The MPLS-OPS Archive

["Intae Jung" <yjim@lge.com>]

Hello,
I'd like to ask a couple of questions regarding a scalability of LDP,
especially when it is deployed in an ATM-LSR, adopting Downstream on
Demand distribution with conservative label retension and ordered control,
where labels (VPI/VCI) are a relatively scarce resource that must be conserved
in ATM switch. The VPI/VCI is a particularly valuable and limited resource that
should be saved when an ATM LSR is not able to support VC-merging capability.
Unfortunately the system I'm managing has no VC-merging capability.
As defined in the standard, LDP attempts to setup or release a hop-by-hop LSP
whenever a new route is added or deleted in an ATM LER through routing protocols,
mainly IGP such as OSPF, ISIS or RIP.
Let me assume a sample ATM-MPLS network as follows.

Network1 - R1 <-> LER1 <-> LSR1 <-> LSR2 <-> LER2 <-> R2 - Network2

<---IP Network---><------------MPLS Network---------><---IP Network--->

R1, R2: Router operating IGP (OSPF or RIP)
LER1, LER2: Edge ATM LSR operating LDP and IGP (iBGP if needed)
LSR1, LSR2: ATM LSR operating LDP and IGP (OSPF or IS-IS)

OSPF or RIP is used in the IP Network and OSPF or IS-IS is deployed
in the MPLS Network. An i-BGP session could be setup between LER1 and LER2
if needed, but not necessarily since LER1 and LER2 are not supposed to provide
BGP/MPLS functionality, in other words, they don't work as a PE(Provider Edge) router.
LER1, LER2, LSR1 and LSR2 support only LDP as an MPLS signaling protocol.

The problem is that the number of LSP's from LER1 to LER2 increases
as the number of networks increases in the Network2, meaning that the number of
routing entries in LER1 increases accordingly, wherease the number of LSPs supported
by LER1 is a lot smaller than the number of its route entries.
Eventually, as the number of LSPs setup exceeds the maximum number of LSP's,
we'll not be able to setup an LSP any more even though new routes are added.
As I mentioned above, LSR1 and LSR2 don't support VC-merging capability,
thus the scalability becomes quite servere.
Therefore w're trying to figure out how to support a lot of routing entries using
a limited number of LSP's. We are trying to come up with a mechanism that setups
just one LSP between LER1 and LER2, which could be shared between all best-effort
traffic between Network1 and Networks2.

The problem we faced is how to setup only 1 LSPs between LERs automatically,
and then how to map a lot of route entries to the shared LSP.
Although two LER's are shown in the example, the number of LSP's between LER's
can be quite large as the number of LER's increases.
We're considering to setup a full-mesh iBGP sessions between all LER's, and
to use BGP next hop gateway in order to get an FEC-to-NHFLE mapping.
In other words, routes from Network2 will be redistributed to LER1 via iBGP session
between LER1 and LER2, and then LER1 will setup just one LSP for the BGP next hop
address and map all routes from Network1 to the LSP.
However it enforces us to setup iBGP sessions between all of LER's even though
we don't need them unless we deploy MPLS. We're also wondering whether
we'll be able to solve the problem using any IGP capability, like tag field of AS-external
LSA in OSPF.

Please enlighten us on this issue.
Any experience or idea will be greatly appreciated.
Thanks in advance.

Regards,

Jung.