NAME
tcp_bbr – TCP Bottleneck Bandwidth and Round-Trip Time Algorithm
SYNOPSIS
To load the driver as a module at boot time, place the following line in loader.conf(5):
tcp_bbr_load="YES"
To enable the TCP stack you must place the following line in the sysctl.conf(5):
net.inet.tcp.functions_default=bbr
DESCRIPTION
Bottleneck bandwidth and round-trip time (BBR) is a congestion control algorithm which seeks high throughput with a small queue by probing BW and RTT. It is a round-up redesign of congestion control, which is not loss-based, delay-based, ECN-based or AIMD-based.
The core design of BBR is about creating a model graph of the network path by estimating the maximum BW and minimum RTT on each ACK.
MIB Variables
The algorithm exposes the following scopes in the net.inet.tcp.bbr branch of the sysctl(3) MIB:
- cwnd
-
Cwnd controls, for example "target cwnd rtt measurement" and "BBR initial window".
- measure
-
Measurement controls.
- pacing
-
Connection pacing controls.
- policer
-
Policer controls, for example "false detection threshold" and "loss threshold".
- probertt
-
Probe RTT controls.
- startup
-
Startup controls.
- states
-
State controls.
- timeout
-
Time out controls.
Besides the variables within the above scopes the following variables are also exposed in the net.inet.tcp.bbr branch:
- clrlost
-
Clear lost counters.
- software_pacing
-
Total number of software paced flows.
- hdwr_pacing
-
Total number of hardware paced flows.
- enob_no_hdwr_pacing
-
Total number of enobufs for non-hardware paced flows.
- enob_hdwr_pacing
-
Total number of enobufs for hardware paced flows.
- rtt_tlp_thresh
-
What divisor for TLP rtt/retran will be added (1=rtt, 2=1/2 rtt etc).
- reorder_fade
-
Does reorder detection fade, if so how many ms (0 means never).
- reorder_thresh
-
What factor for rack will be added when seeing reordering (shift right).
- bb_verbose
-
Should BBR black box logging be verbose.
- sblklimit
-
When do we start ignoring small sack blocks.
- resend_use_tso
-
Can resends use TSO?
- data_after_close
-
Do we hold off sending a RST until all pending data is ack'd.
- kill_paceout
-
When we hit this many errors in a row, kill the session?
- error_paceout
-
When we hit an error what is the min to pace out in usec's?
- cheat_rxt
-
Do we burst 1ms between sends on retransmissions (like rack)?
- minrto
-
Minimum RTO in ms.
SEE ALSO
cc_chd(4), cc_cubic(4), cc_hd(4), cc_htcp(4), cc_newreno(4), cc_vegas(4), h_ertt(4), mod_cc(4), tcp(4), tcp_rack(4), mod_cc(9) Neal Cardwell, Yuchung Cheng, Stephen Gunn, Soheil Hassas Yeganeh, and Van Jacobson, “BBR: Congestion-Based Congestion Control”, ACM Queue, Vol. 14, September / October 2016. Dominik Scholz, Benedikt Jaeger, Lukas Schwaighofer, Daniel Raumer, Fabien Geyer, and Georg Carle, “Towards a Deeper Understanding of TCP BBR Congestion Control”, IFIP Networking 2018, May 2018, https://www.net.in.tum.de/fileadmin/bibtex/publications/papers/IFIP-Networking-2018-TCP-BBR.pdf.
HISTORY
The tcp_bbr congestion control module first appeared in FreeBSD 13.0.
AUTHORS
The tcp_bbr congestion control module was written by Randall Stewart ⟨[email protected]⟩ and sponsored by Netflix, Inc. This manual page was written by Gordon Bergling ⟨[email protected]⟩.