IETF 121 SRv6OPS WG Draft Minutes
Tuesday, 5th Nov 2024
16:30 - 17:30 (local/UTC, room Liffey A)
Introduction (5 mins)
1.1. Administrivia, Agenda Bashing (Chairs)
Operator Presentation (55 mins)
Each presenter has 10-15 minutes including Q&A to share their deployment experience.
2.1 Service Programming (Daniel Bernier, Bell Canada)
This talk will provide details on the “SRv6 service programming” implementation at Bell Canada and detail the underlying technologies and protocols in use. It will illustrate the factors that influenced its introduction and how the SRv6 network programming architecture proved the ideal approach.
- Daniel Bernier: Introduce using SRv6 Stateless SFC to steer traffic on a path with dynamic inserted services.
- Yao Liu: What is the actual behavior of END.SC?
- Daniel: We created a custom function to map multiple network functions under a single SID rather than requiring multiple SIDs or binding SID needing recirculations, allowing them to scale efficiently by using one SID per SFC, regardless of length.
- Boris Khasanov: How does the MEC Leaf create an explicit-path SR-Policy?
- Daniel: Since we are control-plane independent, we can use any option: BGP, Netconf, PCEP, gRPC, etc. Currently using NetConf+Route coloring.
- James Guichard: Are you using specific arguments in the SID to identify where you are in the chain?
- Daniel: Nope, we use a custom implementation that uses specific SID per SFC. But, arguments can be added in the future.
- Cheng Li: END.SC as a new behavior has been defined and not defined in the IANA. What is the difference between this SID and the existing SID behavior END.NSH from RFC9491?
- Daniel SID-to-NSH mapping in SRv6 is a local behavior, meaning the device internally manages the SID instructions (like END.SC) without external dependencies, reflecting the network programming logic of localized operations on programmable devices.
- Cheng: Add this description to the draft.
- Dhruv Dhody: The registry is FCFS, should it be registered?
- Ketan Talaulikar: There is also a "private use" registry and don't request standardization if is used in the private network. Provide feedback on the IDR I-D?
- Daniel: Will provide it on the mailing list. While exposing BGP-LS is ideal in theory, it’s impractical for security-focused functions (like a DDoS engine) that don’t typically support such protocols and are designed to avoid external exposure, making protocol adherence challenging for these devices.
2.2 SRv6 in Smart Grid (Jiangang Lu, China Southern Power Grid)
This talk will share insights from deploying SRv6 technology within China Southern Power Grid (CSG). It will cover key SRv6 features that were implemented to meet specific service requirements, highlighting the operational experiences and benefits achieved through the deployment.
- Boris: About 1st use case, you mentioned that some old access segments still use a MPLS, does it mean that your PEs support SRv6-MPLS Interworking (draft-agrawal-spring-srv6-mpls-interworking)?
- Jiangang: (From Chat) Actually in the existing stage, we take use the fact that all dispatching services operate between substations and the dispatching center. So just upgrade the dispatching backbone network region by region. Still thanks for reminding the document, we maybe consider this in other scenarios
2.3 SRv6 Network Analytics Network Incident Postmortem (Thomas Graf, Swisscom)
The presentation will focus on an SRv6 network incident at Swisscom. Thomas will demonstrate how specific SRv6 operational network metrics (RFC 9487, RFC 9069) contributed to identifying Network Anomaly Detection, based on the architecture outlined in the draft-ietf-nmop-network-anomaly-architecture.
- Dhruv: Please discuss on the ML. This is an interesting case, this could be the foundation for network observability work.
2.4 SRv6 DC Multi-POD scenario (Gyan Mishra, Verizon)
In this presentation, Gyan will explore Verizon’s IT Multi-POD data center architecture, which leverages SRv6 Next-C-SID and organizes each POD as an independent Autonomous System (AS). The discussion will focus on deploying end-to-end SRv6 Next-C-SID across North-South and East-West inter-domain routing paths.
- Jeff Tantsura: I don't see any DC has requirement of TE because of bandwidth and CLOS topology. The feedback loop is really short? Why bother with DC?
- Gyan: A major advantage of using uSID SRv6 in data centers is the ability to extend the SRv6 fabric to the host level, enabling traffic engineering and steering from the host’s perspective. This setup addresses issues like head-of-line blocking, which can be managed with virtual output queues and advanced queuing techniques, even on hardware-limited platforms with high bandwidth.
- Jeff: In high-bandwidth environments with extensive topology, the feedback loop in a centralized SDN solution may be too slow; by the time topology changes are reconciled and new policies are deployed, congestion might have already resolved, making centralized control overly complex and ineffective for real-time congestion management.
- Dhruv: please continue to discuss on the list. Clarify if this is an operational guidance or explanation of a deployment scenario.
- Chairs: If you have any feedback on the SRv6OPS session and how we can improve, please reach out.