| Internet-Draft | SRv6，6G，UPF | July 2026 |
| Cheng & Zhang | Expires 20 January 2027 | [Page] |
This document leverages the five standardized 6G service scenarios defined in 3GPP TR 22.870—multi-agent collaborative communication, compute-network integration, integrated sensing and communication (ISAC), network digital twin (NDT), and wide-area deterministic networking—as core business baselines. It systematically analyzes the multi-dimensional stringent user-plane requirements introduced by next-generation mobile systems. This work further dissects inherent technical limitations within the 5G GTP-U point-to-point tunnel architecture that render it incapable of accommodating dynamic 6G service characteristics. Based on the above analysis, this draft illustrates the inherent programmability benefits provided by SRv6, and accordingly establishes a suite of standardization requirements for SRv6 to serve as the underlay forwarding protocol of the 6G user plane.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 20 January 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
The 5G user plane relies throughout the network on GTP-U point-to-point tunnels to carry N3/N9 interface traffic, with tunnel endpoints statically bound during PDU session establishment. This architecture has four inherent limitations that prevent it from matching the dynamic, multi-dimensional service requirements of 6G:¶
Forwarding path rigidity: The forwarding path is determined at session establishment and cannot be dynamically adjusted based on real-time link state, compute load, or service requirements, making routing detours highly likely and significantly increasing service latency. Moreover, GTP tunnels are anchored between fixed N3/N9 endpoint pairs; there is no native mechanism to dynamically insert or remove intermediate service nodes (such as compute offload points or AI inference accelerators) without tearing down and re-establishing the complete PDU session, introducing excessive signaling overhead in highly dynamic scenarios.¶
High path-reconfiguration latency: The current 5G user plane adopts a serialized, segment-by-segment establishment approach, with a relatively lengthy procedure. Path reconfiguration caused by user mobility, service requirement changes, or network element/link failures results in latency that is difficult to meet for delay-sensitive services.¶
Single-dimensional service identification: 6G scenarios impose stringent requirements on network latency, bandwidth, reliability, and emerging metrics such as computational capacity. Network conditions and service requirements change dynamically in real time. The current approach, which relies primarily on indirect descriptions based on parameters such as GBR and QCI, cannot directly, comprehensively, or precisely describe the differentiated requirements of diverse scenarios.¶
Coarse forwarding granularity: User-plane path construction relies on parameters such as DNN and S-NSSAI at the PDU session level, which prevents fine-grained differentiated forwarding for individual data flows or specific types of sensing packets.¶
The 6G era introduces new services such as dynamic agent collaboration, distributed AI inference, massive ISAC data ingestion, full-network digital twinning, and industrial control. These scenarios demand a forwarding plane that is inherently flexible, maintains minimal state at intermediate nodes, and is capable of binding transport behaviors to application-layer semantics. SRv6 meets these demands through its three programmable capabilities:¶
Path Programmability: An ordered list of SIDs (Segment Identifiers) in the SRH (Segment Routing Header) completely specifies the end-to-end forwarding path. It supports dynamic insertion or removal of transit nodes, primary/backup path separation, and Fast Reroute without control-plane intervention.¶
Service Programmability: Through IPv6/SRv6 extension headers, it can carry QoS, slicing, bandwidth, and latency constraints to achieve flow-granularity differentiated scheduling.¶
Application Programmability: Through standardized fields carried in SRv6 packets for application-specific identifiers (such as AI compute load and application information), the forwarding plane is enabled to sense and respond to application-layer semantics.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
Distributed AI agents (such as autonomous fleet controllers, federated learning worker nodes, and multi-robot coordination systems) exchange high-frequency control messages and model gradient data across geographically dispersed UPFs. These interactions require low-jitter, highly reliable, and non-detoured mesh connectivity. Agent members join and leave at sub-second timescales based on task assignment and mobility, supporting real-time cross-regional interaction and collaborative task scheduling among multiple agents.¶
Key Service Requirements:¶
Sub-second dynamic agent on-boarding and off-boarding, with no interruption to ongoing PDU sessions during path changes.¶
Dynamic insertion and deletion of multi-level intermediate processing nodes (preprocessing, inference, aggregation) in service chains.¶
Differentiation between small control packets and large model traffic flows, with independent queues and differentiated latency guarantees for the two types of traffic.¶
Support for mesh connectivity between any nodes across the entire network, avoiding latency increases caused by cross-region routing detours.¶
Load balancing of massive concurrent service flows across equal-cost links, with in-order packet delivery guaranteed for the same conversation flow.¶
Addressing the deployment of AI models at edge UPFs and multi-level scheduling of wide-area KV Cache, this scenario supports high-speed lossless migration of KV Cache data across edge UPFs and between central cache nodes, ensuring continuous and imperceptible large-model inference services during user mobility.¶
Key Service Requirements:¶
Support for arbitrary mesh topology to enable high-speed lossless migration of cache data across edge UPFs.¶
Path planning must integrate multi-dimensional metrics including compute load and storage utilization for joint scheduling.¶
Service interruption duration during compute instance switching must be controlled within 50ms.¶
Stateful AI inference services must maintain the binding relationship between data flows and local cache.Frequent switching of compute nodes would incur cold-start penalties and significantly increase inference latency.¶
6G networks simultaneously possess both communication transmission and environmental sensing capabilities, achieving deep integration of communication and sensing to support scenarios such as vehicle-infrastructure cooperation and security surveillance. Base stations simultaneously complete communication transmission and environmental sensing, with the user plane needing to carry Gbps-level point-cloud data flows and millions of concurrent micro-sensing data packets (such as radar point clouds, environmental fingerprints, LiDAR returns). Packets are typically 100-300 bytes in size, with millions of flows from densely deployed sensor arrays converging simultaneously on the edge network. Massive sensing data and high-value communication data are transmitted together, requiring that important sensing flows receive deterministic quality assurance.¶
Key Service Requirements:¶
Massive real-time sensing small packets of hundreds of bytes require reduced encapsulation transmission overhead and improved bandwidth utilization.¶
Massive concurrent sensing flows require route aggregation to reduce forwarding hardware FIB entry pressure.¶
Resource isolation between high-priority radar/security sensing flows and ordinary background sensing flows to avoid congestion preemption.¶
Real-time collection of end-to-end latency and packet loss metrics for each sensing packet, with dynamic forwarding path adjustment based on measurement results.¶
Low-frequency raw point-cloud/radio-frequency large packets require support for lossless fragmentation and transmission of oversized packets.¶
Construct a digital mirror that maps the physical network in real time, accurately restoring network state and maintaining a real-time digital replica of the entire network, including all forwarding nodes, links, and queues. All network nodes continuously report latency, jitter, packet loss, queue depth, and utilization metrics to the twin platform, supporting network simulation, fault prediction, and closed-loop optimization.¶
Key Service Requirements:¶
Telemetry measurement traffic must be completely isolated from user production traffic, with no preemption of service bandwidth.¶
Collected data must cover all forwarding hops, not just endpoints.¶
Real measured values for per-hop latency, queue depth, and link load must be obtainable, without reliance on estimated data.¶
Support for on-demand adjustment of telemetry sampling frequency and collection scope to balance measurement overhead and data accuracy.¶
Applications such as remote automated factory control, smart grid protection relay signaling, autonomous vehicle platooning, and remote surgery require stringent bounded latency (e.g., end-to-end < 10ms), jitter < 1ms, packet loss rate < 10^-6, and sub-50ms automatic failure recovery. Industrial control flows must pass through security service nodes in sequence, such as "firewall → intrusion detection → controller"; any detour or out-of-order delivery could cause control logic errors.¶
Key Service Requirements:¶
End-to-end latency, jitter, and packet loss rate must have deterministic upper-bound constraints; optimizing only average metrics is insufficient.¶
Primary and backup paths must be physically link/node-disjoint to avoid shared failure domains.¶
Automatic link failure switching with service interruption duration below 50ms.¶
Industrial control flows must pass through security service nodes such as firewall and intrusion detection in a fixed order; routing detours that disrupt the processing sequence must be prevented.¶
Dedicated end-to-end transmission resources must be reserved for critical services to ensure highly reliable isolated transmission.¶
Synthesizing the above five categories of 6G key service scenarios, six categories of user-plane general capability requirements can be abstracted as the input baseline for the SRv6 capability analysis in Section 4:¶
Dynamic Topology and Seamless Path Scheduling Support for dynamic addition/removal of nodes/service chains, mesh connectivity, fast failure self-healing, and long-duration interruption-free mobility/compute switching.¶
Fine-Grained Differentiated QoS and Traffic Isolation Differentiation between different types of data flows, allocation of independent forwarding resources, and guaranteed prioritized transmission for critical services.¶
Deterministic Reliable Forwarding and Physical Redundancy Physically disjoint primary/backup paths, sub-50ms failure switching, ordered service chains, and dedicated resource reservation. Primarily targeting wide-area deterministic communication, while also addressing vehicle-infrastructure sensing and AI high-reliability interactive services.¶
Application Semantics In-Band Carriage The forwarding plane must be able to recognize upper-layer service-specific identifiers (cache affinity, compute requirements, sensing types, etc.) and adjust forwarding policies based on application semantics.¶
Efficient Data Transmission and Large-Scale Flow Aggregation Reduced header encapsulation overhead for small packets, and convergence of forwarding table entries for massive concurrent flows. Primarily adapting to ISAC scenarios while also optimizing high-frequency control small packets for AI agents.¶
Full-Path Native Operations and Measurement Per-hop real-time collection of network performance metrics, traffic isolation, flexible sampling, supporting full-network digital twin simulation and fault localization.¶
This section, based on the six general 6G user-plane capability dimensions summarized in Section 3, provides a unified summary of SRv6's existing IETF-standardized native capabilities, and distills the new standardization extension directions required for 6G services.¶
The existing SRv6 standards family (RFC8402, RFC8754, RFC8986, RFC9800, etc.) has already formed a complete underlying forwarding capability foundation that can cover the basic forwarding requirements of 6G services:¶
Path Scheduling: Source routing via SR Policy enables explicit definition of complete end-to-end forwarding paths; combined with TI-LFA and BFD, it provides sub-50ms failure self-healing without control-plane intervention; the Make-Before-Break (MBB) mechanism ensures service continuity during path switching.¶
Service Chaining: SRv6 SFC supports sequential forwarding of service flows through ordered service nodes.¶
Differentiated Scheduling: SRH extension TLVs and IPv6 extension headers can carry basic QoS priority and latency constraints, with differentiated SIDs enabling basic traffic queue isolation.¶
Overhead Optimization: The compressed SID scheme defined in RFC9800 reduces small-packet header overhead.¶
Application Carriage: SID Arguments and SRH TLVs reserve standardized extension fields capable of carrying various upper-layer service identifiers.¶
In-Band Measurement: IOAM measurement fields can be embedded in packets to collect per-hop performance metrics such as latency and queue load.¶
Deterministic Assurance: SR Policy supports binding bandwidth reservation parameters, providing basic resource assurance for deterministic services.¶
Mobility Interworking: RFC9433 defines End.M.GTP6.E/End.M.GTP6.D to complete bidirectional encapsulation conversion between GTP and SRv6.¶
Building upon the existing standards foundation, to fully satisfy the integrated service constraints of 6G scenarios such as dynamic AI agents, compute-network integration, massive ISAC traffic, wide-area industrial control, and full-network digital twins, a unified and comprehensive set of SRv6 extension standardization specifications needs to be perfected. The following dimensions can be considered:¶
Path Scheduling: A local SID segment update mechanism is needed to enable dynamic forwarding path adjustment without requiring complete PDU session re-establishment.¶
QoS and Resource Isolation: Unified binding rules between network slicing and SR SIDs are needed, with dedicated endpoint forwarding behaviors defined for industrial control, sensing, and AI inference services to achieve hard resource isolation.¶
Application Semantics Carriage: Unified encoding specifications are needed for service labels such as cache affinity, compute load, and agent priority in SID Arguments or SRH TLVs, along with a control-plane collaborative process that reads application semantics and automatically selects optimal forwarding paths.¶
Deterministic Service Assurance: Support is needed for hard-constraint assurance mechanisms for end-to-end latency, jitter, and packet loss metrics.¶
SRv6 path computation, SID allocation management, and policy mapping rely on deep coordination between the core network control plane (e.g., SMF, AMF) and the transport network control plane (e.g., SDN controller, BGP route reflectors).¶
End-to-End Centralized Path Computation and Policy Download: The control plane must be capable of collecting comprehensive network state information—including UPF node status, link TE metrics, compute load indicators, and available slice resources—and performing end-to-end SRv6 path computation. The computed complete SID list, service scheduling rules (FAR/PDR, etc.), and application extension parameters (ARG/TLV) must be downloaded to the ingress UPF in a single transaction using standard protocols (such as PCEP or BGP SR-Policy). This ensures transactional consistency between path selection and policy installation.¶
Dynamic Path Adjustment: During user mobility or compute node migration, the control plane should support updating only the affected portion of the SR path (e.g., replacing the middle SID representing the serving UPF) without tearing down the entire SR policy or interrupting online data flows. The update operation should adopt a Make-Before-Break mechanism: the new SID list is installed in hardware at the ingress UPF while the old list remains active, with traffic switched only after the new path is verified (e.g., via BFD).¶
SLA-to-SID Mapping: The control plane must map the 3GPP-defined 5QI/QFI parameters to SRv6 TLV fields (latency, jitter, packet loss rate, priority) or IPv6 extension headers. This mapping should be pre-configured in the policy table at the ingress UPF to enable rapid translation for each new flow without controller interaction. When network SLA degrades (e.g., measured latency exceeds the target), the control plane should support automatic path re-optimization triggered by telemetry data, migrating affected flows to backup paths.¶
State Synchronization Between SMF and Transport Controller: The Session Management Function (SMF) and the transport network controller must synchronize state regarding active sessions and their associated SID lists.¶
The introduction of source routing (SRv6) in the 6G user plane raises specific security concerns that must be addressed. These are categorized below, with clear traceability to the requirements defined in Section 4.¶
Since the path is encoded in the packet header, external attackers could inject a forged SRH and malicious SID list to redirect traffic to a rogue node or create loops in the network that consume resources.¶
Every SRv6-capable UPF must validate that the packet source IPv6 address belongs to the allowed administrative domain (e.g., using ingress ACLs that restrict source addresses to the operator's prefix range). Additionally, the SRH should be protected by a cryptographic signature or HMAC in the IPv6 extension header, with the signature covering the SID list and critical TLV fields to prevent tampering.¶
Modifying QoS, compute load, or application-semantics ARG could allow attackers to elevate service priority, misdirect traffic to congested nodes, or degrade service quality. The control-plane protocol used to download SID lists and TLVs must be secured with TLS, SSH, or similar security protocols, with mutual authentication between the controller and the UPF. This ensures that only authenticated controllers can issue SID updates. At the forwarding plane, where feasible, nodes should implement integrity verification of ARG and critical TLV fields—either through per-hop verification (e.g., using a shared secret to validate TLV checksums) or through cryptographic protection of the entire SRH.¶
The SID list may inadvertently expose the precise location of a user-plane endpoint, especially when the last segment is a UPF identifier associated with a specific cell site or geographic region.¶
Edge aggregation nodes should strip or obfuscate terminal-specific SIDs before the packet leaves the trusted administrative domain for an external Data Network (DN). This should be achieved by replacing the terminal-specific SID with a generic egress SID that does not reveal location information, or by using privacy-preserving SIDs that are periodically rotated. For flows that remain within the operator's domain, location privacy may be protected through the use of anonymized SIDs.¶
The control plane (controller, PCE, SMF and other control plane entities) is a critical component. An attacker compromising this component could manipulate all forwarding policies. All control-plane interactions—between PCE/SDN controller and UPFs, and between SMF and transport controller—must use secure transport protocols (e.g., TLS 1.3 or higher) with mutual authentication. Role-Based Access Control (RBAC) should be enforced to restrict which entities can modify SRv6 policies. Access logs should be maintained for all policy changes to support forensic analysis in the event of security incidents.¶
This document makes no requests for new IANA registrations.¶