]> Framework for Implementing Lossless Techniques in Wide Area Networks China Unicom
Beijing China het21@chinaunicom.cn
Huawei
Beijing China shihang9@huawei.com
China Unicom
hanzx21@chinaunicom.cn
China Unicom
gaox60@chinaunicom.cn
Huawei
zhoutianran@huawei.com
General rtgwg WAN Lossless Framework This document proposes a comprehensive framework to address the challenges of efficient, reliable, and cost-effective large volume data transmission over Wide Area Networks (WANs). The framework focuses on planning and managing traffic paths, network slicing, and utilizing multi-level network buffers. It introduces dynamic path scheduling and advanced resource allocation techniques to optimize network resouce and minimize congestion. By leveraging cross-device buffer coordination and real-time adjustments, the framework ensures high throughput and low latency, meeting the demands of modern, data-intensive applications while providing a robust solution for large-scale data transmission.
Introduction In recent years, the demand for reliable and efficient transmission of large volumes of data across Wide Area Networks (WANs) has surged. highlighted several critical use cases that emphasize the necessity of low packet loss and high throughput in WANs. These requirements are driven by applications that handle massive datasets, such as scientific research, financial transactions, and multimedia content delivery, while the locations of data production and consumption differ, requiring efficient and timely transmission across WANs. The characteristics and requirements of large data transmission are listed as follows:
  • Large Volume. The datasets involved in these transmissions often reach terabyte levels. Traditional fixed bandwidth dedicated lines, while reliable, can be prohibitively expensive. Enterprises must balance the need for high-capacity data transmission with cost considerations. This necessitates exploring more flexible and economical solutions that can handle large-volume data without incurring excessive costs.
  • Timeliness. Timeliness is a critical factor for data transmission over WANs. For instance, in the field of genetic research, the timely transmission of genetic data can significantly influence diagnostic and treatment outcomes. Delays in data transmission can render the data obsolete, e.g., leading to incorrect results and conclusions. Therefore, ensuring that data is transmitted within a specific time window is essential for maintaining its utility and accuracy.
  • Predictability. Large-volume data transmission tasks typically have predictable patterns, allowing for better planning and resource allocation. This predictability helps in designing network solutions that can efficiently manage the anticipated data load. By leveraging predictable traffic patterns, network administrators can optimize resource allocation, minimize congestion, and enhance overall network performance.
This document proposes a comprehensive framework aimed at addressing the challenges associated with large volume data transmission over WANs. The framework focuses on enhancing traffic management and resource allocation strategies to ensure efficient, reliable, and cost-effective data transmission. By implementing these strategies, the framework aims to meet the demands of modern, data-intensive applications, providing a robust solution for large volume data transmission in WAN environments.
Network Challenges Posed by Large Volume Data Transmission
Limited Network Capacity WANs have finite carrying capacities. When a significant amount of traffic enters the network simultaneously, it can lead to traffic conflicts, resulting in queuing and jitter. These issues are exacerbated by the continuous nature of large data transfers, which can strain network resources over extended periods. Addressing these challenges requires advanced traffic management techniques that can efficiently utilize available network capacity.
Congestion Hotspots Packet loss often occurs due to probabilistic simultaneous influxes of large volumes of traffic. This congestion is exacerbated by mechanisms such as Equal-Cost Multi-Path (ECMP) routing, where multiple flows compete for certain bottleneck links, leading to congestion and packet loss. Packet loss in WANs does not lead to permanent data loss since lost packets can be retransmitted. However, retransmissions increase transmission latency, causing delays in data delivery. Moreover, packet loss can trigger congestion control mechanisms, which reduce the network's throughput to prevent further congestion. This reduction in throughput can significantly affect the performance of data-intensive applications, making it critical to minimize packet loss.
Inefficient Buffer Utilization The network itself has a certain buffer capacity to partially mitigate short-term processing deficiencies. However, current mechanisms only utilize the local device's buffer and do not fully exploit the overall buffer capacity across multiple devices. This fragmented buffer utilization leads to inefficiencies in handling bursty traffic. Advanced congestion management strategies are necessary to coordinate buffer usage across the network, maintaining high throughput and low latency to ensure efficient and reliable data transmission.
Framework This document proposes a comprehensive framework to address the challenges of efficient, reliable, and cost-effective large volume data transmission over Wide Area Networks (WANs). The framework focuses on the planning and management of traffic paths, network slicing, and the use and management of multi-level network buffers.
Adaptive Planning and Management of Network Resouce When users seek efficient transmission of large datasets, they can rent temporary network bandwidth in addition to their fixed leased lines (a.k.a guranteed bandwidth). This temporary bandwidth is cheaper by sharing but offers weaker Service Level Agreements (SLAs). Due to the predictable nature of the traffic, users can pre-request resource scheduling from the network, including traffic paths and even network slices. The network can allocate resources based on availability, avoiding prolonged congestion through effective planning. If serious congestion occurs, the network scheduler can recalculate paths and slice resources. Network devices can flexibly choose the best available path from multiple pre-allocated paths, particularly when head-end devices detect local or remote congestion. By adjusting the current and incoming traffic path selection, network devices can optimize traffic distribution and alleviate congestion dynamically.
Specific Requirements:
  • Network Resource Reporting and User Request: Network devices report attributes such as bandwidth, latency through control plane protocols like IGP and BGP-LS. Users provide the overall needs of bandwidth and latency for large volume data transmission, including guaranteed dedicated resources and flexible resources with weaker guarantees. In addition to know network parameters such as bandwidth and latency, the system also needs to know whether network forwarding nodes have the ability to share the buffer with other devices,and confirm the scope of the wide-area lossless network domain. In the centralized mode, the central controller needs to know the network device's capability of buffer and buffer size to do path planning.The information transferring can be done through the BGP-LS protocol extension. In distributed mode, the network forwarding nodes can realize multi-level network buffering and path switching by knowing the neighbour's capability of buffer and buffer size. The information transferring can be done through the IGP protocol extension.
  • Network Resource Allocation and Policy Distribution: Controllers calculate out IP-based dedicated lines (IP tunnels with segment routing) within the WAN domain based on available flexible bandwidth and buffers. Using SR-policy, data traffic is steering into IP tunnels at ingress nodes and directed to dedicated network slicing. Configuration of buffer allocations are distributed via protocols like BGP and PCEP from the controller to the network devices who are executing and enforcing these configurations.
  • Network State Measurement and Telemetry: Real-time bandwidth measurement based on measurement packets helps in sensing utilized and available bandwidth on network links. This information is reported to the controller via telemetry mechanisms and used to adjust paths and slice resources. For example, when a link nears its bandwidth limit, traffic can be rerouted to idle path resources to improve overall network bandwidth utilization.
Use and Management of Multi-Level Network Buffers Since temporary bandwidth is shared and not dedicated, it exhibits weaker SLA guarantees. If traffic experiences jitter during transmission, network device buffers can absorb packets to reduce packet loss.
Specific Requirements:
  • Single Device Buffer Sharing and Management: Single devices should implement fine-grained buffer divisions based on traffic priority and slice. These buffers should be isolated to avoid mutual interference. Initial buffer resource allocation is determined by the controller and configured across all devices in the domain via control plane protocols.
  • Cross-Device Buffer Coordination: Given the nature of large data transmissions, a single device's buffer might be insufficient for absorbing bursty traffic. Therefore, multiple devices' buffers of the same fine-grained type (e.g., same priority and slice) should be used collectively. For example, if device C in the path A->B->C is congested and its buffer is insufficient, it should notify upstream devices B or A to utilize their similar buffers to absorb some traffic. This involves:
    • Control Signaling: Using control signaling packets to notify upstream devices to buffer packets, reducing the burden on the congested device. If upstream device buffers also reach a threshold, further notifications should be triggered upstream. Control signaling should include buffer index (e.g., slice ID), control instructions, and parameters. Controller configuration or segment routing can help determine upstream device addresses. Upon congestion relief, upstream devices should be notified to release buffered traffic. This notification mechanism can be inspired by IEEE PFC mechanisms but requires more granular backpressure.
    • Trigger Conditions for Buffer Coordination: The local device-triggering cross-device buffer coordination requires pre-set conditions. Controllers can configure device-specific thresholds to customize trigger conditions for each device, slice, and priority.
Requesting Source Rate Control Network devices can send rate control requests to the source via data packet marking or separate control packets. This method is useful during widespread network congestion, leveraging source rate reduction to manage traffic. Although this feedback mechanism involves a larger control loop and slower adjustments, efficiency can be improved through fast reverse notifications.
Conclusion The proposed framework addresses the challenges of large volume data transmission over WANs by enhancing traffic management and resource allocation strategies. By implementing dynamic path scheduling, advanced resource allocation, and efficient buffer management, the framework ensures efficient, reliable, and cost-effective data transmission. This approach meets the demands of data-intensive applications, providing a robust solution for large volume data transmission in WAN environments.
Security Considerations TBD.
IANA Considerations TBD.
Informative References Use Cases and Requirements for Implementing Lossless Techniques in Wide Area Networks China Unicom China Unicom Huawei Huawei This document outlines the use cases and requirements for implementing lossless data transmission techniques in Wide Area Networks (WANs), motivated by the increasing demand for high- bandwidth and reliable data transport in applications such as high- performance computing (HPC), genetic sequencing, multimedia content production and distributed training. The challenges associated with existing data transport protocols in WAN environments are discussed, along with the proposal of requirements for enhancing lossless transmission capabilities to support emerging data-intensive applications.
Acknowledgements TBD.
Contributors TBD.