]> Framework for High Performance Wide Area Network (HP-WAN) ZTE Corporation
xiong.quan@zte.com.cn
ZTE Corporation
huang.guangping@zte.com.cn
China Mobile
yaokehan@chinamobile.com
New H3C Technologies
linchangwang.04414@h3c.com
hpwan This document defines a framework to enable the host-network collaboration for high-speed and high-throughput data transmission, coupled with fast completion time and newtork utilization of High Performance Wide Area Networks (HP-WAN). It particularly enhances the congestion control by introducing signaling mechanisms and collaborative functions, including rate negotiation, traffic scheduling, resource reservation, admission control and rate control to meet job-level objectives.
Introduction Data-intensive applications always demand high-speed data transmission over Wide Area Networks (WANs) such as scientific research, academia, education as discussed in and other applications in public networks as per . These applications which mainly focus on job-bases massive data transmission over long-distance WANs impose stringent requirements on throughput, completion time, and reliability. Traditional transport protocols often exhibit limitations in high-speed WAN environments, including slow convergence, long feedback loops, and inefficient handling of incast congestion as discussed in . Consequently, uncoordinated multi-flow competition leads to rate instability and suboptimal resource utilization. The High Performance WAN (HP-WAN) framework addresses these challenges by enabling the host-network collaboration. High, reliable and effective data throughput coupled with completion time and newtork utilization is the fundamental requirement for HP-WAN. The high-level requirements are: *Job-Aware Scheduling: Schedule multiple data transfer requests based on available capacity and required completion times. *Collaborative Signaling: Allow hosts to signal traffic characteristics to the network for proactive congestion avoidance. *Transport Reliability: Ensure reliable delivery through traffic scheduling and admission control to minimize packet loss and delay. *Rate-Aware Routing: Compute optimal paths, reserve resources according to negotiated QoS policies and control the rates in pre-congestion conditions. This document defines a framework to enable the host-network collaboration for high-speed and high-throughput data transmission, coupled with fast completion time and newtork utilization of High Performance Wide Area Networks (HP-WAN). It particularly enhances the congestion control by introducing signaling mechanisms and collaborative functions, including rate negotiation, traffic scheduling, resource reservation, admission control and rate control to meet job-level objectives.
Conventions used in this document
Requirements Language 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 when, and only when, they appear in all capitals, as shown here.
Definition of Terms This document uses the terms defined in and :
Framework for HP-WAN
Overview The framework is formulated to enable the host-network collaboration upon more active network involvement. The client and server could adjust the rate efficiently and rapidly with the negotiated rate-based congestion control in a fine-grained way. The network could enhance the capability to regulate the traffic and schedule the resources which could provide predictable network behaviour and mitigate incast network congestion preemptively. The following diagram illustrates the functionalities between Client/Server and WAN including: *Host-network collaboration signalling or configuration *Active network-collaborated traffic enforcement and scheduling *Negotiated rate-based congestion control algorithms |Edge Node|...|Transit Nodes|...|Edge Node|<------->| Server | | | +----+----+ +-------------+ +----+----+ | | +--------+ | | +--------+ *collaboration | | *collaboration signalling/configuration| | signalling/configuration +-------------------------------+ \_________/ \______________________________/ *Negotiated rate-based *Active network-collaborated congestion control enforcement and scheduling algorithms Figure 1 HP-WAN framework ]]>
Signalling in Distributed Model The following diagram illustrates the workflows among client, server and network nodes (e.g. edge nodes and transit nodes).
The RSVP signaling can be instantiated as per . *The request of scheduled traffic will be signaled from the client to the network based on the negotiated rate. Furthermore, the traffic pattern and job-based requirements, such as completion time, should be included in the request. *The edge node will perform admission control and acknowledge the traffic, reserving the resource quota, but it will reject access when the network capacity cannot guarantee the job's completion time. *The acknowledgement will be signaled back from the network to the client, including the response with the negotiated rate and QoS policy for the client to send traffic. *The notification will be signalled from the client to the network to notify the completion of traffic, and the network will release the resource quota and cancel the acknowledgement of this job. *The update may signal to the client from the network to update the acknowledgement of the negotiated rate when new traffic requests are received. |*Rate negotiation | | Requests | | |*Traffic scheduling| |----------------->| |Acknowledgement |*Admission control | | Acknowledgement | |(negotiated rate) | *Reserve resource quota |<-----------------| |<------------------------|*Negotiated rate-based traffic engineering| | | |<########################################>| | |New Request | | | | |------------------------>| | | | |Update(negotiated rate) | | | | |<------------------------| | | | |Notification(completion) | | | Notification | |------------------------>| *Release resource quota |----------------->| |Acknowledgement(cancel) |<########################################>| Acknowledgement | |<------------------------| | |<-----------------| | | | | | V V V V V Figure 2 The workflow of signalling between hosts and network ]]>
Configuration in Centralized Model Host-and-network collaboration could also be performed using configuration in centralized model. The hosts could communicate with the network through YANG data model. The job metadata and service request from the application is as per and . It may be considered as centralized approach where a controller or an orchestrater orchestrates data processing and resource allocation across hosts and network infrastructure. For instance, for the SDN for End-to-end Networked Science at Exascale (SENSE) system in Research and Education (R&E) networks, the orchestrator and resource Manager (RM) have the capability of hierarchical planning and resource reservation in the network. The orchestrator communicates the requests from applications and interacts with the RM for resource reservation. The following diagram illustrates the workflows among orchestrator, controller, client, server and network nodes (e.g. edge nodes and transit nodes). *The request of scheduled traffic will be initialed from the Application to the Orchestrator with traffic pattern and job-based requirements included. *The Orchestrator will perform rate negotiation among hosts and networks. If the network resources is efficient, the Orchestrator will perform admission control and acknowledge the traffic. *The Acknowledgement will be configured to the client, including the response with the negotiated rate and QoS policy for the client to send traffic. *The Controller of the network will reserve resource quota to guarantee the job's completion time. |Edge Node|...|Transit Nodes|...|Edge Node|<-------->| Server | | | | | | +----+----+ +-------------+ +----+----+ | | | | | +--------+ | | | | +--------+ | | | | | | | +-------------+ +-------------------------------+ +---------------+ Figure 3 The workflow of configuration between hosts and network ]]>
Traffic Workflow The client could send traffic according to the negotiated rate policy to achieve a high throughput within the completion time. The edge node could adjust the maximum rate of traffic when the network performs the rate control in pre-congestion situation. And it also could send fast feedback with the advised rate when the traffic rate does not apply to the network. |-------------->|----------------->|----------------------->| | | | | | | | | | | | | *Rate control(maximum-rate) | | |*Fast Feedback(advised-rate)|<################################>| | |<---------------------------| | | | | Traffic(advised-rate) | Traffic(advised-rate) | Traffic(advised-rate) | |--------------------------->|-------------->|----------------->|----------------------->| | | | | | | | | | | | | | | | V V V V V Figure 4 The workflow of traffic between host and network ]]>
Functions The functions are described in the sections below including transport-related technologies such as rate negotiation, admission control, traffic scheduling and enforcement and routing-related technologies like traffic engineering, resource scheduling, rate control and load balancing.
Rate Negotiation In HP-WAN, the host could negotiate the sending rate with the network due to the predictability of jobs. The client communicates the traffic patterns of high-speed flows to the network to negotiate rate. The network responds to the negotiated rate and QoS policy for the client to send traffic. There are three kinds of rate policy as follows:
  • Optimal rate or optimal rate range negotiation. The network provides resource reservation for high-speed data to guarantee the transmission capacity and achieve optimal rate transmission. The client could transmit flows according to the negotiated optimal rate or optimal rate range.
  • Minimum rate negotiation. The network provides the minimum resource guarantee. The client could transmit at a rate not less than the negotiated rate.
  • Maximum rate negotiation. The network provides an upper limit for resource guarantee. The client could transmit at a rate not greater than the negotiated rate.
Admission Control The network node should perform admission and traffic control based on negotiated QoS and rate. By combining the admission control with congestion control, it can provide high throughput associated with completion time while efficiently using the available network capacity. The strategies of admission control are different based on the QoS policy. For example, one strategy is to immediately grant or reject admission to a reservation request on its arrival time, which is called on-demand admission control. If a reservation request can not be granted or rejected at the time of its arrival, it will be put in a queue, which is called queue-based admission control. Furthermore, a time-slot based admission control is used for scheduling the elastic and flows requests.
Traffic Scheduling and Enforcement The network node (e.g.edge node) performs rate-based traffic scheduling and enforcement. For example, traffic classification may be needed based on the traffic type. If it needs to prioritize critical traffic for acceleration, it should upgrade the priority of QoS. Moreover, if the traffic needs a guaranteed QoS, it should provide guaranteed bandwidth for this flow. It also could perform the aggregation of mouse flows or the fragmentation of an elephant flow if needed. Splitting data across multiple paths for load balancing can increase the throughput and provide redundancy. If one path experiences congestion, alternate paths compensate, ensuring timely delivery. The traffic enforcement at network edges can used to regulate data flow to eliminate congestion and minimize the flow completion time. For example, it could enforce the rate limits based on the negotiated rate to access traffic.
Optimization of Congestion Control Algorithms The client should perform the improvement of congestion control algorithms based on the negotiated-rate from the network. The negotiated-rate can be viewed as an initial congestion signal to assist the client in selecting a suitable sending rate with the network resource scheduling acknowledgement. And it also needs to adjust the rate reasonably and rapidly when receiving the fast feedback from the node nearing the client.
Negotiated Rate-based Traffic Engineering The negotiated rate-based traffic engineering should be provided by routing technologies and the signaling from client will assist the network operator's traffic management and corresponding resource planning and scheduling. The edge node may get information (topology, bottleneck link bandwidth, queue and buffer) from a centralized controller or through IGP advertisement. The network should provide resource scheduling at nodes along the path and it is not bandwidth allocation but quota reservation which can be used for admission control as per . The client and network can also negotiate rate based on the quota of each job. Quota is expressed as a vector of resource quantities (bandwidth, buffer, queue, etc.) at a given priority, for a time frame. The network can make dynamic bandwidth reservation upon different time frames defined by quota. It will differ based on the different QoS policy. For example, it is required to reserve the minimum bandwidth quota for the minimum rate policy.
Fast Feedback The fast feedback for HP-WAN enables the edge node to send fast feedback with the advised rate when the traffic rate is not applicable to the network. It could also adjust the sending rate of traffic when congestion occurs and resume it when congestion is mitigated.
Rate Control Networks may proactively enforce rate control as per to mitigate pre-congestion conditions. As defined in , intermediate nodes detecting a early congestion could signal the ingress nodes to reduce the maximum rate. And the ingress nodes could also provide quantitative, fast feedback to clients when the ingress nodes can not handle and need to prevent from overwhelming the network.
Applicability of Host-network Collaboration Signalling There are several existing signalling options for HP-WAN host-network collaboration signalling such as RSVP and GRASP. There will be two deployment scenarios in HP-WAN. The first one will be the central controller deployment which will have a hierarchical planning and resource reservation in the network like CERN deployment and the SENSE architecture. In this case, the host-newtork signalling (between client and edge node) may be peer-to-peer solution and both GRASP and RSVP may be applicable. And the second case will be distributed or hybrid deployment in the network which needs distributed signalling along the path for resource reservation. In this case, the host may signal from the client to the network nodes along the path. RSVP may be applicable but not GRASP. GRASP is peer-to-peer signalling and is designed for synchronization and negotiation between autonomic service agents, which reduces the need for hierarchy and allows the intelligence to be distributed rather than centralized. However it is not applicable when the signalling should be performed along the end-to-end path. Although RSVP may not be deployable with complex configuration and management which requires precise configuration across all network devices along the path. It will also add administrative complexity between host and network in HP-WAN with operational issues. But SR, slicing, diffServ QoS and SDN-based approaches may be used to largely improve RSVP in HP-WAN. Moreover, RSVP reservations often allocate fixed resources in the nodes along the path, which can lead to underutilization if the reserved resources are not fully used. The extensions may be required to applied to HP-WAN that the bandwidth and rate vary over time and it requires scalable throughput, dynamic bandwidth reservation and efficient use of capacity.
Security Considerations It is required to create the trusted relationship between the clients/servers and the network before host-and-network collaboration. The network may perform resource reservation based on authentication (e.g. and ) and authorization (e.g.).
IANA Considerations Currently this document does not make an IANA requests.
References Normative References Key words for use in RFCs to Indicate Requirement Levels In many standards track documents several words are used to signify the requirements in the specification. These words are often capitalized. This document defines these words as they should be interpreted in IETF documents. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements. Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words RFC 2119 specifies common key words that may be used in protocol specifications. This document aims to reduce the ambiguity by clarifying that only UPPERCASE usage of the key words have the defined special meanings. Informative References RSVP Cryptographic Authentication This document describes the format and use of RSVP's INTEGRITY object to provide hop-by-hop integrity and authentication of RSVP messages. [STANDARDS-TRACK] RSVP Cryptographic Authentication -- Updated Message Type Value This memo resolves a duplication in the assignment of RSVP Message Types, by changing the Message Types assigned by RFC 2747 to Challenge and Integrity Response messages. [STANDARDS-TRACK] The OAuth 2.0 Authorization Framework The OAuth 2.0 authorization framework enables a third-party application to obtain limited access to an HTTP service, either on behalf of a resource owner by orchestrating an approval interaction between the resource owner and the HTTP service, or by allowing the third-party application to obtain access on its own behalf. This specification replaces and obsoletes the OAuth 1.0 protocol described in RFC 5849. [STANDARDS-TRACK] Current State of the Art for High Performance Wide Area Networks Lancaster University Jisc Pittsburgh Supercomputing Center ZTE Corporation China Mobile High Performance Wide Area Networks (HP-WANs) represent a critical infrastructure for the modern global Research and Education (R&E) community, facilitating collaboration across national and international boundaries. These networks include global education and research networks, such as GÉANT, Internet2, Janet, ESnet, CANARIE, CERNET, and others, and also refer to large scale commercial dedicated networks built by hyperscalers and operators. They are designed to support the ever-growing transmission of vast amounts of data generated by scientific research, high-performance computing, distributed AI-training and large-scale simulations. This document provides an overview of the terminology and techniques used for existing HP-WANs. It also explores the technological advancements, operational tools, and future directions for HP-WANs, emphasising their role in enabling cutting-edge scientific research, AI training and massive R&E data analysis. High Performance Wide Area Network (HPWAN) Use Cases and Requirements -- From Public Operator's View China Mobile ZTE Corporation Bulk data transfer is a long-lived service over the past twenty years. High Performance Wide Area Networks (HP-WANs) are the backbone of global network infrastructure, enabling the seamless transfer of vast amounts of data and supporting advanced scientific collaborations worldwide. Many of the state-of-the-art dedicated networks have been mentioned in [I-D.kcrh-hpwan-state-of-art]. For non-dedicated networks like public operator's network, the case is different in terms of QoS policies, security policies, etc. This document presents use cases and requirements of HPWAN from public operator's view. Problem Statement for High Performance Wide Area Networks ZTE Corporation China Mobile China Telecom China Unicom CAICT High Performance Wide Area Network (HP-WAN) is designed for many applications such as scientific research, academia, education and other data-intensive applications which demand high-speed data transmission over WANs, and it needs to provide high-throughput transmission within a completion time. This document outlines the problems for HP-WANs. Signaling Solution for HP-WAN ZTE Corporation ZTE Corporation New H3C Technologies This document proposes a technical solution for the host-network collaboration signaling to enhance the congestion control in High- Performance Wide Area Networks (HP-WAN). It also describes the RSVP extensions as an instantiation of this signaling solution. HPC/AI Service Intent Model ZTE Corporation HPE Lancaster University This document defines a common service intent model for High Performance Computing (HPC) and AI workloads over High Performance Wide Area Networks (HP-WANs). The model allows heterogeneous workload managers and orchestration platforms to express endpoint, communication pattern, timing, performance, data movement, policy, and admission requirements for network services without exposing technology-specific tunnel realization details. HPC/AI Scheduler Job Metadata Model ZTE Corporation HPE Lancaster University This document defines a scheduler-facing metadata model for High Performance Computing (HPC) and AI workloads. The model captures common job, workload, scheduler, tenant, timing, and task metadata that can be mapped from heterogeneous workload managers and orchestration platforms and used as context for network service intent. SRv6-based Rate Control China Mobile ZTE Corporation ZTE Corporation This document describes a rate control mechanism for Segment Routing over IPv6 (SRv6) network slices. It addresses the challenge of balancing resource utilization and congestion avoidance in over- committed slice deployments. The mechanism leverages a token-based scheduler to differentiate between Committed Information Rate (CIR) and Peak Information Rate (PIR) traffic, and defines procedures for calculating initial PIR values and dynamically adjusting them based on network conditions. Dynamic rate adjustments are triggered by localized congestion or underutilization, enabling proactive rate control and efficient bandwidth sharing among slices sharing common physical links. SRv6-based Rate Notification ZTE Corporation ZTE Corporation China Mobile This document specifies a rate notification mechanism for Segment Routing over IPv6 (SRv6) networks. It enables a transit or egress node to dynamically notify the ingress (headend) node about a recommended rate range (MinRate and MaxRate) when localized congestion is detected or when underutilized bandwidth is identified, allowing the headend to perform proactive traffic shaping and rate enforcement. This mechanism enhances transmission efficiency in SRv6 networks by enabling fine-grained, congestion-aware rate control. RSVP Extensions for Rate-based Resource Quota ZTE Corporation ZTE Corporation This document proposes RSVP extensions for rate-based resource quota to enable dynamic resource reservation, achieving effective rate control in multi-flow data transmission.