[10:03:03] ehalep joins the room [10:03:09] ehalep leaves the room [10:12:24] jhs joins the room [10:14:41] ehalep joins the room [10:14:45] ehalep leaves the room [10:16:11] ehalep joins the room [10:16:25] ehalep has set the subject to: IETF 78 ForCES WG [10:24:51] ehalep leaves the room [10:25:23] ehalep joins the room [10:25:26] ehalep leaves the room [10:25:56] jhs leaves the room: offline [10:26:15] ehalep joins the room [10:26:41] ehalep is now known as Evangelos Haleplidis [11:36:06] Evangelos Haleplidis leaves the room [11:36:21] Evangelos Haleplidis joins the room [11:40:03] Evangelos Haleplidis leaves the room [11:40:20] Evangelos Haleplidis joins the room [13:11:05] Evangelos Haleplidis leaves the room [13:11:19] Evangelos Haleplidis joins the room [13:14:00] jamal joins the room [13:14:12] ok, this works ;-> [13:19:22] avri joins the room [13:20:51] hello avri [13:21:49] hi, having trouble getting the audio [13:23:03] oh - did you hear that? [13:23:46] no, i think i have the channel, but keep losing it. [13:24:06] Evangelos will scribe the meeting [13:24:27] you can get the slides and follow [13:24:53] ok, i hear something but not talking. [13:25:01] oh, now i hear [13:28:04] Starting the meeting. [13:28:16] Jamal speaks: Agenta. Loaded meeting. [13:28:28] Joel and Weiming have priority [13:28:42] Kentaro and last is Evangelos. [13:28:55] Joel and Weiming try to get as much resolution as possible [13:29:02] Note well slide [13:29:26] spadgett joins the room [13:30:03] angga joins the room [13:30:15] Slides are available online [13:30:44] Last meeting 4 rfcs published [13:30:52] Slides link: https://datatracker.ietf.org/meeting/78/materials.html#wg-forces [13:31:09] After this meeting Jamal plans to get RFC errata [13:31:29] Documents to be published: Applicability statement, and Implementation Report. [13:31:54] 2 more important LFBLib draft [13:32:32] We want to make LFBLib proposed standard? [13:32:48] Joel: There is room for argument. We want for the things to be defined, we want that definition. [13:33:16] Joel: We want vendors to use the same definitions. MIBs go proposed standards.same [13:33:44] sound keeps coming and going. is the mike being used? [13:33:50] or is it just me? [13:34:20] Jamal: Strong sentiment for the CEHA effort try and leave CECE plane off for now. Attack CEHA perspective. [13:34:20] adrianfarrel joins the room [13:34:45] adrianfarrel leaves the room [13:35:05] Jamal: After all this work the wg will shutdown around March from the looks of it. [13:35:22] We must move as fast as possible [13:35:48] Joel on the mic for Ethernet port Evolution [13:36:45] A number of people have work on the library of the lfbs. [13:36:59] we're trying to figure out what is the basic connectivity [13:37:06] we started with a single answer. [13:37:36] IEEE - not so simple [13:37:42] Separate physical layer. [13:38:34] You need the right kind of LFBs for the right arguments. [13:39:04] We tried to separate the classes. [13:39:09] What structure do we need? [13:39:23] The next conceptual thing (Ethernet Handler). Decide what is local or remote. [13:39:50] The reason is nonlocal bridging stuff, we can put whatever lfb's later. [13:40:07] adrianfarrel joins the room [13:40:15] adrianfarrel leaves the room [13:40:40] avri leaves the room: Replaced by new connection [13:40:41] avri joins the room [13:41:01] What we have to do for local packets. Logical or Physical port it came. If it is a VLAN packet. [13:41:27] You have to look at what VLAN and decide where to go. [13:41:33] Back up a slide. [13:41:53] Next slide. [13:42:20] It gives us a nice isolation from the Ethernet Handler. [13:42:44] You need to be able to allow EtherType and VLAN demultiplexing. [13:42:53] Next slide. [13:43:02] Still we need. [13:43:18] adrianfarrel joins the room [13:43:23] One interesting question: Where input packet validation happens. [13:43:48] adrianfarrel leaves the room [13:44:47] we have to figure the components of the PHY. [13:44:54] Weiming on the mic. [13:45:23] adrianfarrel joins the room [13:45:32] W: i just thing that this looks quite good, but there are maybe need a little bit more discussion on the ethernet. (Go back to Reusability and separation) [13:45:45] adrianfarrel leaves the room [13:45:46] Weiming: How can EtherHandle decide which is for bridging. [13:45:49] adrianfarrel joins the room [13:46:35] Joel: If the Ether Handler can see the MAC address, it can have a table for decision. [13:47:34] Weiming: Is it possible there might be a need for packets to go both way. [13:47:54] Joel: Multicast addresses which are used for bridging, they will go both ways. [13:48:10] Joel: If you do bridging, you have to pass it along. [13:50:27] Jamal: When we use your slides we can revisit these issues [13:50:38] Weiming on the mic. [13:50:59] Slides: On update of FoCES LFB Library Draft [13:51:37] Weiming: want to present on some issues on the update of the LFBlib draft [13:51:42] next slide [13:52:04] we are prepared to update it very soon. [13:53:20] Next slide [13:54:49] Removed all related FEPO and FEO LFBs. Remove ICMP processing. Maybe ICMP is handled by Control Plane. Not proper to define inside the FE. [13:55:28] On IPv4 and IPv6 a better way to process these two stacks is to separate their related LFBs [13:55:43] we will usually have separated LFbs for individual IPv4 and IPv6 [13:55:59] for instance we could have IPv4Calidator and IPv6Validator. [13:56:10] The same thing may apply to processing like ARP LFB. [13:56:49] on the specific LFB Classes we decided to defin a relative simple sample scheduler to reflect the requirments of QoS. [13:57:11] In this LFB we would have multiple inputs and the function of queues will be included. [13:57:15] and it will have one output [13:57:55] specific redirect lfbs, cause redirect lfb will be very different from port lfbs. we need specific ones. [13:58:29] Next slide [13:59:25] Other Consensuses use FE model function like alias to describe a lot of things including Address Resolution Table, will be used in ARP and in encapsulation. [13:59:30] Jamal: Alias? [13:59:49] Weiming: One component is defined in one place and will be used in another place. [14:00:16] Joel: Is refering to the model defintion of alias. Any IP should be defined once. [14:00:44] Joel: Things that will be used in two places, are defined in one place, and are referenced using alias from the other place. [14:02:38] Weiming: Another consensus. For LPM, we decide that we will use two separate tables to describe this procces. Longest Prefix Map table and another Next Hop table. May only be incremented by one table. In this case the ForCES need to map this kind application based on abstraction of individual implementations. We suggest that for LPM we need to make some specific statement [14:02:47] to this kind of application [14:03:04] Weiming: Need to put some text to describe it. [14:03:05] Next slide [14:03:24] No questions on the consensus [14:03:37] Proposed discussions for work: [14:05:03] Joel on the mic [14:06:14] Joel: this match is the slides i presented. The EtherInPort i have it as a different name. These are tiny issues. [14:06:21] Weiming on the mic [14:06:33] Next slide [14:07:31] Question: Steve Padgett: What are we talking like in port... [14:07:55] Evangelos Haleplidis leaves the room [14:08:05] Evangelos Haleplidis joins the room [14:08:40] Weiming on the mic [14:09:03] For Input port we're trying to define one input and two outputs. [14:09:38] I don't know if it is proper to have it as L3 and L2. we can discuss. [14:09:50] Jamal: Want to define ...? [14:10:02] Joel: This is the port that makes the local/non-local decision. [14:10:11] the name is up in the air. [14:11:15] ... these are two ports that one is local and non-local. [14:11:15] [14:11:32] Weiming: We need to discuss components [14:11:33] next slide [14:11:49] This is a table that is based one existing EtherInPort LFB [14:12:24] Inside this component we're trying to add component to reflect the requirements for L2 forwarding. [14:12:47] for forwarding a booleanType, if it is able or not. [14:13:08] Something we may try hard to do how should we define the classification on these two ports. [14:13:14] Joel: Rephrase [14:13:31] Weiming: We can define the boolean type very easily but we don't think it's enough [14:14:03] Joel: For the Ethernet hanlder LFB the decision for local or not, it's look in a table with a MACs [14:14:18] There are cornercases we have to see [14:15:15] Steve: where is stuff like flow control is present? [14:16:06] Steve: LinkSpeed is read-write? Maybe link speed is ok for the PHY... [14:17:01] Joel: A typical value is an enumeration. These are supported values. we support that. [14:18:00] Jamal: Is there a slide for the PHY LFB? [14:18:21] Weiming: Yes. [14:18:24] Stewart Bryant joins the room [14:18:56] L2ForwardingEnable makes sense in this level. [14:19:41] Joel: My thought was that this is where you make choise for local-non local. [14:20:20] Next Slide [14:20:34] EtherPhy. [14:20:47] many questions. [14:22:06] Joel: There is not MAC mux/demux in the PHY. [14:22:24] We can allow multiple outputs connected to the same PHY. [14:22:32] Next slide [14:23:40] Joel: I don't know if ifindex should be here or not. [14:23:55] Weiming: If we don't have it it seems quite difficult to reflect the IP addresses [14:24:07] Joel on the mic [14:24:54] Weiming: Some functions inside LFB may use this index. [14:25:04] Joel: ARP never uses Ifindex. [14:25:40] You can have an array, one which points to the interface and other to the address information [14:25:42] next slide [14:25:48] etheroutport [14:25:59] next slide [14:26:07] next slide [14:26:25] next slide: EtherClassifier LFB (2) [14:26:39] We use something like dispatch table and vlan table to demux [14:26:49] Joel: this is the right components. [14:26:53] discuss offline. [14:26:55] next slide. [14:26:58] next [14:27:03] EtherEncap LFB(2) [14:27:19] we may nee to have things like ARPtable, even VLanTable [14:28:06] Jamal: Should be somewhere else and alias'ed? [14:28:30] Joel: I think the ARP and tthe V6 goes elsewhere. [14:28:31] [14:28:55] That information goes here. [14:29:03] This is the mirror of demultiplexer [14:29:29] Jamal: I understand. This grows infinitely. [14:29:46] Joel: There are not infinite encapsulations. [14:30:04] Infiniband is a different encapsulation. This is to handle Ethernet. [14:30:12] Jamal: If i had multiple VLAN tags. [14:30:22] Joel: IEEE has restrictions. [14:30:25] Jamal: MPLS? [14:30:38] Joel: this is below MPLS, i want it to go this output port [14:30:49] Jamal: This is only for the EtherType [14:30:53] Joel: And the VLANs. [14:31:12] Joel: The idea is the mirror of the other case. [14:31:51] Jamal: Discussion off-line. [14:32:01] next slide (go to IPv4Validator LFB(2)) [14:32:19] next slide (go to ArpLFB(2)) [14:32:28] we need the Ip addresses to ARP. [14:32:43] Joel on the mic: Disagree. [14:33:03] Joel: Is this a valid packet that should have been received in this interface? [14:34:21] Weiming: Alias may not be enough for the ARP [14:34:44] Joel: we don't know where they should be yet [14:34:46] next slide [14:34:57] comments on lfb class definitions. [14:36:29] next slide. (Comments on the draft status) [14:37:32] Kentaro Ogawa on the mic. Next set of slides. [14:37:54] adrian - question for you; the LFBlib is considered important enought to be tied to the rest of the standards for interop purposes .. [14:38:16] wed like to push it as a standards doc - we can chat afterwards or if you have time to come over [14:38:32] Kentaro: [14:38:37] msg from adrian - who is unable to get into the room: [14:38:43] Can you drop something in the room that if they need their AD they should ping him direct. Email at adrian@olddog.co.uk may be quickest [14:38:43] Next slide [14:39:25] Next slide [14:39:35] Current CE HA: Cold stanby [14:39:47] the first step FE1 associates with CE. [14:40:03] DTime1: Time to prepare to associate. [14:40:12] DtimeA: Time to complete. [14:40:22] Next slide [14:40:48] After CE1 failure, Dtime2: time to detectCE1 failure. [14:41:06] Then we have Dtime1 and DTimeA again. [14:41:11] and Dtime3 for state recovery [14:41:14] Next Slide [14:41:21] CE HA: Hot Standby [14:41:23] Next slide [14:41:38] dtime1 and dtimeA removed, as FE1 is connected to CE2. [14:41:42] dtime3 is reduced. [14:41:50] only dtime2 is needed. [14:41:55] Next slide [14:42:13] Changes needed. [14:42:37] Joel: I got a big problem with that the FE should accept commands for all CEs. [14:42:50] Joel: We can't operate that way. [14:43:09] Joel: No problem with establishing a backup association. [14:43:33] Joel: But it brakes the structure. [14:43:45] Jamal: If you assume that the CEs .. [14:44:14] Joel: That is error prone. The problem is the way we defined before... it's pretty fast... [14:44:36] Jamal: If you cold standby, all CEs can be up and running. [14:44:43] Joel (made that comment) [14:45:47] Joel: Not to complicate the protocol for something that may not be faster. [14:47:40] Jamal: Objection noted. [14:49:24] Removing dtime3 slide [14:49:52] next slide [14:50:02] Optional CECE Synchronization [14:50:31] Joel on the mic: [14:51:18] As i said on teh last meeting. CECE is useful. This is important, i can argue that control recovery is not important as forwarding recovery [14:51:59] Jamal: why is the problem to use forces? [14:52:23] Joel: The problem is, you get confused underlying state... [14:52:52] mirroring the state operation is almost useless if you can deal with the rest of the issues. [14:53:05] Jamal, not include CECE synchro [14:54:17] Joel: Steve's point is: Events from FE to CE. You need to reflect them to other CEs. You can just ignore them. To do ForCES sets won't reflect it. [14:54:32] Jamal: Reflecting the events to all CEs [14:54:39] Joel: that would do it. [14:55:14] Request to make this a wg document. [14:55:34] Jamal: objection? [14:55:38] without CECE. [14:59:51] Jamal will send notification for the CECE HA to be a wg item. [15:00:41] Joel: There is an interoperability issue. [15:01:02] Adrian: There is need to be changes to the document to say that this is useful information [15:01:08] spadgett leaves the room [15:03:02] Meeting adjourned [15:03:15] Stewart Bryant leaves the room: Disconnected. [15:04:45] angga leaves the room [15:12:01] Stewart Bryant joins the room [15:12:13] spadgett joins the room [15:14:19] angga joins the room [15:15:55] spadgett leaves the room [15:26:12] Evangelos Haleplidis leaves the room: Inside the ForCES wg room [15:32:24] Evangelos Haleplidis joins the room [16:02:41] jamal leaves the room [16:04:12] Evangelos Haleplidis leaves the room [16:07:23] Stewart Bryant leaves the room: Disconnected. [16:21:46] angga leaves the room [16:35:42] angga joins the room [16:45:47] angga leaves the room [20:49:00] avri leaves the room [21:31:39] angga joins the room [22:27:02] angga leaves the room [23:13:57] Stewart Bryant joins the room [23:17:50] Stewart Bryant leaves the room: Disconnected.