NFVRG IETF 95 Buenos Aires

Wednesday, 6 April 2016 16:20 Ð 17:20

Thursday, 7 April 2016 14:00 Ð 16:00

 

Chairs:  Diego Lopez (diego.r.lopez@telefonica.com)

                   Ramki Krishnan (ramki_krishnan@dell.com)

 

Note takers: Dirk Kutscher, Catalin Meirosu, R—bert Szab—

 

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

 

Welcome and administrative matters

Presenter: D. L—pez

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-0.pdf

-       Mailing list: https://www.irtf.org/mailman/listinfo/nfvrg

-       Web site: http://trac.tools.ietf.org/group/irtf/trac/wiki/nfvrg

-       Proceedings: https://www.ietf.org/proceedings/95/nfvrg.html

-       Dirk as note taker for the first session. Catalin and R—bert for the second session

-       Mehmet Ersue acted as Jabber scribe for the first session. Diego for the second session

-       The list of NFVRG and NFV research-related events will not be updated in this and following meetings. Too many already going on. Use the NFVRG list for announcements, limited exclusively to research-related events

 

Open Source and Project Updates

 

On OSM

Presenter: Diego L—pez

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-1.pdf

-       Open Source MANO stack aligned to ETSI NFV (information and data) models

-       Intended to be ready to be offered to cloud and service providers

-       Based on existing code seeds, already demonstrated

-       Discussion

o   Robert Szabo: do you consider hierarchies for resource orchestration?

¤  Not for the moment

 

VMs, Unikernels and Containers: Experiences on the Performance of Virtualization Technologies

Presenter: F. Huici

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-2.pdf

-       VMs provide good isolation but are heavyweight. Containers provide light-weight isolation

-       But there are ways between these two extremes

-       Discussion

o   Did you use CGroups with dockers or just LXC?

¤  Just LXC

o   Ramki: we have an active draft on containers. Found some security topics that should be discussed. Have you assessed security for different solutions?

¤  Focus was more on performance, haven't looked into security in detail

High-Performance Service Chaining

Presenter: D. Dolson

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-13.pdf

-       Zero-copy packet code

-       PCI pass-through ad SR-IOV

-       Slice your network traffic

-       Minimize East-West traffic and touches per packet

-       SFF within each SF thread

-       Minimize packet overhead

-       Discussion

o   Ramki: You are proposed co-locating SFF with SF. If you need an overlay, do consider doing this separately?

¤  We could do that in a separate thread

¤  Ramki: Unless overlay is combined with such as model, full deployment is difficult

o   Felipe HuicI: KVM and Open vSwitch what are you using?

¤  Linux-KVM

¤  Felipe: Have you looked into cost of different NFV functions?

¤  Yes, should be in report

o   Diego: When looking into performance, are you considering power cost per bit?

¤  Report has some information on this

 

PPA Project Announcement

Presenter: R. Krishnan

https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-7.pdf

-       Introducing Policy-based Platform-aware Scheduling

 

News from the RG chairs

 

Progress on Document Structuring

Presenter: R. Krishnan

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-3.pdf

-       The RG is implementing the one-document-per-area policy decided in Yokohama

-       The Policy-based Resource Management draft is targeted towards an RFC in Q1 2017

-       3 drafts adapted, which are relevant to the scope. Moving to github

-       Specific examples from other drafts could be extracted from other drafts and included in the area-specific draft

 

News and Proposals on the NFVRG Activities

Presenter: D. L—pez

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-4.pdf

-       Inter-RG collaboration

o   Received a proposal for a joint session on network management in Berlin (together with NMRG, NMLRG, SDNRG) and look at challenges for SDN/NFV. This coordination is still work in progress.

o   Started contacts with NWCRG to explore joint activities on opportunities opened by NFV for them. They asked for a presentation at the NFVRG meeting in Berlin.

-       Collaboration started with OPSAWG. NFV Operational Taxonomy, operational gap analysis; volunteers welcome - contact Diego

-       We received a request from IDR on the applicability of BGP to multi-domain NFV and SDN - Susan looking for operators or people that could make some tests (draft-hares-flowspec-combo-01)

-       Work ongoing in the BMWG regarding VNFBaaS.

-       Changing focus, ongoing near-term items:

o   policy-based resource management

o   analytics for visibility and orchestration

o   security

-       Proposal to change the inactive item with three new ones, in areas that have potential to be interesting for the community:

o   Service orchestration and lifecycle aspects: SDN Design patterns, DevOps and CI, FCAPS aspects

o   Real-time properties: monitoring, measurements, management, orchestration. Exemplary application - VoLTE, augmented reality

o   Reliability and fault detection in NFV-based an mixed environments

-       Documents are expected to be focusing on these items. Diego will send additional info on the mailing list

-       Discussion

o   Ahn Lee: is it possible to cover additional work item proposals?

¤  Possible, use the list.

¤  Ahn Lee: What timeline for sending the messages?

¤  Probably sending it next week, hoping for a discussion that lasts a couple of weeks. For additional items - write a short description of what you mean to address and why, not a draft

¤  Ahn Lee: Intention to send a few slides

 

Update on Area Drafts

 

draft-irtf-nfvrg-policy-based-resource-management

Presenter: R. Szab—

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-15.pdf

-       Motivation: consolidate contributions in the policy management area, on request from the RG chairs

-       Editors would like to summarize and harmonize the contributions. started with three adopted RG drafts (bottom-up approach). At the moment, content from two drafts and placeholders for content from the third one.

-       Within scope:

o   single domain, multi-domain, multi-admin

o   service assurance policies that can be delegated to RM

-       Out of scope:

o   VNF (service specific) config

-       Content like a whitepaper, the draft will promote alternative approaches as considered by the community; it could be potential input for IETF WGs on what might be standardized in the future.

-       Implementation examples and experimentation are part of the ToC

-       Next steps: Architecture section, depending on feedback on limiting the scope on the list

o   For the Resource Management in Service Chaining there will be no architecture considerations in the draft.

-       Ways of working: opening up the document by migrating it to Github. Will take contributions from suggestions given through NFVRG and the mailing list.

-       Discussion

o   Dean Bogdanovic: I don't understand why the NFV PoP would be constrained on CPU/storage; in my experience the network is the issue

¤  Diego: capacity in cloud environments comes with constraints - we're adding additional constraints to be considered in addition to network.

¤  Ramki: it is easier to optimize in large-DCs, while in-network DCs it is difficult and distributed DC is a different context.

¤  Robert: even in base stations there will be an execution environment.

¤  Dean: in PoPs and COs, there is more than enough physical space and power, so it's just a matter of providing the resources. Some of the base station compute could be put in these PoPs and COs; a CO is the size of this room.

o   Yun Bi: Scope seems a bit bold. Could you identify more precisely what will be done?

¤  Robert: We started with a bottom-up approach, and welcome suggestions on the mailing list. This goes well beyond onto architectural considerations. So where should the boundary driven in particular with respect to MANO

¤  Ramki: There is content in the draft, please send comments to the list.

o   John Strassner: Are you imagining a continuum of policies, or separate elements with defined interfaces

¤  Robert: we donÕt know yet.

¤  John: Then I might have something for you.

o   Dean: everybody talks about resource management, but nobody talks about capability management. The resources would be different from the capabilities executed on top of them

¤  Robert: I agree

¤  Ramki: the idea is to also be aware of other capabilities (hw accelerators)

¤  Dean: IPv4 and v6 would give different reliability over the same hw.

¤  Robert: certain level of capabilities should be considered

¤  Ramki: we should cover capabilities across the board, including Intel EPA. I will talk about an example we will bring, shortly.

¤  John Strassner: not only do you need to advertise capabilities, you need to advertise constraints.

¤  Dean Bodganovic: Intel has EPA, but it would be good to do it across the board because there are constraints from Cavium, etc.

 

Update on Area Drafts

 

draft-irtf-nfvrg-gaps-network-virtualization

Presenter: C. Bernardos

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-8.pdf

-       Organized the gaps against standardization & research initiatives. Went through the comments received from the mailing list

-       Next steps:

o   Include other areas in the analysis (IoT and NFV/SDN/MEC)

o   Change the document structure (major change), to per-topic rather than per-WG/SDO. As well identify key missing elements and future work

o   Broaden the documetn scope by identifying research challenges, potentially based on the gaps already pointed out. Asking for comments from the mailing list

-       Discussion

o   Diego: Will see how some of this work could be contributed to the OPS WG

 

draft-irtf-nfvrg-nfv-policy-arch

Presenter: R. Krishnan

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-16.pdf

-       DNew in the draft - merged the policy-based NFV placement and scheduling example from [1] onto section 8.2

-       Policies are written based on performance parameters provided by a "measurement collector"

-       Next step - draft content to be merged onto the Policy-Based Resource Management

 

draft-irtf-nfvrg-resource-management-service-chain

Presenter: S. Lee

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-5.pdf

-       Next steps - refining the system model, merge in the Policy-Based Resource Management draft in terms of architectural considerations, examples and framework

-       Discussion

o   Diego: How mature is the framework you are going to build?

¤  We have to model and refine it

o   Ramki: Are you considering something based on open source?

¤  We can use ODL, but it is different from our approach

 

draft-irtf-nfvrg-service-verification

Presenter: M-K Shin

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-6.pdf

-       Added a new challenging issue: Stateful VNFs with multiple physical views. States of all instances for one logical VNF must be considered when verifying correctness. Also important to know the ordering of the functions

-       Next steps: looking for input and additional drafts to merge with as part of the security area

-       Discussion

o   Diego: This is verification in advance of deployment?

¤  It is not clear how to achieve maintaining the dependency between different VNFs

¤  Ramki: this could be useful for both pre and post-deployment

 

draft-irtf-nfvrg-unify-recursive-programming

Presenter: R. Szab—

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-18.pdf

-       Changes: policy considerations - hierarchical resource management architecture, with tenant, operational and policies

-       Report on experimentation (actual report included in a draft sent the SFC WG, but we have a section here as well). used for SF-instantiation and SFP creation with implementing the API we propose over Openstack, Docker, and a Mininet environment for emulating the network.

-       Yang model for a programmatic API - as results from the experimentation. It is really convenient for the management in order to provide IP layer configuration in order to provide management plane configuration for the EM

-       There are pointers in the draft for components that are available as open source

-       Discussion

o   Ramki: Is the Yang data model focused on contributions for VNFs?

¤  We would like to jointly handle the VNFs and the forwarding overlay, and the Yang model tries to handle this. We also handle some of the inside characteristics (like whether there is acceleration for a VNF in one domain but not another one)

o   Ramki: Think about the I2NSF work, some of this seems related

¤  Will do

 

Update on Existing Drafts

 

draft-cai-nfvrg-recursive-monitor

Presenter: C. Meirosu

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-10.pdf

-       Discussion

o   John Strassner: extended Datalog, what way?

¤  Construct to model the nodes and relationships in the graph

¤  John Strassner: Check Openstack congress language, which might be easier to you; please feel to send questions

¤  Ramki: Datalog wrappers are created in in open source communities

 

draft-unify-nfvrg-devops

Presenter: C. Meirosu

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-21.pdf

-       Discussion

o   Dean Bogdanovic: Working operational data, this is not necessary the right RG for this work

¤  Addressing challenges

¤  How do you do that if the information sources are not here

¤  Ramki: scope is in the operational side.

¤  Catalin: how to share relevant data sets and terminate when tests are done; related issues privacy and trust

o   Steven Wright: challenges exporting raw data between autonomous entities; there are methods anonymizing data. DevOps challenge is to do this in an organized way, business agreements may be needed in place

o   Diego: DevOps is part of the new proposed areas. This is really interesting

Introduction to New Drafts

 

draft-bernardos-nfvrg-multidomain

Presenter: C. Bernardos

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-9.pdf

-       Discussion

o   Dean: How do you combine hierarchical with cascading?

¤  We are starting now, so we do not have a solution yet

¤  Dean Bogdanovic : Any analysis on hierarchical vs. cascading

¤  Carlos: we are doing it

o   Ramki: entry point in cascading one?

 

draft-liu-nfvrg-analytic-framework

Presenter: V. Liu

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-17.pdf

-       Trying to put analytics part in the NFVO in order to allow the NFVO to trigger policy by itself

-       Three parts in the draft: Monitoring and data collection and policy inventory: how to do the collection, what parameters to be collected, what to do with it. Developed a collection framework to collect all data from the systems

-       Analytics model, based on K-nearest neighbor

-       The analytics architecture is based on the OPEN-O implementation

-       Next steps: refine the parameter inventory for the monitoring data, policy generation and selection and more real-time model and non-real-time models

-       Discussion

o   John Strassner: Did you look at variants of KNN, because the basic version is memory intensive for large datasets. You might consider having an inner and outer loop for integrating the analytics, with the outer loop trying to fine tune results from the inner loop.

o   Ramki: K-n algorighm. Are you using it for placement?

¤  Yes

 

Update on OPNFV Brahmaputra Release

Presenter: H. Kirksey

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-12.pdf

-       OPNFV looking to increase the adoption of NFV

-       Focusing a lot on integration upstream, and trying to avoid reinventing the wheel

-       Brahmaputra release had a major simultaneous release process

-       Performing unit testing and including a number of performance benchmarking projects.

-       Able to realize fault management (from 7-8 minutes to about 1 second)

-       Able to demonstrate SFC and L3VPN, as well as work on resource virtualization by creating a layer on top of OpenStack

-       OPNFV has deployed OpenStack in both bare metal environments and integrated with several controllers, in addition to the standard Neutron network plugin

-       Discussion

o   Linda Dunbar: OPNFV does not really do code

¤  We might host a project from OpenSAF, but we do not implement if not needed

¤  Linda: who deployed it?

¤  We have a distributed test lab around the world, including bare metal capabilities where we deploy to

 

(Late) Update on the Open-O Project

Presenter: H. Deng

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-20.pdf

 

Applying BGP FS in NFV and SDN Environments

Not presented, but referenced in the second session agenda discussion

Slides: https://www.ietf.org/proceedings/95/slides/slides-95-nfvrg-19.pdf