Markdown Version | Session Recording

Session Date/Time: 10 Dec 2021 16:00

ICNRG

Summary

The ICNRG session included an update on the group's recent activities, including two new RFC publications and several drafts nearing completion or requiring further review. Three presentations followed: Carlos Gamarais introduced Xeno, a data-centric solution unifying data in motion, in use, and at rest; Red Samson and Hami Locker presented GT Systems' SPAN Network Architecture, discussing a hybrid approach integrating IPFS and NDN for distributed content; and Junxiao Shi detailed NDNts API design and challenges for building NDN libraries for the web. Key discussions revolved around technical implementations, security considerations, and the unique constraints of different deployment environments.

Key Discussion Points

ICNRG Status Update

• The research group celebrated the publication of two new RFCs:
  • RFC 9138: "Design Considerations for Name Resolution Service in ICN"
  • RFC 9139: "Adaptation of CCNx to Low-Power Wireless Personal Area Networks (LoWPAN)" – recognized for its potential in constrained networks and IoT.
• Several drafts are nearing publication:
  • "MNS architecture considerations" is completed and awaiting publication.
  • "ICN in LTE/4G networks" is awaiting final reviewer comments.
  • "CCN Info," a specification for a tool to collect information about elements on a path in a CCN network, is ready for IRSG review after minor comments were addressed.
• Call for Reviews: The "Flick specification," a manifest technology for ICN, requires thorough review, particularly on newly rewritten sections, with a goal for publication early next year.
• Two other documents, "Traceroute" and "Ping" for ICN, are technically mostly complete and will soon proceed to a Last Call on the mailing list.
• Ongoing drafts include "Path steering" (a soft state approach for path-sticking), "Reflexive forwarding" (for symmetric forwarding, used in RISE), and "Time TLV" (a new TLV structure for recording time, intended for standalone publication).
• Future Work Topics: Initial thoughts for upcoming work include:
  • Further exploration of ICN in IoT, particularly with LoRa systems.
  • Developing best practices or convergence for supporting applications with ICN.
  • Rethinking distributed computing with ICN as an enabler, building on existing RISE and CFN work.
  • Taking the next step on Quality of Service (QoS) in ICN, leveraging insights from RFC 1964 to specify mechanisms.
• Participants were encouraged to propose additional topics for discussion.

Presentation: Xeno by Carlos Gamarais (Adlink)

• Carlos Gamarais presented Xeno, described as a data-centric solution designed to unify data in motion, in use, and at rest by blending publish/subscribe and distributed query patterns, with built-in support for geo-distributed storage and computations.
• Xeno's architecture features a layered approach: a protocol layer (operating over various network stack layers), data transportation primitives (content-agnostic), and data-oriented abstractions (content-aware APIs).
• Naming in Xeno is inspired by ICN, using "name segments" grouped into "keys" with wildcards (* for a single segment, ** for zero or more). Fine-grained data selection can be achieved with predicates, projections, and properties executed at the query or publish node.
• The system defines various entities (Clients, Peers, Routers) and operations (scout for discovery, init/open/close for hop-by-hop sessions, declare/undeclare for optimizing entity roles, and write/pull/query for data exchange).
• Distributed queries are handled with consolidation strategies (none, newest, full) to provide eventual consistency, applied at different points in the network path.
• Xeno is transport-agnostic, having been tested over protocols like UDP, TCP, TLS, QUIC, Thread, and Bluetooth, creating an abstract overlay network.
• Key highlights include high efficiency (minimal 4-6 bytes wire overhead), support for the cloud-to-thing continuum, both push and pull models, and features like zero-copy, reliability, fragmentation, and batching.
• Discussion:
  • A participant raised concerns about integrating Xeno with existing industrial buses (e.g., Bacnet) where low-level data access is challenging. Carlos noted current work focuses on microcontrollers with network capabilities rather than direct integration with legacy industrial systems but expressed interest in offline collaboration on this challenge.
  • Regarding security, Carlos clarified that Xeno does not provide built-in content integrity mechanisms like some ICN protocols; application-level handling of integrity and signatures is expected. User access control and link authentication are part of the roadmap.
  • It was noted that Xeno has not been implemented on top of NDN or CCN, as it aims to provide similar core functionality with a focus on low wire overhead.

Presentation: GT Systems SPAN Network Architecture by Red Samson and Hami Locker

• Red Samson and Hami Locker presented GT Systems' SPAN Network Architecture, which evolved from efforts to improve adaptive bitrate video streaming.
• Their core concept, "secure peer assist," involves content slicing and tagging (with hashes), intelligent routing based on these tags, and a peer-to-peer approach with super-nodes to form a mesh for distributed storage and computing.
• The presenters discussed the potential of integrating IPFS (for its distributed storage capabilities) with NDN/CCN (for efficient distribution). A unified namespace was proposed as a critical area for standardization to integrate these approaches effectively.
• The "Hybrid Adaptive Routing System" combines name resolution (for persistent distributed storage) and name-based routing (for fast access and delivery), with NDN-like routing logic, and is designed to operate at the network layer.
• Modeled benefits include significant cost reduction (up to 50%), reduced latency by placing data and functions at the network edge, and enhanced security through content encryption and NDN's packet signing mechanisms.
• Hami Locker outlined a multi-stage optimization model:
  1. Optimal placement of data and functions (compute/storage) across core, edge, and peer levels.
  2. Exploiting network programmability and virtualization for elasticity and reduced over-provisioning.
  3. Integrating name resolution and name-based routing for a "best of both worlds" solution.
  4. Utilizing AI (combining global optimization with local agents) for better resource allocation and routing decisions.
• Red Samson further elaborated on the concept of distributed AI agents to manage network optimization, allowing local, trained agents to feed information to a broader optimizing intelligence, aiming for a "universal content distribution network."
• A potential real-world application involves partnering with a greenfield network provider deploying satellites in Medium Earth Orbit, seeking to re-architect their backbone away from traditional IP.
• The technical implementation envisages running a custom protocol stack alongside a classic TCP/IP stack on open routers, with NDN handling tunneling requirements.
• The presenters also touched on implications for the metaverse, suggesting that a network acting as a distributed cloud computer would fundamentally rely on standardized publish/subscribe operations.
• Discussion: Due to time constraints, Dirk Koscher encouraged participants with detailed technical questions to reach out directly to Red Samson and Hami Locker, and noted that a white paper is available at gtsystems.io/whitepaper.

Presentation: NDNts API Design and Challenges for the Web by Junxiao Shi

• Junxiao Shi presented NDNts, a set of NDN libraries implemented in modern JavaScript/TypeScript, designed to run in Node.js and web browsers, capable of operating standalone or connecting to NDN forwarders.
• The talk focused on low-level API design (packet decoding, fragmentation, retransmission) and the unique challenges posed by the web environment, primarily code size and the browser's role as an operating system.
• Junxiao emphasized the importance of good low-level API design, as interactions with these are unavoidable for library developers and specialized applications, even with high-level abstractions available. NDNts benefits from observing pain points in existing libraries.
• For TLV decoding, NDNts employs an "Evolvability-aware TLV Decoder" (EvDecoder) that automatically handles the NDN specification's requirements for unrecognized or out-of-order TLV elements.
• The traditional "Face" concept in NDN libraries is evolved into "Endpoint" in NDNts. An Endpoint offers enhanced capabilities such as automatic Interest retransmission, Data verification, Data buffering (for multi-segment responses), and Data signing/prefix announcements, allowing application developers to focus more on their core logic.
• Web Challenges: Code size is a primary concern for web applications (target of 170KB compressed for a 5-second "Time to Interactive" load). NDNts addresses this by making certain complex features, like data buffering, optional imports rather than including them in the core library.
• NDNts supports various transports under the Endpoint layer, including HTTP/3 (over QUIC/UDP) recently added for more efficient video streaming, avoiding TCP-over-TCP issues common with WebSockets. WebRTC transport for browser-to-browser communication is planned.
• Security: NDNts' keychain uses the Web Crypto API and IndexedDB in browsers, generating non-extractable keys for enhanced protection.
• Security Limitations:
  • The Web Crypto API requires a secure context (HTTPS), which prevents NDNts security features from working in low-internet environments (e.g., public Wi-Fi hotspots) where plain HTTP might be used. No solution has been found for this.
  • Web Crypto's limited support for cryptographic algorithms can hinder interoperability with other NDN specifications (e.g., FLIC's AES-CCM, which is not supported).
• User Naming/Identity: For anonymous users, NDNts generates random, self-renewing identity names and certificates. For authenticated users, the challenge is to provide a streamlined user experience while integrating with NDN-based authentication.
• Discussion:
  • Junxiao clarified that NDNts transports automatically attempt to reconnect on failure with exponential backoff, though some (like Web Bluetooth) may have device-specific limitations.
  • The primary motivation for NDNts is to provide the first comprehensive NDN library capable of running directly in web browsers, enabling experimentation and development of web applications on NDN.

Decisions and Action Items

• Action Item: Jose indicated he would contact Carlos Gamarais offline to discuss Zeno's integration with existing industrial systems and provide specific examples of data exposure challenges.
• Action Item: Individuals with questions for Red Samson and Hami Locker about the GT Systems SPAN Network Architecture are encouraged to reach out to them directly. The white paper is available at gtsystems.io/whitepaper.
• Action Item: Red Samson will share the URL for ACM ICN 2022 to the ICNRG mailing list if it is publicly available.

Next Steps

• Future Meeting Schedule:
  • If the IETF 113 meeting in March is held in person, the ICNRG will aim to meet during that time.
  • If IETF 113 remains online, an online interim meeting will be scheduled before or after the IETF week, similar to the current format.
• Community Engagement: Continued discussions on the technical content presented are encouraged via the ICNRG mailing list or by contacting the presenters directly.
• Upcoming Conference: Participants are reminded of ACM ICN 2022, which will take place in Osaka, Japan. The Call for Papers and other details are expected to be announced soon.