IETF Last Call Review of draft-ietf-teas-actn-poi-applicability-20
review-ietf-teas-actn-poi-applicability-20-opsdir-lc-buraglio-2026-09-19-00
review-ietf-teas-actn-poi-applicability-20-opsdir-lc-buraglio-2026-09-19-00
Hi, I have been selected as the Operational Directorate (opsdir) reviewer for this Internet-Draft. The Operational Directorate reviews all operational and management-related Internet-Drafts to ensure alignment with operational best practices and that adequate operational considerations are covered. A complete set of _"Guidelines for Considering Operations and Management in IETF Specifications"_ can be found at https://datatracker.ietf.org/doc/draft-ietf-opsawg-rfc5706bis/. While these comments are primarily for the Operations and Management Area Directors (Ops ADs), the authors should consider them alongside other feedback received. - Document: draft-ietf-teas-actn-poi-applicability-20 - Reviewer: Nick Buraglio - Review Date: 2026-09-14 - Intended Status: Informational --- ## Summary - Has Issues: I have some minor concerns about this document that I think should be resolved before publication. ## General Operational Comments Alignment with RFC 5706bis Provide an overview of the draft’s operational feasibility, readability, and alignment with RFC5706bis guidelines. The document describes how the ACTN architecture (RFC 8453) applies to Packet Optical Integration (POI) deployments, specifically the coordination between IP/MPLS Packet PNCs (P-PNCs) and Optical PNCs (O-PNCs) under a Multi-Domain Service Coordinator (MDSC). The document analyzes topology discovery, service setup, and multi-layer path coordination scenarios using YANG data models at the MPI interface. This review was done on version 20. It is very detailed and clearly authored, fairly easy to read with a well-structured gap analysis in Section 6. From an operational / engineering standpoint, however, the document has some notable deficiencies. Section 8 (Operational Considerations) is thin relative to the complexity of the architecture it describes, several operationally significant failure modes are deferred or left unaddressed, and the security section relies on fairly generic statements without guidance commensurate with the sensitivity of cross-layer controller infrastructure. The concerns below are raised in the spirit of making this document more useful to operators who will actually attempt to deploy ACTN-based POI architectures, as it is fairly complex in nature and would likely need significant operational resources for design, ops, as especially troubleshooting in real-time. Having done some of this work, the challenges are many and very complex. There are also some significant holes in this document regarding the realities of operating such networks that are left as out of scope, but nevertheless are very common. I almost feel like this is multiple documents pushed together into one monolithic document that should be something like 3-4 discreet drafts (operational guidance, security considerations, architectural overview). As it stands, with an operators hat on I would be hard pressed to find this document useful from an *operations* perspective. ## Major Issues ### Section 8 (Operational Considerations) is inadequate for the architecture's complexity Section 8 lists high-level categories (capacity planning, interoperability, performance monitoring, fault management) but provides no actionable guidance for any of them. For an architecture that coordinates multi-layer provisioning across packet and optical domains from multiple vendors, the following important operational realities are missing or left wanting: **State synchronization and consistency.** The MDSC maintains a topology and service database derived from PNC notifications (RFC 7923, RFC 8641, RFC 8650). The document does not discuss what happens when the MDSC state diverges from PNC state—due to PNC restart, notification loss, or partial failure. In production environments, controller state divergence is a well-known operational hazard in multi-vendor deployments. To start, I would expect some guidance on the following, even it it is simply "consider this as a probblem-space when implementing" either as an operator or a developer of a controller: - How the MDSC detects and recovers from state inconsistency. - What happens to in-flight provisioning operations during a PNC restart or MDSC failover? the operational realities of this shouldn't be under-stated. This is a very common issue with such services, and one that should be addressed. **MDSC high availability.** The MDSC is the single coordination point for multi-layer provisioning. Its failure or restart has direct impact on the ability to provision, modify, or restore services. The document does not address MDSC redundancy, failover behavior, or the operational impact of MDSC unavailability on already-provisioned services. This is a fundamental deployment concern. I don't know of a large network that would deploy a PCE without redundancy, and I suspect this is a significant gap here as well. **Brownfield migration.** Section 3.3 briefly mentions the PCEP brownfield case, but the document otherwise assumes a greenfield ACTN deployment. Most service provider networks that would benefit from POI are brownfield environments with long existing optical management systems and IP EMS/NMS. There is no discussion of how operators should migrate from siloed domain management to ACTN-based coordination, what the incremental steps are, or more practically, what co-existence constraints apply. At minimum, the document should acknowledge this and note whether the architecture is compatible with staged adoption. **Operational skill gaps.** A key motivation stated in Section 1 is that packet and optical operations are siloed due to different skill sets, I've seen this personally. The proposed ACTN architecture does not eliminate this problem—it shifts it to the MDSC operator who must now understand the behavior of both layers, an issue that in reality may be very challenging to overcome. The document should acknowledge that deploying ACTN POI requires new multi-layer operational competencies and may require organizational changes, rather than implying automation alone resolves the skill gap, i.e. not a technical problem, but a potential organizational shift. This is pretty difficult to articulate in an informational doc, but it is very much an operational/organizational reality. **Timing and latency of provisioning workflows.** The multi-step provisioning workflows in Sections 4 and 5 involve sequential interactions between the MDSC, P-PNCs, and O-PNCs. For bandwidth-on-demand or maintenance-driven rerouting scenarios, latency in the control loop definitely matters. The document provides no guidance on expected or acceptable provisioning times, nor on timeout and retry behavior at the MPI. Operators need to know whether these workflows are suitable for near-real-time bandwidth adjustment or only for pre-planned changes. I've also seen this in PCE deployments. Finding the latency requirements for an HA system to properly syncronize is a hidden requirement in many cases, and latency of provisioning goes hand in glove. Not exactly the same, but adjacent and informative in practice. ### Scalability analysis is explicitly deferred with no resolution path Section 4 states: _"Reporting the entire inventory and detailed topology information of packet and optical networks to the MDSC may present scalability issues... The analysis of the scalability of this approach and mechanisms to address potential issues is outside the scope of this document."_ This is acceptable for a first analysis, but Version 20 of the document should either: - Provide at least a qualitative characterization of the scalability constraints (e.g., order-of-magnitude topology sizes where concerns arise), or - Reference specific work in progress that addresses this, or - Explicitly identify this as a gap requiring further work. Simply deferring scalability without a forward reference or gap identification is insufficient at this stage of the draft's lifecycle. Operators deploying at scale—which is the explicit target audience—need at least a framework for assessing whether a given deployment is within tractable bounds. There are some adjacent references in some commercial optical planning systems, and it may be useful to reference those in study. There are also limits within many optical shelves that require zone-based deployments for topology isolation (i.e. OSPF-TE areas to reduce the overhead for the shelves to process LSAs). This may also inform some data points in that regard. Detailing out scaling is hard, but operators need some guidance, even if it is simply "please verify with vendors". The document should either: - Reference the specific work that will close this gap and indicate its current status, or - Expand the gap statement to clarify what a complete SR-TE augmentation must cover (binding SID allocation, SID list configuration, path reporting, revertive behavior), so that implementers and WG contributors understand the scope of missing work. The note in Section 5.3 about binding SID persistence across router or PNC reboots is operationally important and deserves explicit treatment rather than a parenthetical remark. Loss of binding SID mappings after a reboot would break inter-domain SR-TE stitching. Operators need to understand whether this is a device requirement, a PNC requirement, or both. ### Failure recovery workflows are described inadequately or relegated to appendix Appendix A.2 describes maintenance window and router port failure scenarios, but the descriptions are informal and incomplete. The router port failure scenario (A.2.2) explicitly states _"the scenario is not clear"_ in the non-LAG case. A document at version 20 should not contain an acknowledged unresolved scenario in a normative appendix without at least characterizing why it is unresolved and what the open design questions are. More broadly, the document's failure model needs more work: - Multi-layer fault isolation (i.e. defining which layer caused the alarm?) is explicitly deferred to draft-ietf-teas-actn-poi-assurance, which is itself still undergoing changes / scoping. - The cascading failure scenario—where an optical failure causes IP link failure, triggering FRR, with the MDSC subsequently needing to restore the optical path—is described only at a high level with no discussion of timing interactions between the optical restoration, FRR convergence, and MDSC-driven re-optimization. - There is no discussion of partial failure states: what happens if the MDSC successfully reconfigures the P-PNC but the O-PNC operation fails, or vice versa? Given that multi-layer resiliency is one of the primary motivations for POI (explicitly stated in Section 1), the treatment of failure recovery is surprisingly thin in the main body of the document. --- ## Minor Issues Operational considerations that should probably be in Section 8. ### Coherent pluggables (ZR/ZR+) exclusion deserves more explicit treatment Section 2 excludes scenarios with coherent optical interfaces on IP routers ("Scenarios using coherent optical interfaces on the IP routers are outside the scope of this document"). This exclusion is noted without explanation. Coherent pluggable optics (400ZR, OpenZR+, OpenROADM-compliant ZR) are a repidly growing deployment model where the boundary between packet and optical domains changes fundamentally—the router itself becomes a DWDM endpoint, collapsing the cross-technology Ethernet link concept. Given the operational relevance of this deployment model and the rate that it has beenn increasing in the wild, the document should [briefly] address why this scenario is out of scope, and ideally note whether this is expected to be addressed in future work (which is very likely worth the time, IMO). ### Notification subscription management is not discussed The document relies heavily on RFC 7923 / RFC 8641 / RFC 8650 subscriptions for the MDSC to receive real-time topology and service change notifications from PNCs. However, there is no discussion of: - How the MDSC manages subscription lifecycle across PNC restarts - What happens if notification delivery fails or lags behind network state - Whether periodic polling is expected as a fallback - Subscription filter granularity (subscribing to everything from a large PNC may be impractical at scale) ### A gap exists in specificity appropriate to controller infrastructure Section 7 correctly notes that TLS should be used for RESTCONF, NETCONF, and PCEP, and that NACM/RBAC provide access control. However, it may be useful to cite current bests practices rather than simply acknoledge the need for them. - No minimum TLS version is specified. Given that the MDSC controls both packet and optical path setup across an entire operator network, it represents a *very* high-value target. At minimum, the document should recommend TLS 1.2 or later (per BCP 195/RFC 9325) and discourage older versions. - The MDSC compromise scenario is not discussed. If an attacker gains control of the MDSC or a PNC, they can reroute all traffic in the network. The document should acknowledge the privilege level represented by the MDSC and note that operators should treat MDSC access with the same care as they would a device with routing protocol adjacency to all routers. See draft-ietf-spring-srv6-security, which may be of use for reference. - Mutual authentication at the MPI is mentioned implicitly (TLS) but not required explicitly. The document should state whether client certificate authentication is expected, or whether password-based authentication is acceptable. - Audit logging of MDSC and PNC configuration operations is not mentioned. For a system that can provision and modify paths across both layers, audit trails are an operational and often a regulatory requirement. - The supply chain and trust model for multi-vendor PNCs is not discussed. When an O-PNC from one vendor interoperates with a P-PNC from another, the MPI is a trust boundary. Again, migth be too specific but definitely worth considering. ## Nits - **Section 1**: Readability: I don't really see the need to point out that 5g and fiber networks are high bandwidth, this may read easier if "such as ultra-fast mobile broadband for 5G and fiber connectivity services. " was left out. It feels like an unnecessary distinction that will age out as newer technologies emerge. - **Section 2**: The reference to [RFC7424] for LAG/ECMP seems out of place; RFC 7424 is about LAG component link utilization optimization, not LAG configuration in general. IEEE 802.1AX is already cited for LAG; the RFC 7424 reference should be reviewed for accuracy. Both Figure 2 and Figure 3 are fairly busy and hard to follow. I picked up your intention, but they're reasonably hard to digest. I don't really have a suggestion as to what I would change, but even something as simple as expanding them somehow to make them less congested might help? - **Section 5.3**: "The MDSC also request" should be "The MDSC also requests." - **Appendix A.2.2**: "the scenario is not clear" is an unusual phrase for a standards-track-adjacent document. This should be reformulated as a gap statement with a specific description of what is unclear and what additional specification would be needed to resolve it. - **Section 6**: The gap summary notes that "TE Tunnel model... needs enhancement to support scenarios where multiple parallel TE paths are used in load-balancing." This is described as "noted" but "not analyzed in this document." Given that multiple parallel TE paths for load balancing is a standard deployment practice in SP networks, this gap deserves at least a gap identifier rather than a footnote-style observation.