SDNRG Minutes by Haomian Zheng IETF 94 - SDN Research Group Meeting Monday, November 2, 2015 9:00-11:30 Monday Morning session I Room 502 Agenda 1. Administrative and Agenda Bash - Chair (5 minutes - 5/150) 2. Existing SDN Research Group Topics 2.1 DNS based PKI Model for SDN Authentication and Authorization (15 Minutes - 35/150) Presenter: Hosnieh Rafiee (Remote) [Jamal] Use case #1, Controller, why change after defect? [Hosnieh] NE need to send information to SDN controller. There need to be authentication for this NE to the SDN controller. Otherwise SDN controller accepts every NE, can be attacked. [Jamal] This thing is of connectivity, some NE need to get approved from the controller, traffic Congestion afterwards you are about to send the message; [Hosnieh]: right, there is congestion, also possible wrong info to the controller; [Jamal]: I wonder how you look into something like openstack. [Hosnieh] All protocols should be satisfy this, e.g., openflow. Firstly the authentication and then send the information. [Jamal]: trigger authentication will result in traffic congestion? No; [Hosnieh]: yes, risky at the first communication. [Jamal]: no certification will result in no traffic congestion, as no authentication; [DK]: it’s about identity management, openstack can be a manager, use the list for further discussion; 2.2 SDN and ForCES based optimal network topology discovery Presenter: Evangelos Haleplidis (Remote) [DK] Very big advantage discovery mechanism compared with openflow; LLDP packet tested; [DK]: some result also included in benchmark testing doc; [bhuvan]: SDN benchmark draft author; in this draft the metric and methodology were defined, welcome to comment; [DK]: you have backup slides; [Evangelos]: yes, some parts of the model info included; [cotalo]: what happen number of SDN switches increase? [Evangelos]: we don’t expect this grow very much. Only a big grow will result in scalability issue. Currently it went well. [cotalo]: the change in SDN will result in topology change, which may happen; [Evangelos] yes; [bhuvan]: increase switch will be exponential instead of linear; suggest adding more detail statistics; [DK]: is this already concluded? And published? [bhuvan]: yes. 2.3 Cooperating Layered Architecture for SDN (15 Minutes - 20/150) Presenter: Luis M Contreras https://tools.ietf.org/html/draft-contreras-sdnrg-layered-sdn-04 [Michael] The architecture like service orchestrator; how does it scale? How this can be localized to one area? [Luis]: you are talking about multiple controller interactions/relationship, it’s in the scope, but not fully considered yet. [Kostas Pentikousis(KP)]: there is draft talks about inter-domain, and policy issues; would happy to give some input; [Hui Deng]: What is the relationship between SDN orchestrator and SDN global controller? hierarchy controllers? How it will orchestrate? Suggest to clarify the architectures in between. [Luis]: Offline or mailing list discussion, welcome to hear suggestsions; [DK]: hands up; quite a few read & support RG; 2.4 SDN Controller Benchmarking and Terminology (15 Minutes - 50/150) Presenter: Bhuvaneswaran Vengainathan https://tools.ietf.org/html/draft-bhuvan-bmwg-sdn-controller-benchmark-term https://tools.ietf.org/html/draft-bhuvan-bmwg-sdn-controller-benchmark-meth [Jinzhu] Performance matrix; constraint consideration? Number of switches in the test? [Bhuvan]: configurable in the test; [Jamal]: consider the table sizes; atomic; row numbers…; 2) to consider CPU sufficiency; how robust is the controller? [Bhuvan] [Adlo]: single /multi table pipelines; [Bhuvan]: that is useful; controller black box; less useful [adlo]: [Kostas]:1) Terminology already published as RFC7426, together with architecture and layers. 2) valuable if don’t care protocols, [Bhuvan]: Our next objective is to make it generic. [DK]: Not much protocol discussion as this is a research group; test numerical result will be very helpful to industry. Some questions Missed during the discussion, Help 2.5 PCECC: an IETF-based SDN Controller (15 Minutes - 65/150) Presenter: Quintin Zhao http://tools.ietf.org/html/draft-zhao-pce-central-controller-user-cases http://tools.ietf.org/html/draft-zhao-pce-pcep-extension-for-pce-controller [DK]: any support you can find in RG? [p]: ask for help polish from scientific dimension; hopefully more researchers can provide input; 3. Operators, Service Providers and Research Projects: Challenges, Findings and Opportunities 3.1 SDN as enabler of proactive network operation based on "network science" (20 Minutes - 85/150) Presenter: Kohei Shiomoto [DK]: I am interested in correlation work in research field about bigdata; multi-technology virtualization; will be function-driven. I would like to see more application with bigdata in future. 3.2 Practice of deploying SDN and VNFs in the data center (15 Minutes - 100/150) Presenter: Jinzhu Wang [pedro] Surprise to hear interface is not verified, Do you have any knowledge about SDN/NFV presentation (e.g. manu?) They seems provide such stuffs; [Jinzhu] We think manu defines some reference point, it’s info model, instead of protocol. For example you can choose NETCONF or REST to transfer the information. But in this one we only limited to NETCONF. In our deployment our implementation based on Opendaylight, and our vendors provide such supporting. [Jamal] VM is used as a SDN controller? [Jinzhu] This is one of our deployment at a public data center. We setup VM in openstack, Neutral will send this message; [Jamal] vnfm is from nova or something like that? [DK] VNFM is a functional component, as a functional manager. Function requests are sent to VNFM for the virtual network; [Bhuvan] There is one thing complicating: instantiate a vn from vnfm to SDN controller; why different from launching from VM, seems to be same? Any provisioning / orchestration into sdn controller? [Jinzhu]: take virtual routers as VMs; or (proposed) use VNFM to manage the virtual routers; now out of scope of openstack; openstack did not prohibit such VNFM. [??] option1: more independent orchestrator; you are using openstack as an orchestrator? [Jinzhu] In this deployment there is no orchestrator, as it is Data Center. Different from what ETSI is talking. [??] This resource control and functional control should be run independently? SDN controller is not giving anything to VNF manager? If you make it general and use identical architecture, it can get some notification and start coordination. [Jinzhu] Management entities may need to help in hierarchical structure; (move to offline discussion & mailing list) [John] multi-vendor use multi-controller? [Jinzhu] yes. [John] then how to interact? [Jinzhu] VNGM by us (China Mobile), vendors now provide full solutions, currently in opendaylight, but not standardized. 3.3 "SUPERFLUIDITY" [An SDN] Project Grand Challenges (15 Minutes - 115/150) Presenter: Pedro A. Aranda Gutierrez [Aran] For monitoring, the patient is more critical than anything else; [Pedro]: some resource will be devoted from this project. 3.4 "COHERENT" Towards a control framework for 5G mobile networks (15 Minutes - 130/150) Presenter: Kostas Pentikousis 4. SDN RG Charter, Scope and Existing I-Ds (15 Minutes - 145/150) Presenter: Daniel King (SDN Chair) =========== Late Request =========== "LawNFO" Decision Framework for VNF Placement Presenter: Michiaki Hayashi