Re: [mpls] Working Group Last Call on draft-ietf-mpls-summary-frr-rsvpte

Hi Alexander,

Thanks for your suggestions and comments. We discussed this further amongst authors. We think the PLR has to do a proactive check of the MTU on the bypass tunnel for proper FRR to work correctly at all times.
We will introduce an addition in section 3.3.1 of the I-D to describe this:

“When using procedures defined in this document, the PLR MUST ensure bypass tunnel assignment can satisfy the protected LSP MTU requirements post FRR. This is to avoid any packets from getting dropped due to exceeding the MTU size of the bypass tunnel after FRR.”

Let us know if this addresses your concerns.

Regards,
Tarek

From: Alexander Okonnikov <alexander.okonnikov@gmail.com>
Date: Tuesday, May 14, 2019 at 8:21 AM
To: "Mike Taillon (mtaillon)" <mtaillon@cisco.com>
Cc: Markus Jork <mjork=40128technology.com@dmarc.ietf.org>, "mpls@ietf.org" <mpls@ietf.org>, "draft-ietf-mpls-summary-frr-rsvpte@ietf.org" <draft-ietf-mpls-summary-frr-rsvpte@ietf.org>, "mpls-chairs@ietf.org" <mpls-chairs@ietf.org>
Subject: Re: [mpls] Working Group Last Call on draft-ietf-mpls-summary-frr-rsvpte
Resent-From: <alias-bounces@ietf.org>
Resent-To: <mtaillon@cisco.com>, Tarek Saad <tsaad@juniper.net>, Rakesh Gandhi <rgandhi@cisco.com>, <adeshmukh@juniper.net>, <mjork@128technology.com>, <vbeeram@juniper.net>
Resent-Date: Tuesday, May 14, 2019 at 8:21 AM

Hi Mike,

If we don't talk about gap in procedure of Path messages merging by MP, then I don't see other gaps in RFC 4090. But I see that gap with MTU adjusting appears with introduction of Summary FRR.

Current behavior is follow:

1) PLR sends backup LSP Path to MP with actual MTU in ADSPEC;
2) MP sends backup LSP Resv to PLR with actual MTU in FLOWSPEC (derived from ADSPEC);
3) PLR merges FLOWSPECs of protected and backup Resv messages (according to RFC 2210) and sends Resv upstream with actual MTU in FLOWSPEC.

New behavior with Summary FRR:

1) PLR sends SFRR-Active Path to MP;
2) MP does summary refresh for backup LSPs;
3) PLR sends Resv with unchanged MTU of protected LSP upstream, as before failure. This is because FLOWSPEC for backup LSP has not been updated, and resulting FLOWSPEC thus has not been updated.

I cannot agree that head-end not need to be aware about path MTU changes. If head-end is irrespective to actual path MTU, it could cause blackholing of packets that are larger than actual path MTU. Also, I'm not aware about implementations of PLR that take into account MTU in SENDER_TSPEC of protected LSPs and choose bypass tunnels accordingly. My knowledge that many of them agnostic to this.

Thank you.

13 мая 2019 г., в 20:38, Mike Taillon (mtaillon) <mtaillon@cisco.com<mailto:mtaillon@cisco.com>> написал(а):

Hi Alexander

By gap, I mean its not mentioned/covered at all.

To give alternate perspective, I disagree with trying to merge state of primary and backup LSP.
And would think it unncessary for either the MP or the headend to be aware, or make any MTU changes post FRR.

I do agree with your last statement where PLR should ensure the MTU of backup can accomdate the primary LSP MTU (plus any encap added to transport over the backup LSP).
And surely most implementations are already doing this, or FRR wouldn’t be that successful...

Do you not agree this issue/gap is out of scope of (ie. not specific to ) this document ?

-mike

On May 13, 2019, at 11:31 AM, Alexander Okonnikov <alexander.okonnikov@gmail.com<mailto:alexander.okonnikov@gmail.com>> wrote:

Hi Mike,

Do you mean the gap in merging of Path messages by MP in part of choosing ADSPEC of protected LSP rather than merging ADSPECs (choosing minimal MTU, particularly)? Oh, agree. Though if we would assume that MP did perform merging of ADSPEC (and probably other objects where applicable), there would be problem with signaling backup LSP MTU to MP after failure.

I agree that guaranting enough MTU size on all links in the network is good practice, but in reality it not always could be provided, or could be provided with significant penalty on manageability.

Per my understanding, reliable solution would be for head-end:

1) to specify minimum LSP MTU as a constraint (like resource affinities, BW, etc.) for CSPF and take link MTUs from TEDB into consideration, and

2) to signal to downstream LSRs minimum LSP MTU (by virtue of SENDER_TSPEC, like BW), such that PLRs would be able to make decision about availability of bypass tunnels, which can accomodate requested MTU.

Thank you!

9 мая 2019 г., в 15:28, Mike Taillon (mtaillon) <mtaillon@cisco.com<mailto:mtaillon@cisco.com>> написал(а):

Hi Alexander,

I beleive MTU handling post FRR is not covered in base RFC4090 and is therefore an existing gap.
It kinda defeats the purpose if headend needs to adjust MTU after FRR to prevent drops… and would presume most deployments assume that backup MTU can accomandate MTU of primary LSP (plus any added MP labels).

Issue deserves discussion, but think its out of scope from this document.

-mike

On May 7, 2019, at 4:54 PM, Alexander Okonnikov <alexander.okonnikov@gmail.com<mailto:alexander.okonnikov@gmail.com>> wrote:

Hi authors,

As far as Summary FRR LSPs are not being signaled via Path messages over bypass tunnel after failure, information on head-ends about actual path MTU of protected LSPs can be corrupted. For example, path MTU of protected LSP is 1500 bytes (provided that ADSPEC is used), and path MTU of bypass tunnel is, for example, 1500 bytes. As far as Path messages for protected LSPs are not being sent over bypass tunnel, MP will use ADSPEC received in Path messages of those protected LSPs previoulsy (before they have been rerouted onto bypass tunnel), i.e. 1500 bytes in place of 1496 bytes. To avoid this problem PLR would have to signal path MTU of its bypass tunnel in B-SFRR-Active object (alternatively, MP could inherit this value from ADSPEC of PSB of the bypass tunnel), and 2) MP would have to choose minimal of MTU values from ADSPEC objects while merging Summary FRR protected LSP. But, even in this case MP will have to generate trigger Path messages (with updated ADSPEC) for protected LSPs and then, after receiving Resv messages with updated FLOWSPEC, send them to PLR. I.e. summary refresh in MP->PLR direction with high probability will be inapplicable, due to trigger messages.

Thanks.

7 мая 2019 г., в 23:46, Markus Jork <mjork=40128technology.com@dmarc.ietf.org<mailto:mjork=40128technology.com@dmarc.ietf.org>> написал(а):

As a co-author,  I believe this document is ready for publication.
-Markus

On Apr 30, 2019, at 4:33 AM, N.Leymann@telekom.de<mailto:N.Leymann@telekom.de> wrote:

Working Group,

This mail initiates the two weeks working group last call on draft-ietf-mpls-summary-frr-rsvpte which is considered mature and ready for a final working group review.

Please read this document if you haven't read the most recent version yet, and send your comments to the mpls wg mailing list (mpls@ietf.org<mailto:mpls@ietf.org>), not later than 17th of May.

There is one IPR disclosure against draft-ietf-mpls-summary-frr-rsvpte

This working group last call ends May 17th, 2019 (there is at least in some countries a public holiday this week, therefore the call is a bit longer than usual).

Best regards

Nic

_______________________________________________
mpls mailing list
mpls@ietf.org<mailto:mpls@ietf.org>
https://www.ietf.org/mailman/listinfo/mpls<https://urldefense.proofpoint.com/v2/url?u=https-3A__www.ietf.org_mailman_listinfo_mpls&d=DwMFaQ&c=HAkYuh63rsuhr6Scbfh0UjBXeMK-ndb3voDTXcWzoCI&r=Kd6NW5ctLWr7GB646PVmMByvi8wQxPILpKhHvGQeRHY&m=8ugk0fj6GyOtsYCkrvM6x1jpwt7T7bGkBhQ9Le-SI6k&s=0zgSQxWhO9k94N9On89so_wtqNQ0mXruYDy6Dka6-Lc&e=>