[00:32:25] weihongbo joins the room [09:16:34] weihongbo leaves the room: I'm happy Miranda IM user. Get it at http://miranda-im.org/. [16:08:34] Shane Amante joins the room [16:08:40] Shane Amante leaves the room [20:05:46] YJS joins the room [20:05:56] Stewart Bryant joins the room [20:06:21] Andy Malis just introduced himself as new chair [20:06:31] round of applause for Stewart's 7 years as WG chair [20:06:53] Matthew preparing slides [20:07:15] Eric Gray - agenda bashing - can move one of Friday presentations to today ? [20:07:59] (LDP Extensions for MPLS-TP PW OAM configuration ) [20:08:40] Stewart also wants the TLV draft to be discussed today [20:09:08] WG Agenda and Status - Matthew [20:09:18] Stewart has joined IESG as routing AD, so stepped down as WG Chair here [20:09:28] Andy Malis is new chair [20:09:36] WG moved to RTG with Stewart as responsible AD [20:09:41] documents : [20:09:48] No new RFC since last IETF [20:09:57] draft-ietf-pwe3-mpls-transport is in RFC Ed Queue [20:10:04] hold due to bfd-base cleared but still waiting on rsvp-te-no-php-oob-mapping [20:10:11] draft-ietf-pwe3-vccv-bfd's hold due to bfd-base is now cleared [20:10:21] draft-ietf-pwe3-cep-mib has been stuck 364 days in "requires new ID" state - chairs need to address [20:10:33] draft-ietf-pwe3-segmented-pw needs revised ID (to address discusses) [20:10:58] Luca at mike - resolved most of issues [20:11:17] except 2 - [20:12:15] pathtrace verification needs some text, and congestion issue [20:12:29] draft-ietf-pwe3-oam-msg-map was heavily revised by Yaakov [20:12:38] It has an IPR statement from Cisco [20:12:52] draft-ietf-pwe3-mpls-eth-oam-iwk was recently updated - when will it be ready for LC > [20:12:55] draft-ietf-pwe3-congestion-frmwk-02 recently expired - need advice from transport advisor [20:13:05] PW redundancy drafts are being updated. should be ready for WG LC soon [20:13:38] MPLS and Ethernet OAM Interworking - Nabil [20:14:37] quick update - objective is to define failure detection and notification procedures between Ethernet PWs and associated native Ethernet ACs [20:15:13] main changes (none of authors was present at previous meeting) [20:15:26] Ray (Huawei) added as co-author [20:15:36] aligned defect state terminology [20:15:48] added CCM interface status TLV [20:16:07] need to address comments from Greg M (on list) [20:16:19] also some editorial issues [20:16:32] need afterwards to issue an update and go to WG LC [20:17:32] question from mike - in oam msg map draft there are two modes - are there here ? [20:18:12] Nabil - this draft only on Ethernet services, but similar [20:19:10] Wim - interface up and down events not standard [20:19:47] YJS - PW msg mapping [20:20:24] MB - who has read latest version of msg mapping - niot too many [20:21:00] (that was eth msg mapping - yjs talking about main mapping draft -12) [20:22:29] Stewart asked for more work - Monique project lead fixing the text - YJS then asked my MM to make the documnet co-herent [20:22:34] YJS agreed [20:23:27] From a tech pov identical, but major re-wite to make it clearer [20:25:03] YJS defines criteria for how oam is handled [20:25:14] and what pw types are covered [20:25:31] Addresses main changes [20:25:46] Added expn ttx [20:26:07] Detecf/fault/failure cleaned up [20:26:28] Editorial issues [20:26:47] Now meets IDnits !!! [20:26:59] Need another WGLC [20:27:15] ~Due to the extent of the changes [20:27:43] LC on the changes, not on the tech unless there are errors [20:31:55] satoru.matsushima joins the room [20:32:03] satoru.matsushima leaves the room [20:32:30] Discussion on defn of defect and fault - and yes we do use it differently from ITU [20:33:50] Stewart - packet PW [20:33:59] discussing since we have changed the mechanism [20:34:51] 4 mechanisms were decided for PD - CW, label, parallel PWs, virtual Ethernet [20:35:26] ver 0 proposed PID in CW, but this would change the PW (and make it longer) [20:35:28] martini joins the room [20:35:42] using a separate label is an extension of normal MPLS [20:36:17] however, although it complies with MPLS, HW may break due to assumption of single label [20:36:24] certainly breaks present chip sets [20:36:38] third method is merge of pid label and PW label [20:36:53] the parallel PW approach - one PW per protocol, and bind in control plane [20:37:21] this is efficient (only 4 needed), however, it breaks fate sharing [20:37:47] also would need to choose between OAM per PW (and combine) and single OAM for all [20:38:08] that would ne a nightmare to configure [20:38:27] the final approach (an old one) was virtual Ethernet - p2p Eth as carrier [20:38:49] works with PWs as presently defined [20:39:24] problem is that the MTU is larger (Eth header) but if jumbo frames are available this is not a problem [20:39:47] so it is proposed to reject parallel PWs because of fate sharing problem [20:40:10] PID breaks impleicit beahvior nad lable stack assumptions [20:40:17] except that those protocols do not fate share anyhow .... [20:40:22] PID in CW needs new HW so reject this [20:40:39] so the best approach is the virtual Eth approach [20:40:59] Stewart recommends making this a BCP (no new normative protocol needed) [20:41:07] Rahul - applicability ? [20:41:24] Stewart - for PW service without identical data links at both ends [20:41:50] Rahul - how does this compare with network transport layer service in MPLS-TP [20:42:38] Andy - need to handle any protocol (CLNP, ...) [20:43:04] obvious candidates - Ethernet, PPP, CAN transport anything between two routers [20:43:15] functonality is like PoS [20:43:41] Rahul - this document should point to TP framework document [20:45:19] Don - what is virtual Eth - what is the significance of the dest MAC ? [20:45:48] Stewart - the LSR that interface to the MPLS carrier forms boundary between SP and client w/o external I/F [20:46:29] DA is of dest router [20:46:52] this is a simple Ethernet PW , used a bit differently from the usual application [20:47:04] if ATM needs to connect to PoS then need common datalink, so can use Eth PW as canonical form [20:47:35] Andy - the only difference between this and a regular Eth PW is that this does not terminate on a physical Eth port [20:48:12] but terminates into a virtual port inside the router [20:49:53] Dave A - to be clear there is bridging as in regular Eth [20:50:43] Stewart - yes, could you help in defining this ? [20:50:50] Dave - I'm not the expert [20:51:01] Lucy - Enhanced ECMP and Large Flow Aware Transport [20:51:42] Why do we need enhanced ECMP ? [20:51:57] ECMP hashes flow IDs [20:52:06] Internet traffic traffic shows small percentage of (video) flows take large part of bandwidth [20:52:16] Hashing ECMP methods can't handle this efficiently [20:53:48] simulations show that hashing leads to congestion on some paths while others are empty [20:54:27] proposal - "enhanced ECMP" [20:54:51] different treatment for large and small flows - use hashing for low BW flows, but explicit table mapping for high BW flows [20:55:49] can recognize large BW flows in several ways, but out of scope [20:56:59] But this recognition is only at ingress PE, P routers recognize via indication inserted by PE [20:57:08] so not very high cost to the network [20:58:12] Simulation based on Caida data [20:58:50] there were 2% large flows that took 30% of the traffic (98% of the flows took the remaining 70%) [20:59:26] Showing graphs - with mechanism there is much more equalization between paths [21:02:15] an observation from the simulation is that compensation for the large flows can be done by this method [21:03:18] Proposal is to embed an indication of large BW flow in the "fat PW" flow Label [21:03:58] the flow label is used by P nodes for ECMP [21:04:17] proposal is to add an indication in the flow label to indicate whether it is a large or small flow [21:04:43] uses one of the TC (EXP) bits [21:05:39] She claims that this does not contradict present usage of TC bits for DiffServ purpose, since don't need TC for flow label [21:05:59] (the flow label is never top of stack) [21:13:56] no one has been jabbering since the scribes were at the mike [21:14:41] However, YJS said that this is an important issue, although the problem can be solved in ways that do not violate the MPLS standards [21:15:27] George asked how we know when the bottom label has the special TC bit [21:17:41] Lucy shows that the bottom of stack (flow) label has TTL=0 (which is illegal for an incoming packet) [21:20:22] Andy - out of time [21:21:22] Himanshu - quick comment - need to address applicability [21:21:51] Lizhong Jin - Control Word Reserved bit for use in E-Tree [21:23:12] Idea - use a bit in the CW of an Ethernet PW to indicate to a remote PE whether the frame comes from a Root AC or a Leaf AC [21:23:42] No new PW signaling is required - the Egress PE can decide whether the frame can be forwarded to a local Leaf AC or not [21:24:27] want allocation of one of the CW bits for the E-tree case [21:24:50] asks to make WG draft [21:25:54] Florin - need to decide in L2VPN how E-tree is to be handled first [21:26:03] Matthew agrees [21:26:22] Shahram - why not 2 different PWs ? [21:27:12] Lizhong - more complex this way [21:27:28] Dave A - no - this is more complex. [21:27:54] Dave - should not accept as WG item, 2 PWs are better [21:28:22] Luca - this was discussed at last IETF, this embeds topology info in the packet [21:28:51] this can be solved with present mechanism by using BGP communities [21:30:46] Nick DelRegno - Mandatory Feature of VCCV Implementations [21:32:05] VCCV has many choices - three control channel choices, no mandatory control channels, no standard TTL setting, [21:32:24] no consistency between vendors in usage [21:33:14] no interop between vendor equipment - this restricts deployments (examples from Verizon networks) [21:34:21] missed submission date, but draft proposes standardizing, e.g., CC type 3 with TTL=1 and type 1 with CW [21:34:33] router alert optional [21:35:29] in Ethernet PWs where CW not mandatory need method [21:43:59] lots of joking around at the mike [21:44:21] Tom and Yaakov tried convincing that the CW should be mandatory [21:44:40] Luca says that for single cell per packet ATM PWs 4 bytes on 52 is a large overhead [21:45:00] Yaakov suggests compensation for the extra overhead [21:45:54] Andy thinks that this draft would be good as a BCP [21:48:21] Tom volunteers [21:49:16] Yuanlong Jiang (Hauwei) - VPLS PE Model with E-Tree Support [21:49:32] explaingin IEEE E-tree scheme [21:50:19] also EVC service defined in MEF (6.1, 10.2) [21:50:35] problem of E-Tree in PBB-VPLS first raised in BBF [21:50:58] two I-Ds proposed to solve the general problem of E-Tree in VPLS in IETF, and also work in IEEE [21:51:28] 2 VSIs and 2 sets of PWs are needed per E-Tree if we incorporate the enhanced Asymmetric VLAN into the VPLS PE model 2 [21:52:02] scalability better if use Tree VSI (T-VSI) attached to S-VLAN bridge with Root VLAN and Leaf VLAN, and work in shared VLAN learning [21:52:12] Traffic from Root or Leaf UNI distributed into Root VLAN and Leaf VLAN [21:52:20] Only one T-VSI and one set of PWs needed per E-Tree [21:53:02] interconnection scenario - Either PE1 or PE2 can do VLAN translation (either when enter or exit PW) Bridge module filters Leaf VLAN traffic on the egress Leaf port [21:53:25] PW processing - [21:53:43] PW works in Tagged mode At least one end of PE provided with the VLAN mapping capability Remote Root VLAN <-> Local Root VLAN Remote Leaf VLAN <-> Local Leaf VLAN At the PW ingress, Root or Leaf VLAN encapsulated in the same PW and transparently label switched At the PW egress, Ethernet frames translated into Local Root or Leaf VLAN [21:55:08] need LDP extension [21:55:24] E-Tree sub-TLV is defined as one of interface parameters PEs negotiate their supports of E-Tree (T-VSI) when the PW is set up Root VLAN ID and Leaf VLAN ID carried in the sub-TLV P bit indicate that PE is attached with “Pure Leaves” R bit is a request flag for “Remote VLAN Translation” [21:55:40] Matthew - also being discussed in L2VPN - need to see how they handle first [21:55:47] Matthew - done ! [21:56:22] YJS leaves the room [22:01:58] Stewart Bryant leaves the room [22:19:31] Stewart Bryant joins the room [22:36:05] Stewart Bryant leaves the room [22:47:20] martini leaves the room: Replaced by new connection