NAME
tcp_rack – TCP RACK-TLP Loss Detection Algorithm for TCP
SYNOPSIS
To load the TCP stack as a module at boot time, place the following line in loader.conf(5):
tcp_rack_load="YES"
To enable the TCP stack, place the following line in the sysctl.conf(5):
net.inet.tcp.functions_default=rack
DESCRIPTION
RACK-TLP uses per-segment transmit timestamps and selective acknowledgments (SACKs) and has two parts. Recent Acknowledgment (RACK) starts fast recovery quickly using time-based inferences derived from acknowledgment (ACK) feedback, and Tail Loss Probe (TLP) leverages RACK and sends a probe packet to trigger ACK feedback to avoid retransmission timeout (RTO) events.
Compared to the widely used duplicate acknowledgment (DupAck) threshold approach, RACK-TLP detects losses more efficiently when there are application-limited flights of data, lost retransmissions, or data packet reordering events.
It is intended to be an alternative to the DupAck threshold approach.
MIB Variables
The algorithm exposes the following scopes in the net.inet.tcp.rack branch of the sysctl(3) MIB:
- net.inet.tcp.rack.misc
-
Misc related controls
- net.inet.tcp.rack.features
-
Feature controls
- net.inet.tcp.rack.measure
-
Measure related controls
- net.inet.tcp.rack.timers
-
Timer related controls
- net.inet.tcp.rack.tlp
-
TLP and Rack related Controls
- net.inet.tcp.rack.timely
-
Rack Timely RTT Controls
- net.inet.tcp.rack.hdwr_pacing
-
Pacing related Controls
- net.inet.tcp.rack.pacing
-
Pacing related Controls
- net.inet.tcp.rack.tp
-
Rack tracepoint facility
- net.inet.tcp.rack.probertt
-
ProbeRTT related Controls
- net.inet.tcp.rack.stats
-
Rack Counters
- net.inet.tcp.rack.sack_attack
-
Rack Sack Attack Counters and Controls
Besides the variables within the above scopes the following variables are also exposed in the net.inet.tcp.rack branch:
- net.inet.tcp.rack.clear
-
Clear counters
- net.inet.tcp.rack.opts
-
RACK Option Stats
- net.inet.tcp.rack.outsize
-
MSS send sizes
- net.inet.tcp.rack.req_measure_cnt
-
If doing dynamic pacing, how many measurements must be in before we start pacing?
- net.inet.tcp.rack.use_pacing
-
If set we use pacing, if clear we use only the original burst mitigation
- net.inet.tcp.rack.rate_sample_method
-
What method should we use for rate sampling 0=high, 1=low
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_bbr(4), mod_cc(9) Neal Cardwell, Yuchung Cheng, Nandita Dukkipati, and Priyaranjan Jha, The RACK-TLP Loss Detection Algorithm for TCP, February 2021, RFC 8985. M. Allman, V. Paxson, and E. Blanton, TCP Congestion Control, September 2009, RFC 5681. M. Mathis, Nandita Dukkipati, and Yuchung Cheng, Proportional Rate Reduction for TCP, May 2013, RFC 6937.
HISTORY
The tcp_rack congestion control module first appeared in FreeBSD 13.0.
AUTHORS
The tcp_rack 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]⟩.