Faculty of Electrical Engineering and Computer, Ningbo University, 818 Fenghua Road, Ningbo 315211, China.
Sensors (Basel). 2021 May 20;21(10):3568. doi: 10.3390/s21103568.
The increase in network applications diversity and different service quality requirements lead to service differentiation, making it more important than ever. In Wide Area Network (WAN), the non-responsive Long-Term Fast (LTF) flows are the main contributors to network congestion. Therefore, detecting and suppressing non-responsive LTF flows represent one of the key points for providing data transmission with controllable delay and service differentiation. However, the existing single-queue management algorithms are designed to serve only a small number of applications with similar requirements (low latency, high throughput, etc.). The lack of mechanisms to distinguish different traffic makes it difficult to implement differentiated services. This paper proposes an active queue management scheme, namely, SQM-LRU, which realizes service differentiation based on Shadow Queue (SQ) and improved Least-Recently-Used (LRU) strategy. The algorithm consists of three essential components: First, the flow detection module is based on the SQ and improved LRU. This module is used to detect non-responsive LTF flows. Second, different flows will be put into corresponding high or low priority sub-queues depending on the flow detection results. Third, the dual-queue adopts CoDel and RED, respectively, to manage packets. SQM-LRU intends to satisfy the stringent delay requirements of responsive flow while maximizing the throughput of non-responsive LTF flow. Our simulation results show that SQM-LRU outperforms traditional solutions with significant improvement in flow detection and reduces the delay, jitter, and Flow Completion Time (FCT) of responsive flow. As a result, it reduced the FCT by up to 50% and attained 95% of the link utilization. Additionally, the low overhead and the operations incur O(1) cost per packet, making it practical for the real network.
网络应用的多样性和不同服务质量要求的增加导致了服务差异化，这比以往任何时候都更加重要。在广域网 (WAN) 中，无响应的长期快速 (LTF) 流是导致网络拥塞的主要原因。因此，检测和抑制无响应的 LTF 流是提供具有可控制延迟和服务差异化的数据传输的关键之一。然而，现有的单队列管理算法旨在为具有相似要求（低延迟、高吞吐量等）的少数应用程序提供服务。缺乏区分不同流量的机制使得实现差异化服务变得困难。本文提出了一种主动队列管理方案，即 SQM-LRU，它基于影子队列 (SQ) 和改进的最近最少使用 (LRU) 策略实现服务差异化。该算法由三个基本组件组成：首先，流检测模块基于 SQ 和改进的 LRU。该模块用于检测无响应的 LTF 流。其次，根据流检测结果，将不同的流放入相应的高或低优先级子队列中。最后，双队列分别采用 CoDel 和 RED 来管理数据包。SQM-LRU 的目的是在满足响应性流的严格延迟要求的同时，最大限度地提高无响应 LTF 流的吞吐量。我们的仿真结果表明，SQM-LRU 优于传统解决方案，在流检测方面有显著的改进，并降低了响应性流的延迟、抖动和流完成时间 (FCT)。因此，它将 FCT 减少了 50%，达到了 95%的链路利用率。此外，低开销和操作的开销为每个数据包的 O(1) 成本，使其适用于实际网络。
