ICNRG Meeting Notes - IETF 113 (Hybrid) Vienna
2022-03-25, 11:30 to 13:30 UTC
Thanks to Matthias Wählish for being in the room!
Chairs’ Presentation: Status, Updates
Good progress on documents recently and more in the pipe.
Important: IPR issues in documents.
Comment - Colin Perkins: An IPR declaration was made on an individual
draft and not on the RG draft -needs clarify and potentially update to
reflect the RG version.
Comment - Dirk Trossen: Draft issued at the previous company and needs
to be cleared with them.
Comment - Colin Perkins: The company already has made a declaration it
just needs to be clarified.
Q - Colin Perkins: NRS consideration it's the RFC?
A - Dirk Kutscher: Yes.
NDN for Data Intensive Science Experiments (N-DISE) - Edmund Yeh
Has been going on for a year.
Q - Ken Calvert: All these are all running over Internet 2? Is there
layer 2 stiching?
A- Edmund Yeh: Yes they are running over Internet 2 and ISnet. And in
layer 2. The architecture shows VLANs that are provisioned (may take
months).
Q - Ken Calvert: These are dedicated hosts at each site?
A - Edmund Yeh: The hosts at participant sites are dedicated but at
Starlight it is shared but the loads are low.
Q - Dave Oran: Is this caching in DRAM only or there is there I/O to
external storage involved?
A - Edmund Yeh: This is DRAM caching.
Q - Dirk Kutscher: A lot of work has been done to set up the network
and accelerate the data transfers. What will you do with the data?
A - Edmund Yeh: The proposal had a section on joint data movement and
scheduling then budget was cut and the data processesing/work flow was
also cut. But it's the ultimate goal.
Q- Thomas Schmidt: It is surprising that the popularity of requests is
Zipf like a Netflix rating.
A - Edmund Yeh: An analysis of requests in high energy physics was
done. The "hot sets" are very much following Zipf - like the most
requested in Netflix.
Q - Ken Calvert: We are doing similar work from Sandi - in the 3 hop
scenario did you develop a bespoke or are you using a generic algorithm
to break up the chunks?
A - Edmund Yeh: Chunking is inherent in the application (ask student).
Will have a discussion off line.
A - Yuanhao Wu: Chunking is provided from the DBDK NIST group.
Q- Oran: 1. At 21 Gbps do you do signature validation? You are not
using the integrity features of NDN?
A - Edmund Yeh: No we don't. The provenance checking will be part of
the future. High energy physics has higher level access securiy but
packets will be signed in the future.
Q- Dave Oran: 2. 4X improvement for FPGA is not very good - what is
the bottleneck? I am interested in the work.
A - Edmund Yeh: Those are reliminary results and the work is
continuing in Jason (?)'s' group.
Comment - Dave Oran: Not exploiting parallelism is a likely cause
(like in network coding) because the clock itself is low - will send
email to continue the discussion.
Comment - Edmund Yeh: The Work has focused on the caching but in the
next months, the work on the FPGA will continue.
Alternative Delta Time Encoding for CCNx Update - Cenk Gündoğan
Comment - Dirk Kutscher: This is the right type of document for a RG
the work itself and the protocol extensibility. It should be accepted as
a RG document. Comment needed from the group/on the list.
Ping & Traceroute Update - Spyridon Mastorakis
Comment - Dave Oran: The other co-chair will have to lead the review
since I am co-author. Also, comment to Colin Perkins: the same person
should review both documents since they go together (Colin says: makes
sense).
Path Steering Refresher - Dave Oran
Q - Ken Calvert: I have not read the draft. Picture of the encoding?
The idea that the next hop is in charge?
A - Dave Oran: Yes it's just a face pointer in the FIB entry
Q - Ken Calvert: Is this at the AS level or node level?
A - Dave Oran: The intent is for diagnostic so every hop would be
identified - and please read the draft!
Comment - Dirk Kutscher: This is a powerful tool that has the
consumers influence how they use the network.
Comment - Dave Oran: Less messy than IP as it uses the symmetric
routing of ICN.
Reflexive Forwarding Re-Design - Dirk Kutscher
Q - Thomas Schmidt: Every consumer can send a reflexive demand at a
producer. Isn't this like DOS?
A - Dirk Kutscher: That's something we explicitly avoid. Will talk
about it later.
Q - Thomas Schmidt: Have the producer has seen interest 1?
A - Dirk Kutscher: Yes
Comment Dave Oran: We dont use FIB on the reflexive back we use the
PIT entry of the original interest to find the ingress face to use as
the next hop.
Comment Dirk Kutscher: Uses the symmetric forwarding properties.
Q - Ken Calvert: You are using the PIT to forward interest? What is
the difference with RICE?
A - Dirk Kutscher: In RICE we used dynamic FIOB modification behavior
for the interests that have the token?
A - Dirk Kutscher: It's a standard interest with RPT field but with a
different name.
... I missed a comment/question ...
Wrap-Up, Next Steps - Chairs
Comment: Ken Calvert: Going through FLIC - will update.