[MPLS] LDP GR query

Hello Lixing Wang,

As I had mentioned earlier, when there is a link failure (in this case, it is identified using BFD), the LDP session is removed and hence RTA and RTB will delete the lsp and reestablish the session. BFD can provide failure detection on any kind of path between systems, including direct physical links, virtual circuits, tunnels, MPLS LSPs, multihop routed paths, and unidirectional links (so long as there is some return path, of course.) So here even though the link failure is not occuring on the connected interface, BFD can identify the link failure.

Now, coming to the case where the 2 LSR are helper router (this is identified during LDP Session establishment where Reconnect timer is set to zero), then the moment LSR comes to know (usually by hello hold time expire) its neighbor is down, LDP session is brought down and it will not enter GR mode.

Regards
Ramesh Huliyar

----- Original Message ----
From: "little_star@huawei.com" <little_star@huawei.com>
To: Ramesh Huliyar <ramhuliyar@yahoo.co.in>; Yu Yi <yuyi@nortel.com>; mpls@lists.ietf.org
Sent: Tuesday, 17 October, 2006 12:43:04 PM
Subject: Re: [MPLS] LDP GR query

Hi Ramesh,

Please see the below graph

If the link between the two LAN-SWITCH is down ,RTA and RTB will think the other side LSR restart,so these two LSR are all helper router.Thay will perform GR flow.Do you think it's good than the non-GR flow(the normal flow which session down).

I don't think so ,we should distinguish the LSR whether restart or not.If the LSR really restart,the GR flow is OK,but if not,I think,the LSR should delete the lsp and reestablish the session.

Lixing Wang

----- Original Message -----

From: Ramesh Huliyar

To: little_star@huawei.com ; Yu Yi ; mpls@lists.ietf.org

Sent: Tuesday, October 17, 2006 12:52 PM

Subject: Re: [MPLS] LDP GR query

Hello Lixing Wang,

What I understand from your problem statement is that, "if there is a change in the lsp (deleted, next hop different) established previous to the GR, it will delay the updation of lsp or basically convergence". If the next hop for the FEC changes, label advertised from the current next hop will be chosen (lable retention mode). Now to speed up the convergence in case of deleted FEC, I think we should have a mechanism in LDP control plane, where it can signal that the process of signalling FEC-to-Label is completed. (Similar to the one in BGP, where we can signal the end of UPDATE message)

Regards
Ramesh Huliyar

----- Original Message ----
From: "little_star@huawei.com" <little_star@huawei.com>
To: Yu Yi <yuyi@nortel.com>; Ramesh Huliyar <ramhuliyar@yahoo.co.in>; mpls@lists.ietf.org
Sent: Tuesday, 17 October, 2006 8:13:00 AM
Subject: Re: [MPLS] LDP GR query

Hi Yu Yi,

When two LSR fall into GR flow,then some LSP will be updated only after GR,such as the lsp which should be deleted.

I think GR flow will delay the updating of LSP,so it's better don't perform GR flow in mistake.

Lixing Wang

----- Original Message -----

From: Yu Yi

To: little_star@huawei.com ; Ramesh Huliyar ; mpls@lists.ietf.org

Sent: Tuesday, October 17, 2006 9:33 AM

Subject: RE: [MPLS] LDP GR query

I think for this scenario RTA and RTB both consider the peer restart and perform the GR process. So RTA will re-send all mapping and RTB also do so. and then after receiving mapping all the LSP's stale flag are cleared and a perfect GR is finished.

The traffic won't be lost if the forwarding plane is ok.

From: little_star@huawei.com [mailto:little_star@huawei.com]
Sent: 2006$BG/(B10$B7n(B17$BF|(B 8:50 AM
To: Ramesh Huliyar; Yi, Yu (SUNP:1G11); mpls@lists.ietf.org
Subject: Re: [MPLS] LDP GR query

Hello,

Then how about another scenario?

Another example that the LDP session will be down when the LSR$B!G(Bs neighbor can$B!G(Bt send out it$B!G(Bs packet, including hello packet or keepalive packet, if the control plane of it$B!G(Bs neighbor is very busy or occurs some error ,then the LDP session will be down. In this case this two LSR will conclude that it$B!G(Bs neighbor restart and the LSR will implement GR flow.

I mean the LDP protocol has some problem.In this case the link is ok,so the BFD can't be used.

When they re-establish the session,they will perform GR flow together.

----- Original Message -----

From: Ramesh Huliyar

To: Yu Yi ; little_star@huawei.com ; mpls@lists.ietf.org

Sent: Monday, October 16, 2006 8:36 PM

Subject: Re: [MPLS] LDP GR query

Hello,

I feel that RTB LDP session will go down with RTA immidietly when the link goes down. After the re-establishment of the LDP session, RTB will be sending FT session TLV with Recovery Time 0. This will cause the RTA to remove the stale FEC-to-label mapping. But until this happens, traffic will be black holed.

To overcome this, other methods of identifying link failures in RTA have to be employed, such as BFD, this will help RTA to use alternate FEC-to-Label mapping, because link failure does not start LDP GR but it will bring down LDP session in case of link level LDP.

Regards
Ramesh Huliyar

----- Original Message ----
From: Yu Yi <yuyi@nortel.com>
To: little_star@huawei.com; mpls@lists.ietf.org
Sent: Monday, 16 October, 2006 2:51:33 PM
Subject: RE: [MPLS] LDP GR query

For your case it is fine. RTB should be in normal case(normal session down)(no any GR flow) and delete all LSPs which are established on the port. After session up again RTA will re-send the same label mapping which are sent before session down. But as you know during this period the MPLS traffic are lost. So this is not a complete GR case.

  GR is applicable only if the forwarding plane is fine.

From: little_star@huawei.com [mailto:little_star@huawei.com]
Sent: 2006$BG/(B10$B7n(B16$BF|(B 3:07 PM
To: mpls@lists.ietf.org
Subject: [MPLS] LDP GR query

Hi

I have query related to LDP Graceful Restart.

RFC 3478 say

When an LSR detects that its LDP session with a neighbor went down,

   and the LSR knows that the neighbor is capable of preserving its MPLS

   forwarding state across the restart (as was indicated by the FT

   Session TLV in the Initialization message received from the

   neighbor), the LSR retains the label-FEC bindings received via that

   session (rather than discarding the bindings), but marks them as

   "stale".

RFC 3478 say that a router will implement GR flow when it detect that its LDP session with a neighbor went down.

But there are many cases which can cause the LDP session went down ,and it$B!G(Bs neighbor don$B!G(Bt restart actually.

Consider this scenario:

Suppose that the link between RTB and LAN-SWITCH is broken down, then RTA will detects that its LDP session with RTB went down, then RTA will implement GR flow, even the LDP session between RTA and RTB is re-established after the link between RTB and LAN-SWITCH is resumed. Because the FT TLV in the initialization message of RTB don$B!G(Bt indicate that RTB whether reboot or not, so RTA is still implement GR flow.

Now RTA is in GR flow , but RTB isn$B!G(Bt in GR flow.

But what will happen if RTA is in GR flow , at the same time RTB isn$B!G(Bt in GR flow? I think it will fall into confusion.

Another example that the LDP session will be down is that the LSR$B!G(Bs neighbor can$B!G(Bt send out it$B!G(Bs packet, including hello packet or keepalive packet, if the control plane of it$B!G(Bs neighbor is very busy or occurs some error ,then the LDP session will be down. In this case the LSR will conclude that it$B!G(Bs neighbor restart and the LSR will implement GR flow.

_______________________________________________
mpls mailing list
mpls@lists.ietf.org
https://www1.ietf.org/mailman/listinfo/mpls

The MPLS WG Archive

[MPLS] LDP GR query