Load balancing over LSPs established by LDP

[yjim0140]

Hi there,

As you know, IGP like OSPF supports a load balancing between equal-cost

multi-paths, and LDP tries to establish an LSP every time a new route is added by OSPF. By the way, I'm wondering whether LDP would setup multiple LSPs for the equal-cost multi-paths to the same destination and do load-sharing between multiple LSPs established by LDP. In summary, my question is whether LDP, not CR-LDP nor RSVP-TE, supports a load-balancing between multiple LSPs.

Thanks in advance.

Regards,

Yongjun Im.

[yjim0140]

I'm assuming a load-balancing in an ATM-LER, meaning an end-to-end load-balancing across more than 2 LSPs for the same prefix between ATM-LERs, since an ATM-LSR has only cell-switching capability.

[mazhar71]

Hi ,

I really couldn't understand your question well but I will try to unswer as I can.

On frame mode each router has an LDP ID. Label space is global label space. So incoming interface is not so important due to global label space. therefore only one LDP sesion is enough between 2 frame mode router.

On cell mode label space is interface wide label space. So incoming interface is important for forwarding. Therefore different LDP sessions are built between each cell mode interfaces.

Regards

Mazhar

[yjim0140]

You understand my question correctly. What I'm saying is whether more than 2 LSPs, for the same prefix, which are established by corresponding LDP sessions on cell-mode interfaces, is able to do load-balancing over those LSPs, especially in a cell-mode MPLS router using distributed CEF.

Thanks in advance.

Yongjun.

[melkomy]

As far as I know that it's possible to load balance over multiple LSPs

EX

D -----MGX1--3---LSC---1---MGX2---S

So in D to S direction.......the RPM in MGX1 will have 3 LVCs to the S destination.

In S to D direction.....S will have only 1 LVC to the D direction....Hence in S to D direction,that single LVC will only ride on one of the 3 links bet LSC and MGX1

in D to S direction,when MGX1 learns the S prefix......it'll make a label binding request for that prefix over each LC-ATM interface (3 links hence 3 requests).

The LSC will likewise make 3 requests for that prefix......MGX2 reply to LSC and LSC reply to MGX 1.

When all this is done,MGX1 will have 3 LVCs for S and load balance on these 3 LVCs as long as the RPm in MGX 1 has 3 IGP adjancies over these links.

Regards,

Komy

[yjim0140]

Thanks a lot.

If 3 LVCs have been established on the MGX1, how can it evenly distribute the packets, with the identical destination prefixes, i.e., destined to the same destination S, over 3 LVCs?

Regards,

Yongjun

[melkomy]

As I told you,there'll be 3 LVCs (with 3 VPI/VCI) to the LSC and CEF will load balance bet them as CEF look up in its FIB table and find 3 equal cost paths to the same destination.

Another issue to keep in mind is whether you've the VC merge feature enabled on your BPX or not.

With NO VC merge you'll have 3 VCs from D and 3 to S but with VC merge you'll have the 3 VCIs from D mspped to one VCI to S.....so you don't need the VC merge feature for load balancing ...it's only to save VCs

Regards,

Komy

[k.anagnopoulos]

Although a bit irrelevant to the discussion ,an important issue to keep in mind at least with BPX,is that you cannot control over which BPX trunk(if more than one)the LVCs will get nailed.This happens when all trunks are controled by a single partition(one LSC).

In this case load sharing is difficult when multiple LVCs exist for the same destination.

Unless using TE (possible now in cell mode),the only trick i'm aware of for equally distributing LVCs over trunks is "MPLS redundancy for IP+ATM networks",but you are limited to two trunks(two LSCs).

Have you experienced any such problem?

Regards,

Kostas