********************************************************************** IETF 100 PALS - Monday, 13 November 2017 - 17:40-18:40 Room: Orchard 35/60 min allocated; ** Please note the slot placement may be adjusted.) ********************************************************************** Chairs: Stewart Bryant and Andy Malis Secretary: David Sinicrope (x = slide sets NOT received as of 12 November 2017 17:00 Singapore time) 1. 15 min - Agenda bash, WG Agenda and Status - Andy MALIS and Stewart BRYANT Andy went through the posted slides. It was noted in slide 5 the MD5 no longer meets security requirements. There will be work in the MPLS WG on a new security mechanism. It was noted this is not just an LDP problem but a problem for all TCP based protocols. 1. there is no default mechanism and 2. no negotiation mechanism. The Rtg Area should work with Transport and Security Areas to resolve. 2. 20 min - Use of Ethernet Control Word RECOMMENDED - Stewart BRYANT https://datatracker.ietf.org/doc/draft-bryant-pals-ethernet-cw Objective: Calling out potential misordering issue with sending Ethernet packets in PWs with out the control word. Stewart went through the posted slides. Himanshu: - what are they doing wrong? Stewart: some vendors have a switch that causes the router to look down the packet and if it sees a 00 the implementation skips x bytes down the header then does a 5-tuple has for load balancing. (it assumes that the pseudowire type is Ethernet). Some operators have turned on these on either by not understanding or by evolution of the network that then causes problems. They have asked that we publish recommendations on how to address the problem. Ignas: had 44 different discussions with operators and only 2 operators were aware of the problem. Need to raise awareness of the issue. The group will do live editing of the two recommendations in the draft. Section 4 and Section 7. (Note: the text resulting from the meeting will be posted in a subsequent revision of the draft.) Loa: the recommendations talk about quite a bit of work to be done in the MPLS WG. Can you give the MPLS WG a heads up? Stewart: yes, how much is needed Loa: need to make sure that people understand what problem we are solving Stewart: OK will talk offline about the details Himanshu: Regarding section 7: This doesn't look right. LSRs are doing the ECMP but LSRs can't recognize the FAT PW. Stewart: The operator would need to understand and discuss how to resolve the issue. Himanshu: but the proposed paragraph does not look correct. How does the P router decide whether to enable or disable and whether to enable ECMP. Greg Mirsky: THe paragraph is basically saying dont do it by default and if you do, then think about it before you do it. Stewart: If you need to do ECMP, then try to use other techniques. If you can't do the other techniques, then analyse the situation. Adrian: Is the intention only to talk about ECMP only at egress PEs? Stewart: no, the problem is broader. Some carriers wanted control word sequence number checking. Adrian: is this intended to work on P routers doing ECMP? Stewart: yes Adrian: Prouters do ECMP in a wide variety of ways and is it better to describe the desired behavior rather than make recommendations Stewart: trying to give guidance John Drake: The text does not mention P routers and is not enforceable. *** Andy displayed text for the group to edit the proposed text. Andy recorded the changes. The result will be posted in the next revision of the draft. The result will be further vetted on the PALS list. *** Section 4 and Section 7 on use of the CW was accepted as proposed by the room. Section 4 and Section 7 text on use of ECMP was discussed. Adrian: concerned the solution is too prescriptive. A solution should decribe what should happen in the network and let implementors implement accordingly. Himanshu: To identify a micro flow in a PW one can use the FAT or ELI, should make the context clear. Stewart: context is clear, Ethernet PW. Himanshu: but needs to be clearer and let the implementor decide. Matthew: It could be clearer that the ELI and FAT are to identify microflows Italo: do these recommendations work for MS PW? Stewart: yes could use FAT for the PW Greg: If ELI is above the PW, it could be used for the next segment. It is up to the implementation. Matthew: an SPE would not pass an ELI/EL across. Stewart: can't assume the EL stays. FAT is the safer way since it is below the PW label. Loa: if you are looking though label stack for the ELI do you know if you are above or below the FAT label? You don't know. Stewart: the FAT label is the Bottom of Stack and immediately follows the PW label. Italo: don't recommend one or the other, just note that one should be used. Matthew: not making assumptions about hash on EL. Just need to say that if ELI is used then ELI should be used on every segment. Onto section 7 Matthew: not sure this should be in the draft, this is talking about P routers so should be in the draft in the MPLS WG. Andrw Dolcanow: agree with Matthew Stewart: Where the DPI problem came up is in the context of PWs. Andrew: why are we only talking about DPI, there could be other hashing or loadbalancing that this impacts. Stewart: someone on the list noted that we need to do more than what the draft says, e.g., should use sequence numbers Matthew: if there is load balancing going on in the network how do we design PW so that they can handle it? Can't say how to design LSRs Stewart: but the problem is with PWs, should we not warn operators? Matthew: no, but should be more general. Andy: But the feature assumes an Eth PW Matthew: But the issue occurs on P routers and is outside of PWs. Stewart: we should at least say this is an interesting feature but should be accessed before use. Himanshu: take to MPLS group since largely in scope of LSRs not PW. MPLS WG should say what the right behavior is at the LSR. Loa: so operational considerations section, no problem with PALS describing a behavior related to PWs. However, the solution should possibly be in one of the MPLS drafts. Need more description brought to MPLS WG. Andrew: saying that the CW and getting into the DPI, is naive. Should simply say that using DPI should be considered carefully before deployed. Stewart: what to do from here. Andy: the proposed text for section 7 should be replaced by text that describes the problem. In section 4, should it be capital MUST/SHOULD or must/should Dave: given it is deployment guidance and not implementation requirement it should be lower case. John: If the PEs are seeing misordered packets, then only choice is to use seq numbers Stewart: Not get SP to turn off DPI based loadbalancing? Himanshu: but not likely to get them to do it, so sequence numbers are only alternative. Stewart: will update the text and post to the list for comments. 3. (added during agenda bash) Liaison telling liaison partners about the changes in the control word draft: - Broadband Forum (for action) - ITU-T SG15 WP3 (for action) - IEEE 802.1, RAC (for information and review) - MEF (for action) o briefly state the issue o point to the recommendation and clarification (i.e., the draft) o ask that the new draft/RFC be considered and taken into account in previous and future works. Andy closed the meeting at 19:01pm ********************************************************************** Overflow (Will be presented if time permits.) ********************************************************************** xx. - None currently ********************************************************************** REMOTE INFORMATION FOR THE PALS SESSION(S) ********************************************************************** Remote Participation Info: http://www.ietf.org/meeting/100/remote-participation.html - No WebEx - IETF 100 Agenda with Audio and Jabber links: https://tools.ietf.org/agenda/100/