IETF125 ICNRG Meeting
Session: Monday Session III
Date Time: 16/06/2026 1400
Chairs: Dave Oran, Ryo Yanagida
Notetakers: Xinjiao Li, Yitong Li
Presentation 1: In-Network Retransmission Control in ICN
Speaker: Kazuhisa Matsuzono
Presentation 2: Reflexive forwarding implementation on Cefore (with demo)
Speaker: Yohei Okamoto
Presentation 3: ICN Service Mesh for plug-and-play ICN
Speaker: Kenji Kanai
Discussion
- Question 1 (Ryo Yanagida): Concerns about transition overhead
while bridging Http to ICN. Is using an HTTP API gateway a
reasonable transitional approach to bridge current networks and
future ICN-based transport, even if it introduces some overhead?
- Answer 1: The goal is not to replace existing applications
immediately but to enable ICN gradually (e.g., via a service
mesh), allowing systems to transition from HTTP-based
communication to ICN transport over time.
Speaker: Shintaro Mori
Discussion
-
Question 1 (Ryo Yanagida): How is the overhearing effect
actually used in the network layer of ICN—especially in terms of
naming and data handling?
- Answer 1: Wireless nodes can overhear broadcast packets and
cache the named content. Later, they can serve the same data if
requested, though integrating this with ICN forwarding is
technically challenging.
-
Question 2 (Dave Oran):In broadcast networks with opportunistic
caching, multiple nodes may cache the same data. How do you decide
which node should respond to a request without causing collisions?
- Answer 2: Offline discussion later.
Speaker: Xiang Li
Discussion
-
Question 1-1 (Rod Van Meter): How do you manage software
versions and ensure the right code runs on the right compute node as
computation moves across the network?
- Answer 1-1: A resource management node collects server status
information, including service/software distribution and
available computing power, and uses that to assign user requests
to appropriate servers.
-
Question 1-2: How does this relate to older distributed systems work
from the 1980s–1990s?
- Answer 1-2: The authors have looked at prior distributed
computing architectures. Their approach differs by using an
ICN-based design and distributing resource management across
nodes to make the system more flexible and decentralized,
especially at the edge.
-
Question 2 (Ryo Yanagida): Does managing multiple identifiers
(and mapping between them) introduce overhead and complexity?
- Answer 2: Yes, but this is temporary. The current design keeps
multiple identifiers to explore their relationships; the goal is
to eventually simplify the system and converge on a smaller,
unified namespace.
-
Question 3 (Dirk Kutscher): Why split the system between IP
addresses and names? Why are IP addresses still needed?
- Answer 3: Names identify services, while IP addresses are kept
for two practical reasons: to associate computing resources with
specific nodes/locations, and to return computation results
directly and efficiently. The authors see this as an interim
design and expect the naming spaces may be simplified or merged
later.