Abstract:Routers employ queues to temporarily hold packets when the scheduler cannot immediately process them. Congestion occurs when the arrival rate of packets exceeds the processing capacity, leading to increased queueing delay. Over time, Active Queue Management (AQM) strategies have focused on directly draining packets from queues to alleviate congestion and reduce queuing delay. On Programmable Data Plane (PDP) hardware, AQMs traditionally reside in the Egress pipeline due to the availability of queue delay information there. We argue that this approach wastes the router's resources because the dropped packet has already consumed the entire pipeline of the device. In this work, we propose ingress Random Early Detection (iRED), a more efficient approach that addresses the Egress drop problem. iRED is a disaggregated P4-AQM fully implemented in programmable data plane hardware and also supports Low Latency, Low Loss, and Scalable Throughput (L4S) framework, saving device pipeline resources by dropping packets in the Ingress block. To evaluate iRED, we conducted three experiments using a Tofino2 programmable switch: i) An in-depth analysis of state-of-the-art AQMs on PDP hardware, using 12 different network configurations varying in bandwidth, Round-Trip Time (RTT), and Maximum Transmission Unit (MTU). The results demonstrate that iRED can significantly reduce router resource consumption, with up to a 10x reduction in memory usage, 12x fewer processing cycles, and 8x less power consumption for the same traffic load; ii) A performance evaluation regarding the L4S framework. The results prove that iRED achieves fairness in bandwidth usage for different types of traffic (classic and scalable); iii) A comprehensive analysis of the QoS in a real setup of a DASH) technology. iRED demonstrated up to a 2.34x improvement in FPS and a 4.77x increase in the video player buffer fill.

Improving PIE's performance over high-delay paths

Mred: A New Adaptive Active Queue Management Algorithm

Research on some active queue management algorithms

A New Queue Discipline for Reducing Bufferbloat Effects in HetNet Concurrent Multipath Transfer

Insights from Curvy RED (Random Early Detection)

A New Rate-Based Active Queue Management Scheme over Wired-cum-Wireless Networks

Bounding Queue Delay in Cellular Networks to Support Ultra-Low Latency Applications

Fair Scheduling for Delay-Sensitive VoIP Traffic

Accelerating End-host Congestion Response using P4 Programmable Switches

Low Delay Random Linear Coding and Scheduling Over Multiple Interfaces

Optimizing Flow Completion Time Via Adaptive Buffer Management in Data Center Networks

A New Fair Active Queue Management Algorithm

Performance Evaluation of Active Queue Management Algorithms in Large Network

A Novel Packet Dropping Mechanism for Active Queue Management

iRED: A disaggregated P4-AQM fully implemented in programmable data plane hardware

Towards Power Efficient High Performance Packet I/O

A robust active queue management algorithm in large delay networks

A Practical On-line Pacing Scheme at Edges of Small Buffer Networks

Design and Analysis of a Stable Queue Control Scheme for the Internet

Explicit Dropping Notification in Data Centers

Re-Architecting Buffer Management in Lossless Ethernet