IETF 83 - NVO3 BOF - Minutes --------------------------------- Location: Maillot meeting room, Le Palais des Congres de Paris, Paris, France Time: 28-Mar-2012, 1300-1500 - Wednesday, Afternoon Session I Chairs: Benson Schliesser & Matthew Bocci Agenda posted at http://www.ietf.org/proceedings/83/agenda/agenda-83-nvo3.html Outline of the Agenda: * Intro (Chairs) * NVO3 Problem Statement (Thomas Narten) * Overlay Networking Framework (Marc Lasserre) * NVO3 Requirements for Tunneling (Igor Gashinsky) * VPN Interactions (David Black) * Network Virtualization Overlay Control Protocol Requirements (Lawrence Kreeger) * Charter Presentation (Thomas Narten) * Charter Discussion Chairs presented the Note Well. Thanks to Steve Ulrich for taking notes. --------------------------------- Thomas Narten NVO3 Problem Statement draft-narten-nvo3-overlay-problem-statement-01 ** high-level motivations review Q (unknown): question as to the nature of the network service - virtual network service. A: semantic differences ** VM requirements (tenant perspective) - address space independence ** VN requirements (DC perspective) - flexible VM placement - irrespective of the physical network boundaries for placement and migration - reference to ARMD and address resolution issues - VLAN space limits - L2 broadcast domain ** summary of requirements / conclusion - referenced the hesitance to embrace a BGP driven model COMMENT (John Scudder) - questioned the hesitancy re: BGP COMMENT (Rajiv Asati) - referenced the BGP comment again COMMENT (Kireeti Kompella): - comment re: BGP reticence again - question re: scale of millions of VMs COMMENT (Igor Gashinsky) - 400K entities within domain COMMENT (Kireeti Kompella) - interested in the scale requirements COMMENT (Igor) - 20K server config, 20 VMs / server - pointing at their current pod requirements COMMENT (unknown): - how does this rationalize against the SDNP BoF --------------------------------- Marc Lasserre Overlay Networking Framework draft-lasserre-nvo3-framework-01 http://www.ietf.org/proceedings/83/slides/slides-83-nvo3-0.ppt ** draft objectives ** reference --------------------------------- Igor Gashinsky NVO3 Requirements for Tunneling http://www.ietf.org/proceedings/83/slides/slides-83-nvo3-2.pptx COMMENT (unknown) - question about middleboxs - i didn't parse this --------------------------------- David Black VPN Interactions http://www.ietf.org/proceedings/83/slides/slides-83-nvo3-1.pptx ** questions time yielded --------------------------------- Lawrence Kreeger Network Virtualization Overlay Control Protocol Requirements draft-kreeger-nvo3-overlay-cp-00 ** questions Q (unknown): asking what the definition of hypervisor A: hypvervisor software instance that allows you to share resources on a single physical server --------------------------------- Thomas Narten Charter Presentation - L2oIP baseline with IPoIP as a potential optimization - intent is to develop layer agnostic architecture - reinforcement of the intention to consider requirements outside of the realm of the VM use cases - LISP may have utility - in providing map-n-encap mechanism. ack'd a concrete example of overlay mechanism. analysis required ** BGP discussion topic suggestion is that the BGP savvy folks drive a dialog between interested / concerned parties to better understand their requirements. --------------------------------- Charter Discussion ** ken calbert - question regarding whether or not there is any prioritization relative to the deliverables slide. encouraged review of the gap analysis. encouraged the review of prior work. ** dave dysan - ack's the utility of the work and feels it needs to be worked encouraged coverage of the following topics. - name space - what space around - control-plane - willing to provide text around control plane integration and consideration of middlebox coverage. (sounded like interest in service chaining use cases) ** kireeti kompella how much coverage re: storage is in scope david black - considers storage to be an IP problem. all storage is handled within the network. (L2 or L3) kireeti - hinting at the requirements for lossless FCoE in such an environment benson - acks we don't have answer to his question kireeti - how much L2 is in scope? kireeti - where does the overlay start/stop and what's in scope? kireeti - shades of inter-as vpns ... ** dmm - acks the need that the work be done - notes deliverables gap in the LISP - 2nd order impacts in the interactions. some of this is an interactions analysis - reviving the LISP interactions draft lofted taken to the list - http://tools.ietf.org/html/draft-meyer-loc-id-implications-01 ** thomas (orange) interested in use case w/L2 trunk between the VM host and the TOR which may be NVO3 aware. ** luigi - DT ... ** igor comments re: NIC offload - important to support both. iterated issues with $5K TOR having support for large MACs ** thomas narten - comments re: scope, scale and nature of an identifier - how much L2 - work order - best addressed in // - lofts the challenge of crafting the requirements doc w/i 6 mos. ** kireeti comment re: small amount of state to facilitate scale. w/corresponding discussion for caching. suggested that resurrection of the dmm draft be done. dino - comments re: state vs. stretch ** igor misc. comments re: convergence characteristics and state. ** XXX curiosity re: the hierarchical nature of the overlay ** XXX2 how to handle encapsulation mismatch and requirements therein. ** pat thaler (BRCM) ack re: need for a diversity of location and brief discussion relative motivations for encapsulation (ToR/edge device - hypvervisor) ** XXX3 (microsoft) ack re: need for work - then discussion re: encap placement and nature NIC and capabilities therein can be placed ** john hudson (BRCD) - agreement that there's validity in the work - feels that 6 mos might be too late to make this work. impending product availability within the next CY - emphasis in doing this right and simply ** rajiv asati - acks validity of work - encouraged coverage of the troubleshooting capabilities ** igor - ack's the need to outline the requirements for VPN and/or encap integration - feels that the charter requirements need to happen very quickly - proclivity for an IPoIP mode of operation to emerge ** XXX4 - looking for additional tunneling options ** sri (for VMware) - encouragement for normalization of nomenclature ** thomas narten - ack the requirement for OAM attention - security considerations require attention ** dino L2 vs. L3 - underlay vs. overlay - commentary re: MACs flat nature and the inability to do any aggregation. notes that IP based orientation is preferred from scaling perspective.