Minutes IETF98: mboned
||Minutes IETF98: mboned
IETF 98 Chicago
Tues, Mar 28, 2017
Zurich B (Held jointly with PIM WG)
Note takers: Toerless Ecker, Joel Jaeggli
Jabber Log: https://www.ietf.org/jabber/logs/mboned/2017-03-28.html
Audio log: https://ietf.org/audio/ietf98/ietf98-zurichb-20170328-0900.mp3
Video log: https://www.youtube.com/watch?v=UaI5E-JWjiY
09:00 Status of WG items
Chairs, 5 min
- ready for last call ?
- in IESG process - Tim Chown and Mikael Abrahamson are doc shepherds
- sandy to discuss
Zhang, 10 min
- Sandy Zhang, ZTE presenting
- Trying to create higher level abstraction than existing per-protocol
Slide 2: What is the multicast model
Slide 3: Multicast Information Model 01 update
- model has been verified in ODL BIER project
Slide 4: Multicast Information Model 01 update (example usage, picutre)
Slide 5: open daylight
Slide 6: BIER project in ODL: wiki.opendaylight.org/view/BIER:Main
Slide 7: Multicast UML Class Diagram
Slide 8: Multicast Data Model overview
Slide 9: Multicast Data Model - Information
Slide 10: Multicast Data Model - Overlay
Slide 11: Multicast Data Model - Transport
Slide 12: Multicast Data Moel - Underlay
Slide 13: Next steps
Jake Holland, Akamai: What about IGMP proxying, MD proxying ?
Sandy: into transport section ?! yes, can be done.
Greg: how many people have read the draft?
how many folks paying attention to Yang work ? 25%
Jake liked the pictures that helped him to understand relationships between
ca. 6 people supporting adoption, none opposed
Call for adoption will be taken to the list.
toerless Q: any other examples of this type of more compound data models in
IETG ? Lenny: as opposed to the single protocol? Stig Venaas: BESS is also
working on service model for VPNs, not only for multicast Joel Jaeggli: service
models is the area most similar to this. multiple pieces required that need to
be tied together. Fair number of those. Success of that entire area is not
guaranteed though, so we do not know if this would work very well. Once you
have a bunch of primitives sticking them together into complex structure.
Pioneering yes, but not alone!
Asaeda, 5 min
Asaeda NICHT, co-author: Kerry Meyer Cisco
Slide 1: Change since last version
SLide 2/3: old vs. new formula calculating arrival time.
Slide 4: Next step
- WGLC (really)
Chair(Lenny): revised multiple times. how many readers of last version ? A: 3
people in room. Stig: it has been multiple last calls, now read it in a lot of
detail, very good shape now, really should get through last call now.
AMT Hackathon Update
Holland(AKAMAI)/Pardue(BBC), 15 min
Lenny: explaining multicast hackathon doing AMT hacking lead by Jake Holland
@AKAMAI, making multicast cool again https://github.com/GrumpyOldTroll/amt
Slide 1: Planned Deployment Overview picture
- both cisco & juniper having virtual router implementations
- Akamai looking for ISP partners to participate in these experiment
- Open source C implementation that a few years back as starting point, good
starting point. - hit some compatibility issues with cisco implementation,
introduced compatibility knob
Slide 2: Lab setup
- mainly working on porting on relay code into openwrt, result checked in.
- running a cople of video streaming apps, including octoshape, also BBC player
Slide 3: Dynamic Adaptive Streaming over Multicast - from BBC
- ways to improve relaying HTTP semantics over multicast
Slide 4: summary of hackathon:
ported amtrelayd to OpenWRT, bugfixes, cleanup
- testbed setup documented
- expreiments running video - native multicast, amt-encapsulated
Slide 5: acknowledgement
- lucas pardue, BBC (multicast video),
- codarren velvindron, orange (remote), installing AMT
- prior work: mboned WG members, initial AMT project
- bill atwood, concordia - previous work on VLC integration
Donald Sharp: Whats difference between AMT and most kernels VIF tunnels, eg:
auto-tunnel across kernel. A(Jake): yes, lots of other tunnel options. One of
main benefits of AMT is that someone remote can join without need of
coordination, eg: other side does not have to be configured for it. Easier to
communicate with remote partners.
lenny: What is the state of AMT gateway implementations ? Know work by UT,
concordia, announced during last few years. A/Jake: Start with rtp and try
that. Better way is to include gateway into applications. Akamai has done this
multiple times, eg: Octoshape, also prior Akamai implementation. Work done by
akamai more focussed on segmented delivery. Not been looking into RTP yet.
Cisco/Juniper supporting relay implementations. Amazon EC2 has CSR1kv that
should include Cisco AMT, as of eg: november last year. Spin up relay and try
it. Cisco has gateway implementation in CSR1kv as well. Can also work on
gateway implementation in open source, needs some work. Currently relays on PF,
not ready for prime time.
toerless: also done open source host based gateway implementation:
bill atwood: has not touched the open source work in a couple of years. But
worked on security in the meantime. because AMT is like all of IP unsecured
packet, that was needed.
Jake: Was looking into video level encryption. Some work to do on that.
Concrened about AMT data packets. Original intent was for AMT to be closer to
the edge. But now in Akamais case going through the core, there is a risk for
data injection, no nonces in data packets, so attackers could just send packets
(guessing just the five tuples).
Will also have table at bit&bytes
Lenny: how many folks working at hackathon, A: 3, 2 local, 1 remote.
Lenny: final remarks:
For MBoned investigating IETF to add secretary. Great opportunity, come see
chairs. Last night there was a video bar bof on Video over the Interner, Greg
can talk about it, multicast topics came up.
No problem statement. Description of BOF caused a lot of interest.
Dave Oran had broader perspective:
IETF is not addressing codecs, network layer, etc. pp. What was presented was
just one slice in it. Lot of discussion about scaleable multicast, multicast
video and oversubscription/congestion.
Jake Holland: There was strong support to address scaleable video problem, some
folks see it as deep caching.
Toerless: overall solution will very much depend on the amount of end-to-end
delay the solution is willing to accept, eg: replay of video where any delay,
30 seconds is acceptable vs. live video.
Mike Mcbride: Trying to understand scope, where theree high bandwidth
requirements ? Greg: no mostly spinning wheels. Jake: There where 4 drafts in
the BAR bof - invitation for them was sent to 98attendees mailing list.
Mike Mcbride talking about VR applications and its immense bandwidth
requirements. Toerless: alsos ee VR presentation LIin Han at Mondays ICCRG
Joel: QUIC exists because similar problems exist with unicast. Solution space
being explored, he'd be surprised if it wasn't overflowing into multicast side.
Joel Jaeggli retiring as AD in wednesday, handing over to Warren Kumari.
Warren Kumari introducing himself. Has not been involved in this much at all.
- Welcome aboard.
Donald Sharp: Multicast forwarding in the linux kernel sucks due to hash
table. Lot of multicast testing in linux kernel: hash table based on high order
byte of multicast group. Performance on linux would suck very much. Problem was
fixed in 4.11 kernel.
Lenny: please mention this also on mboned mailing list.