DESCRIPTIONThe HD congestion control algorithm is an implementation of the Hamilton Institute's delay-based congestion control which aims to keep network queuing delays below a particular threshold (queue_threshold).
HD probabilistically reduces the congestion window (cwnd) based on its estimate of the network queuing delay. The probability of reducing cwnd is zero at hd_qmin or less, rising to a maximum at queue_threshold, and then back to zero at the maximum queuing delay.
Loss-based congestion control algorithms such as NewReno probe for network capacity by filling queues until there is a packet loss. HD competes with loss-based congestion control algorithms by allowing its probability of reducing cwnd to drop from a maximum at queue_threshold to be zero at the maximum queuing delay. This has been shown to work well when the bottleneck link is highly multiplexed.
MIB VariablesThe algorithm exposes the following tunable variables in the net.inet.tcp.cc.hd branch of the sysctl(3) MIB:
- Queueing congestion threshold (qth) in ticks. Default is 20.
- Per packet maximum backoff probability as a percentage. Default is 5.
- Minimum queuing delay threshold (qmin) in ticks. Default is 5.
ACKNOWLEDGEMENTSDevelopment and testing of this software were made possible in part by grants from the FreeBSD Foundation and Cisco University Research Program Fund at Community Foundation Silicon Valley.
FUTURE WORKThe Hamilton Institute have recently made some improvements to the algorithm implemented by this module and have called it Coexistent-TCP (C-TCP). The improvements should be evaluated and potentially incorporated into this module.
HISTORYThe congestion control module first appeared in Fx 9.0 .
The module was first released in 2010 by David Hayes whilst working on the NewTCP research project at Swinburne University of Technology's Centre for Advanced Internet Architectures, Melbourne, Australia. More details are available at:
AUTHORSAn -nosplit The congestion control module and this manual page were written by An David Hayes Aq [email protected] .