Minutes for DMM at IETF-90
minutes-90-dmm-2
Meeting Minutes | Distributed Mobility Management (dmm) WG | |
---|---|---|
Date and time | 2014-07-24 19:20 | |
Title | Minutes for DMM at IETF-90 | |
State | Active | |
Other versions | plain text | |
Last updated | 2014-08-04 |
minutes-90-dmm-2
IETF 90 DMM Notes, Thursday 24 15:25 - 17:20 New individual documents & proposals (75 minutes) --------------------------------------------------------------- Chair presentation Status slide No comments 15:25 - 15:50 Alper (25 minutes) o CSIPTO Introduction o draft-yegin-ip-mobility-orchestrator o draft-yegin-dmm-ondemand-mobility Presentation: CSIPTO overview Presentation: On demand mobility Presentation: Mobility orchestrator Introductin to CSIPTO Slide 2 – SIPTO – what is SIPTO. Slide 3 – CSIPTO – Coordinated SIPTO (RE 13 SA1 work item) Slide 4 – Use Case 1 – UE moves from eNB1 to eNB2. Old flows (requiring IP continuity) will use the old PGW1 to maintain source IP address, new flows (and old flows that do not require IP continuity) can use PGW2 (PGW1 in suboptimal). The PGW1 path will stop to exist when not required anymore. Slide 5 – Use case 1 - Higher layer handover methods may be used (MPTCP, SIP…) for soft handover Slide 6 – Use Case 2 – before mobility two PDN connections are created. One without IP continuity guarantee through a local WG (PDN 1 - LGW which could be co-located at the eNB)and the second with IP address continuity guarantee (PDN 2 - through the PGW at the core). Apps that do not require IP continuity will use PDN1 and apps that do require IP continuity after handoff us PDN2. Slide 7 – Use Case 3 – by default, a PDN connection (without IP continuity guarantee) is set through the LGW. The connection through the PGW is establish only in the UE requires IP mobility. Sri regarding use case 1: When is the second PDN connection establish? Alper: application-specific – if the application requires IP continuity, the PDN through PG1 is used, otherwise, the new PDC connection through PG2 is used Slide 8 – Status of CSIPTO Dapeng: regarding SIPTO what is flat IP Alper: pushing the MA as closed to the edge as possible. IP Mobility Orchestrator Slide 2 – Categorization of session continuity solution. a list of different solutions in different layers (in the IP layer, Above the IP layer and below the IP layer – link layer) Slide 3 – Observation: Tradeoff – sub-IP and IP solutions cannot provide optimized path because of the triangular routing. Above IP can provide optimized path because the IP address can change and the app will still maintain the connection. Carlos: one could do route optimization for IP layer Alper: right will be addressed in the future. Charlie: route optimization is layer 3 Alper: true Slide 4 – Observation: Blocking – lower-layer based solutions inhibiting operation of upper lay by providing a solution even if it is nor optimized. Slide 5: Example of orchestration – using e2e MPTCP with PMIP in access network. Initially use the original IP address, then after handoff, both IP addresses are used in parallel and finally, the original path using IP1 can be dropped. Sri: MPTCP – each IP address is bound to a different interface Alper: no – that is incorrect they can also be in the same interface. Kostas: Any measurements? are you planning to ship Alper: no measurements. Cannot comment on what we are plans Slide 6 – Basic flow involving the mobile host, source GW, target GW and corresponding node. Slide 7 – The IP Mobility Orchestrator – a new entity in the UE for the discovering the capabilities of the network and CN (in terms of mobility support) Slide 8 – Discovery Flow (with the Mobility Orchestrator Slide 9 – Mobility Protocol Selection Algorithm Slide 10 – Handover Algorithm Jouni: Does it include protocol work or only APIs? Alper: Also protocol work Dapeng: How can the Orchestrator no if MPTCP is supported (since it depends on the app) Alper: Through API the app declares what it needs. Brian: This is a general capabilities negotiation – are you aware of the difficulties of standardizing APIs Alper: It is not easy but it is done in MIF for examples. Brian: concerned that the APIs will never be standardized. Alper: we will have to deal with that. 15:50 - 16:00 Stateless U-plane vEPC, Satoru (10 minutes) draft-matsushima-stateless-uplane-vepc Presentation: Virtual EPC Stateless user-plane for vEPC (Virtual Evolved Packet Core) Slide 2 – Summary of 02-03 updates – routing clarification (sec 3.2) and Operational classification (Sec 4.1) Slide 3 – How the I-D fits the re-charter work Item and Milestones Slide 4 – How the I-D fits the re-charter work Item and Milestones Sri: Where is the charging (accounting) done Matsushima-san: No charging in this model. Could add another charging entity Marco: About the green arrow: How often will the BGP routes be updated? Matsushima-san: Not that often John: Also concerned about charging. May use SDN Matsushima: We will find a solution Rajeev: Is the group chartered to do user-plane/control-plane Matsushima-san: Yes 16:00 - 16:10 IP mobility based solutions, Carlos (10 minutes) draft-bernardos-dmm-pmip (PMIPv6 extensions) draft-bernardos-dmm-cmip (MIPv6 extensions) Presentation: DMM IP mobility based solutions (PMIP6 & CMIP6) Slide 2: Where are we in the SMM ocean – difficult to categorize. Could be re-use PMIPv6 signaling, network and client based or access network anchoring slide 3: We extend existing IP mobility protocols – a list of previous papers about client and proxy MIPv6 Slide 4: Client based DMM solution – with the mobility management pushed to the edge of the NW Behcet: I had a draft – did you compare with my draft (I presented in 2012) Carlos: these papers are from 2011 Slide 5: Client-based solution entities – DAR (Distributed Anchor Router) – first hop router. Slide 6: Client based solution operations (I) – after handoff the MN can keep the old address by notifying the corresponding DAR with a BU Slide 7: Client based solution operations (II) – the MN can also use the new IP address (anchored in the new DAR) for new sessions Slide 8: Net-based DMM solution - based on PMIPv6, anchor is pushed to the edge, user plane is distributed, control plane is centralized Slide 9: Net-based solution Entities – MAAR (Mobility Anchor and Access Router) – also one IP address hop from the MN. Concentrates routing, LMA and MAG functionality. CMD (Central Mobility DB) – manages all MNs in the domain. Data packets do not flow through it. Slide 10: Net-based solution. Operations: Initial registration Slide 11: Net-based solution. Operations: Handover Slide 12: Net-based solution. CMD as PBU/PBA proxy Slide 13: Analysis against DMM requirements Slide 14: Solution implemented see www.odmm.net. Also subscribe to odmm@odmm.net Sri: Which extension did you use Carlos: Extension for authentication 16:10 - 16:20 Distributed logical interface concept, Carlos (10 minutes) draft-bernardos-dmm-distributed-anchoring Presentation: Distributed logical interface Slide 2: Where are we in the DMM ocean – difficult to categorize slide 3: What is common to other approaches – routers at the edge assign locally anchored prefixes to the MNs; compatible with RFC5213; D-GWs behave as LMA/MAG; transparent to end-host Slide 4: The Distributed Logical Interface (DLIF) Motivation Slide 5: The DLIF concept – hide the change of anchor from the MN; each D-GW is seen to the MN as multiple routers – one per active anchoring D-GW associated with the MN. Although the MN is physically connected to one physical router, it sees different routers (logical). Slide 6: Solution overview (I) Alper: What is the difference between this and PMIP Carlos: It is similar but the interface is a logical interface and there are protocol implications. Sri: What is the interface to receive a different source address Carlos: This is transparent to the host because it sees a different router (virtual) Slide 7: Solution overview (II) Slide 8: Summary Alper: What is the protocol work Carlos: example: between logical interfaces (virtual routers) Alper: So how does PMIP6 works Carlos: there is only one physical router. 16:20 - 16:40 Controller-based Mobility Architectures, Sri (20 minutes) http://www.ietf.org/proceedings/89/slides/slides-89-dmm-2.pdf An update on some off-line discussions… Slide 1: Stated assumptions Kostas: types – OpenFlow is one word, … Sri: fare enough Hui: What is the purpose of this work? Sri: this is forward looking for next generation. Cannot relate to 5G Hui: Is this related to 3GPP or WiFi Sri: This is radio agnostic Hui: So this could be apply to WiFi or 4G or 5G Sri: Yes Slide 2: DMM – Functional Elements Alper: What about an anchor in the CN network Sri: Fine, OK Charlie: What about authentication is a distributed system – not trivial Sri: Agree Slide 3: Functional architecture with home and access distinction (split control and data-plane) Slide 4: Home and Access distinction Removed from the Control Plane Functions Rajeev: For a generic mobility controller you only have the infrastructure. What is the scope of this group. Sri: Client and NW based Slide 5: Home-DPA/Access-DPN collocation Hui: It depends on the type of the application. Some require a permanently allocated IP address. Sri: I agree, but this is not always required. Alper: We are designing it for the future. In the ACTN BOF discussion there was a layout of such brake out. Sri to Hui: Can you say that in a single operated domain can you say that this is not relevant? Hui: I agree that in this case it is correct. Slide 6: Home and Access Distinction Removed from both Control/Data plane functions Time is running out … Slide 7:.. Slide 8: Home CPA Slide 9: Home DPA . . . Slide n: Demand for the exposure of states to the transport NW Subir: Struggling with the objective – transport network programming? Sri: We are access agnostic. We can control routers 16:40 - 17:20 WG next steps, Chairs/WG (40 minutes) --------------------------------------------------- 16:40 - 17:20 Re-chartering discussion o Charter text discussion and editing o Deployment models, "DMM architecure", functional elements o Discuss work procedures and how to divide work Jouni: We want to complete the re-charter fast to be able to start working on the solutions. We have to complete the charter work this month. Brian: The charter will go through IESG review and a two week phase like WGLC and then, to IESG final approval. I need it on the agenda before the 31st of July. In that case, if the WG gets the charter text ready by the end of July, we can get charter approval by the end of August. Jouni: We need to agree on the text and the milestone dates. Kostas: There is a disconnect between the text that motivates and the milestones. I agree with the items to work on (the milestones). We only need to re-word the motivation part. Dapeng: Will you be willing to provide edits. Kostas: Yes Brian: Don’t rely on anyone else. If you have comments – provide text. Kostas: I have no problem editing the text. I am concerned that it will ignite a whole new round of comments. Brian: Answer about the 4 work items – can the group reach consensus? Behcet: Each draft will have a deployment section. What is the deployment model…. . . . JC: I support the 4 work items. I agree to improve the motivation. Jouni: Danny – write in the minutes the Kostas will provide a new edit by Monday ;-) Rajeev: We should think on mobility in general. It’s a mistake to have text like MIP, PMIP… Use: mobility, end-to-end, network mobility… there is a consumer of mobility and a provider of mobility. Jouni: Good point. Hui: About the centralize and distributed control plane. We have to have something clear for implementation. This is not academic. I don’t see the clear solution. There are lots of other possibilities that we did not state. Sri: in response – We implied that the control plane is virtualized but we did not require it. Jouni: We will have to have some conference calls next week so that we can conclude by the end of next week. Sri: I expect the charter to be broad enough to define a scope. We do not need a list of documents. Jouni: correct ******************************************************************* IETF 90 DMM Notes, Friday 25 Presentation 1: DMM Deployment Model (Anthony Chan) - Description of Network function virtualization with DP-CP separation. - Mobility management options: tunneling with optimization Alper: is it an inter-MAG tunnel? [Alper says that he will take the discussion later] Sri: as a WG we need to identify functions: do these map to the function mapping in {sri's} slides? A: this is what the functions do, these slides are what they are. e.g. anchor allocates IP address in Sri's slides, Anthony's slides are what the functions are. Presentation 2: Enhanced Mobility Anchoring (Anthony Chan) - Mobility management options: enhanced anchor, move IP address, new IP address Sri: we need to ensure that the goals here should be mapped to the functional entities (in Sri's slides) Presentation 3: DMM for Wifi users in Fixed Networks (Behcet) - Motivation: No IP mobility defined for current WiFi/fixed network (and other proposals in dmm today are 3GPP centric) - DMM4WiFi arch: unified gateways (UGW), virtualized control plane (BNG app, SDN app) - Layer 2 mobility - SDN controller calculates shortest path on mobility. - Layer 3 mobility - rfc 5949, predictive and reactive handover. And I2RS type of control. Dapeng: whats the interface between cloud / BR? (slide 6) B: xmpp/Jabber protocol. Presentation 4: Demand for exposure of states to the transport network - Traffic Steering (Marco) - DP anchor (mobility session of a user network - applies default routes ..) - New mobility session - topological binding of address becomes incorrect. Need to handle this. - Anchor relocation and mobility session transfer while keeping routing path short. - Generic description now, no commitment to protocols yet Alper: updating host specific route - so it is in the same category as Softbank/Pete's solution? M: argument is same. But host route does not need to be propagated - default routes can be used. Alper: it is a 3rd approach (vEPC, Pete's and then this one). Sri: is it aligned with softbank's proposal - i.e. with BGP? Alper: high level view is the same. But details are different - 3GPP, etc. Sri: abstract it at semantic level, and leverage work at BGP level. Otherwise we will be redefining everything. Presentation 5: IP mobility protocol for multi homed access (Sri on behalf of Pierrick) - Multi-homed RG - BBF work item 2014.546.03: wireless and wireline operators and user is multi homed through both with load sharing. Hybrid access gateway function. - BBF assumption is that both the operators are the same. - Prefix should be reachable through BNG. - 1:1 mapping to NEMO (with MCoA) is possible. - BBF requirements: do not split a sub-flow (need per flow/packet management) - Mapping to PMIP: PGW is LMA, BNG is MAG. John: handling of link failure Sri: yes. Marco: flow mpobility extension of PMIP required Sri: yes M: are there other candidate protocols for flow steering? Sri: not clear. GRE control plane was proposed, but it requires a high level control plane. Behcet: is there another related work related to PCC that can solve this? Sri: don't know. Presentation 6: Asymmetric Extended Route Optimization (AERO) - Fred Templin - Different from other proposals - not based on MIP, PMIP, etc. - RC 6706 is the basis. - AERO client (handset in airplane), AERO server (embedded in access net), AERO relay - IPv6 is the signaling mechanism, LLA with unique prefixes - AERO mobility by updating the server's neighbor cache - Route optimization: predirect message (server is not in loop) - Internal BGP for updates, long binding to a server. - Used for corporate users in Boeing and civil aviation (airplane is a mobile router) Bhumip: are the servers virtual? Fred: can be physical or virtual B: what is Boeing implementation. Fred: either or is fine - some are virtual and others physical. Raj: Does it work for IPv4? Fred: it uses IPv6 LLA. But IPv4 can be used. Presentation 7: On Demand Mobility Management (Alper) - Categories of IP address: fixed, sustained, nomadic - RFC 5014 - socket API for source address selection defines home /CoA: not sufficient for DMM. - On demand nature not captured in RFC 5014: 3 flags defined for fixed, sustained and nomadic. Sri: is it to replace the existing 5014 flags? A: no, but old flags will be ignored. Sri: behavior of existing apps will not be impacted? A: no. Dapeng: do we need to care for fixed address? A: e.g. server that provides fixed IP addresses for users. Jouni: this is only an API def? A: this plugs into prefix coloring draft, so there is some on-the-wire impact. Discussion Areas to Work on - 4 areas in Charter: - Bi-weekly virtual WG meeting proposal - Small teams on milestones, 2 editors and rest named contributors Alper: deliverables slide - there may be more than 4 RFC, these are just areas. Chair: intention is not to have just 4 RFC. This is a first level of detail. For eg,, we derive RFC for traffic steering, then we may need other RFC for protocols Alper: first call to decide how to split work. Chair; ok Sri: first item decides functional model - should be stds track. Brian: arch and such should be informational. Sri: what is the impact if something changes to functional elements? Brian (AD): can mix std, BCPs. Etc should be no problem. Danny: this is a good approach. Try to divide work and then define table of contents. Chair: yes. Too many documents, so the design team /related approach is needed. Marco: regarding teams - we need to breakdown these items into technical areas, each with a team. Chair: yes. Its likely that multiple document and different docs are needed. Sri: in the past - folks sign up but then there is no progress. If there is no progress, chairs should fire them. Chair: can be strict. Chair: PMIPv6 maintenance continues as side track - things that are broken, and leftover from netext. But not an invitation for new work. H Chan: how many people are going to be actively involved in this work? Chair: will ask this in first call. Sri: given that we don't know the breakup into docs - we should delay this call. Marco: people may have different understanding of each item - so one tel call to converge on these specifics. Chair: yes. Chair: will put details on mailing list, and will be aggressive in setting timeline/conclusion of charter text.