COINRG

Summary

The COINRG interim meeting focused on a rich discussion about the scope of the research group, particularly in the context of routing, in-band processing, and data plane programmability. Presentations included "Semantic Routing" by Adrian Farrel, a summary and forward-looking perspective on COINRG's scope by Dirk Trossen, "Extensible In-band Processing" by Stefano Salsano, and "Flight Plan" on data plane disaggregation by Nick Sultana. The session concluded with a presentation by Nicoletta Randarisa on a student project applying computing in the network to smart city traffic management. A key theme throughout the discussions was the intersection of data plane programmability with traditional network functions like routing, and how research in COINRG can contribute to understanding the implications and potential new capabilities.

Key Discussion Points

Semantic Routing (Adrian Farrel)

Definition and Scope: Semantic routing at the IP level (Layer 3, next-hop forwarding) aims to make forwarding decisions based on information not traditionally used for this purpose (e.g., information carried in packets, existing fields, or new fields). This is intended to enable varying types of service for microflows and better utilization of network resources.
Relation to Network Programming: SDN-based approaches (centralized or distributed routing, pushing forwarding tables) are considered one way to achieve semantic routing. The question for COINRG is whether it's interested in the SDN/programming aspects, especially novel routing algorithms and related issues.
Relation to Computing in the Network: Semantic routing can leverage "computing in the network" by using installed algorithms to build forwarding tables or actively determine per-packet forwarding actions. This is distinct from "active networking" where programs are carried in packets. The research question for COINRG is around programmed algorithms and "mini-compute" on devices, and the programmability required or driven by semantic routing.
Community Feedback: Adrian summarized mailing list discussions, highlighting the need for a unifying abstraction or formal definition of semantic routing, and distinguishing it from existing practices. Concerns were raised about the security implications, particularly with encrypted traffic (like QUIC), and the applicability domain (enterprise vs. inter-domain).
Mauricio Say's reframing: Suggested that for COINRG, semantic routing could be framed as a compelling use case for exploring the implications of packet data plane programmability on routing.

COINRG Scope and Way Forward (Dirk Trossen)

Dirk presented an individual understanding of COINRG's core concepts, intended methods, and potential applicability areas, based on the charter, mailing list discussions, and community input.
Core Concepts: Computing in the Network (as per charter, distinct from active networking), programmability of the data plane, cloud-edge continuum, orchestration, and decentralization.
Methods: Use case-driven requirements analysis (as seen in the ongoing use case draft), identifying benefits, researching new languages/abstractions, and engaging with other IRTF RGs and the IETF (e.g., ICN, energy, privacy, security considerations).
Applicability Areas: Dirk categorized the charter's scope into Vision & Technologies (for compute in the network, distributed computing frameworks/languages) and Applicability Areas (Transport, Privacy, Data Discovery – currently dormant, Routing).
Questions for the Community: Dirk posed questions to guide future work, including:
- How to delineate COINRG's vision from related fields like edge computing.
- What technologies are essential for enabling the COIN vision.
- Which applicability areas (e.g., transport, routing, privacy) should COINRG focus on, and what are the key COIN-specific questions within those areas.
- How to connect COINRG's research to other IRTF RGs and the IETF for early exposure to engineering communities.

Extensible In-band Processing (EIP) (Stefano Salsano)

Goal: To define a generic and extensible mechanism to carry information in IPv6 packet headers, allowing nodes to read, write, and make processing decisions based on this information. This aims to overcome the "ossification" of existing IPv6 extension headers.
Use Cases: Stefano presented examples like contractual networking, deterministic networking, and network slicing, noting that semantic routing could also be a use case for transporting additional information. He requested feedback on other useful use cases.
Domain: The proposed solution is initially targeted at limited or controlled domains rather than the entire internet.
Technical Approach: EIP information could be carried in IPv6 Hop-by-Hop options, Destination options, or Segment Routing Headers (e.g., as TLVs). Content would be use case-driven but could include common elements like timestamping, HMAC-based authentication, or contract identifiers.
Challenges: Stefano addressed wire speed processing (requiring hardware support) and security/privacy concerns, stating that solutions like SRv6's HMAC model could be adapted for limited domains.
Prototype: Stefano announced an open-source prototype for EIP packet processing in Linux, based on eBPF/XDP, aiming for a quick development turnaround. He asked about interest in a P4 prototype.
Discussion with Chairs: Mauricio Say questioned what Stefano expected from COINRG, suggesting that Stefano should identify how EIP aligns with COINRG's existing use cases and what research EIP encourages. Dirk Trossen encouraged Stefano to propose a specific, compelling use case for EIP to be added to the COINRG use case document.

Flight Plan: Data Plane Disaggregation (Nick Sultana)

Problem: The current paradigm for P4 programming is one-to-one (a single program for a single data plane device), which is a mismatch for the broader vision of "programming the network." This leads to exclusivity of resources and a piecemeal approach.
Proposed Paradigm (Data Plane Disaggregation): Flight Plan enables writing a single logical data plane program, which is then automatically split and mapped to a heterogeneous mix of physical data planes (e.g., switches, smart NICs, CPUs) with different capabilities. The goal is to make the network act as "one big programmable switch" without manual effort.
Research Contribution: The system provides automation for splitting, placing, and linking program segments across distributed data planes, considering resource constraints, performance objectives, and potential faults. It works within existing vendor toolchains and hardware.
Key Ideas: Abstraction of programs (using annotations in P4) and hardware features (profiling function execution on different platforms) into a rule-based representation for allocation decisions.
Routing Component: Flight Plan uses a routing component to guide computation through the network, allowing the runtime to make switches more autonomous. Nick highlighted the research space in optimizing across different time scales of forwarding and routing.
Evaluation & Future Work: The system was evaluated through simulation and physical hardware, demonstrating the interoperability of multiple programs, splits, and runtimes. Nick's future work includes extending network programming to include application-level concerns, where the network helps facilitate the realization of application needs.
Discussion: Dirk Trossen commended the work as a "next step" towards a more powerful, networked approach to data plane programmability. Nick provided an example showing how Flight Plan could integrate both network functions (header compression, FEC) and application logic (in-network key-value cache) to improve application performance. Eve Schuler appreciated the "routing in the service of compute" perspective.

Smart City / Smart Lights (Nicoletta Randarisa and students for Mauricio Say)

This was a presentation of a student project from a class on computing in the network, focusing on how data plane programmability could impact applications.
Project: An AI-powered traffic management system for smart cities to optimize traffic flow and address environmental concerns.
Methodology: The project proposed an initial study using cameras to gather city data, which would then be used to train machine learning algorithms to optimize traffic, identify accidents, and plan alternative paths.
Architecture: A distributed architecture was envisioned with local servers (fog computing/edge computing) optimizing traffic for neighborhoods, connected to central main servers (cloud computing) for city-wide optimization. Smart cameras would provide real-time data for monitoring.
Mauricio Say highlighted this project as an example of how the discussions in COINRG about enabling technologies can directly lead to innovative application-level solutions, and thanked the students for their work, emphasizing the role of teaching in expanding the COINRG community.

Decisions and Action Items

Stefano Salsano: Will follow up in the coming weeks/months by identifying a specific, compelling use case for Extensible In-band Processing (EIP) that aligns with the COINRG use case document and cries for a solution like EIP.
Nick Sultana: Invited to potentially present his new work on extending Flight Plan to include application-level concerns at a future COINRG meeting, as it links data plane disaggregation to broader research group interests.

Next Steps

Continue discussions on the mailing list regarding the COINRG scope, specific applicability areas, and the key research questions within them.
Presenters are encouraged to continue collaborating and contributing to the research group.
The next COINRG meeting will be held in approximately one month (at IETF 113), in a hybrid format. Ideas for presentations and contributions are welcomed.

Session Date/Time: 10 Feb 2022 15:00