for packet prioritization in the
network.

> 4. Does LDP create a huge state machine and potentially clog the network
> (especially if the one or more LSPs are disturbed or disconnected)?

LDP state machine is defined here: draft-ietf-mpls-ldp-state-03.txt You
can judge if it's big or not. As far as cloging the network - I would
say is implementation dependent - but remember this is peer-to-peer
protocol.
 
> 5. If there are more than one LDP used in a hybrid environment, could
> there be conflicts regarding LSPs?

I don't understand your question. LDP process is one global per router
to distribute label bindings to it's peers.
 
> 6. Does variable length header processing cause high latency?

Without going into modern high speed lookup processors discussion the
short answer is no.

> 7. How does MPLS gets deployed in the "last mile" - between the egress
> router and the end-system - say on the LAN?

That really depends what is this "last mile". I don't really think that
MPSL to the end-system (since you brought LAN example) probably you mean
end-workstations is at all needed.
 
> 9. Say you have a path (LSP) established between four LSRs, A, B, C, and
> D.  Then when packets are being transmitted on this LSP, C fails.  What
> happens then?  We would be trying to establish another LSP by sending
> LDP packets to D from B.  Or should this be A which established the LSP
> first?  And, what happens to the packets in transit?

Are you asking about: 

A. Prefix based LSPs (established with IP routing plane + LDP)
B  TE LSPs ?

If A:

If there are other paths in the network let's say B - D then after the
IGP convergence B will send packets with a new label to D by itself. 

If B: 

If this is TE LSP then you again have two choices: Using FRR or not.
Anyhow locally protected or not at some point A will need to create new
LSP path to bypass C.

> 10. Is it true that MPLS is good for the network backbone only?

MPLS is good where it is needed and where is used wisely.

R.