CoDel

CoDel
Name	CoDel
Full name	Controlled Delay
Authors	Van Jacobson, Kathleen Nichols
Year	2012
Type	Active Queue Management
Area	Computer networking
Venue	ACM SIGCOMM

CoDel

CoDel is an active queue management algorithm designed to control queueing delay in packet-switched networks. It arose as a response to persistent bufferbloat observed in broadband and datacenter environments, seeking to provide a parameter-free mechanism that required minimal operator tuning. CoDel has been evaluated in academic testbeds, deployed in router firmware, and influenced subsequent research on latency-sensitive networking.

Introduction

CoDel was introduced by Van Jacobson and Kathleen Nichols to address excessive latency caused by oversized buffers in routers and switches. It targets long-term queueing delay rather than queue length, operating independently from specific transport protocols such as Transmission Control Protocol, User Datagram Protocol, and Stream Control Transmission Protocol. CoDel contrasts with earlier Active Queue Management schemes like Random Early Detection and Random Early Detection (RED), emphasizing robustness across diverse link technologies including Ethernet, Wi-Fi, DOCSIS, and SONET.

Background and Motivation

Bufferbloat became prominent in discussions involving contributors from IETF working groups and presentations at venues such as USENIX, ACM SIGCOMM, and ACM CoNEXT. Observers from organizations including Google, Microsoft, Netflix, and Akamai Technologies reported poor interactivity and high latency for applications like Voice over IP, WebRTC, and online gaming on networks provisioned with large buffers. Earlier AQM research by groups at Bell Labs, Stanford University, and MIT produced mechanisms such as RED, BLUE (queue management), and PI controller concepts used by equipment vendors like Cisco Systems and Juniper Networks. CoDel's motivation drew on this history to offer an approach that mitigated bufferbloat without complex parameter tuning.

Algorithm Overview

CoDel monitors the minimum standing queueing delay over sliding intervals and makes drop decisions when the minimum exceeds a target threshold. It measures per-packet sojourn time using timestamping similar to methods used in timestamp-based schedulers studied at Carnegie Mellon University and University of California, Berkeley. When persistent delay is detected, CoDel invokes packet drops at intervals derived from an inverse square-root function to reduce queue occupancy; the control law echoes mathematical ideas encountered in control theory and queueing theory research at institutions such as California Institute of Technology and Princeton University. CoDel’s design choices reference experimental methodologies used by researchers at Stanford University and University College London.

Performance and Evaluation

Evaluations of CoDel in testbeds and simulations have involved tools and platforms like ns-2, ns-3, Mininet, and hardware testbeds maintained by RIPE NCC and GEANT. Benchmarks compared CoDel to RED, tail-drop, and later AQM schemes under traffic mixes including HTTP/2, QUIC, BitTorrent, and bulk FTP flows. Results reported lower median and 95th-percentile latency for interactive flows on links provided by carriers such as Comcast, British Telecom, and Vodafone in measurement studies. Academic evaluations published in venues like USENIX ATC, IEEE INFOCOM, and ACM SIGCOMM showed improvements in latency-sensitive application performance for SSH sessions, VoIP calls, and video conferencing with clients such as Skype and Zoom.

Implementations and Deployment

CoDel has been implemented in operating systems and network stacks including Linux, FreeBSD, and router firmware projects like OpenWrt and pfSense. Major networking equipment vendors and community projects integrated CoDel into devices by companies such as Netgear and Ubiquiti Networks. Cloud providers with networking teams at Amazon Web Services, Google Cloud Platform, and Microsoft Azure conducted experiments inspired by CoDel principles. Deployment efforts included contributions from open-source contributors associated with Kernel.org, and device maintainers collaborating through GitHub repositories.

Criticisms and Limitations

Critics pointed to edge cases where CoDel’s interval-based logic could interact poorly with bursty traffic patterns generated by applications like YouTube or peer-to-peer clients such as uTorrent. Research groups at University of Cambridge and ETH Zurich highlighted scenarios with measurement bias and fairness concerns in mixed-RTT environments, echoing earlier debates around RED and flow fairness studied at University of California, Los Angeles and Columbia University. Vendors concerned with deterministic behavior in real-time industrial networks such as those used by Siemens and Schneider Electric raised issues about predictable latency under CoDel in constrained embedded platforms.

Variants and Extensions

Several extensions built on CoDel’s core ideas, including fq-CoDel (fair queuing with CoDel), Dual-Queue CoDel, and Adaptive CoDel variants. fq-CoDel combined flow scheduling concepts from FQ (fair queuing), research at MIT, and per-flow deficit round robin implementations used in Cisco IOS to provide fairness across competing flows. Derivative work by researchers at University of Washington, Aalto University, and Tsinghua University produced adaptive parameter tuning, integration with Explicit Congestion Notification, and hybrid schemes tailored for Cellular network links operated by carriers like Verizon and AT&T. Extensions influenced newer AQM proposals standardized and discussed within IETF forums and presented at conferences such as ACM SIGCOMM and IEEE INFOCOM.

Category:Computer networking algorithms