| 1 |
.TH CoDel 8 "23 May 2012" "iproute2" "Linux" |
| 2 |
.SH NAME |
| 3 |
CoDel \- Controlled-Delay Active Queue Management algorithm |
| 4 |
.SH SYNOPSIS |
| 5 |
.B tc qdisc ... codel |
| 6 |
[ |
| 7 |
.B limit |
| 8 |
PACKETS ] [ |
| 9 |
.B target |
| 10 |
TIME ] [ |
| 11 |
.B interval |
| 12 |
TIME ] [ |
| 13 |
.B ecn |
| 14 |
| |
| 15 |
.B noecn |
| 16 |
] [ |
| 17 |
.B ce_threshold |
| 18 |
TIME ] |
| 19 |
|
| 20 |
.SH DESCRIPTION |
| 21 |
CoDel (pronounced "coddle") is an adaptive "no-knobs" active queue management |
| 22 |
algorithm (AQM) scheme that was developed to address the shortcomings of |
| 23 |
RED and its variants. It was developed with the following goals |
| 24 |
in mind: |
| 25 |
.IP * 4 |
| 26 |
It should be parameterless. |
| 27 |
.IP * |
| 28 |
It should keep delays low while permitting bursts of traffic. |
| 29 |
.IP * |
| 30 |
It should control delay. |
| 31 |
.IP * |
| 32 |
It should adapt dynamically to changing link rates with no impact on |
| 33 |
utilization. |
| 34 |
.IP * |
| 35 |
It should be simple and efficient and should scale from simple to |
| 36 |
complex routers. |
| 37 |
|
| 38 |
.SH ALGORITHM |
| 39 |
CoDel comes with three major innovations. Instead of using queue size or queue |
| 40 |
average, it uses the local minimum queue as a measure of the standing/persistent queue. |
| 41 |
Second, it uses a single state-tracking variable of the minimum delay to see where it |
| 42 |
is relative to the standing queue delay. Third, instead of measuring queue size |
| 43 |
in bytes or packets, it is measured in packet-sojourn time in the queue. |
| 44 |
|
| 45 |
CoDel measures the minimum local queue delay (i.e. standing queue delay) and |
| 46 |
compares it to the value of the given acceptable queue delay |
| 47 |
.B target. |
| 48 |
As long as the minimum queue delay is less than |
| 49 |
.B target |
| 50 |
or the buffer contains fewer than MTU worth of bytes, packets are not dropped. |
| 51 |
Codel enters a dropping mode when the minimum queue delay has exceeded |
| 52 |
.B target |
| 53 |
for a time greater than |
| 54 |
.B interval. |
| 55 |
In this mode, packets are dropped at different drop times which is set by a |
| 56 |
control law. The control law ensures that the packet drops cause a linear change |
| 57 |
in the throughput. Once the minimum delay goes below |
| 58 |
.B target, |
| 59 |
packets are no longer dropped. |
| 60 |
|
| 61 |
Additional details can be found in the paper cited below. |
| 62 |
|
| 63 |
.SH PARAMETERS |
| 64 |
.SS limit |
| 65 |
is the hard limit on the real queue size. When this limit is reached, incoming packets |
| 66 |
are dropped. If the value is lowered, packets are dropped so that the new limit is |
| 67 |
met. Default is 1000 packets. |
| 68 |
|
| 69 |
.SS target |
| 70 |
is the acceptable minimum standing/persistent queue delay. This minimum delay |
| 71 |
is identified by tracking the local minimum queue delay that packets experience. |
| 72 |
Default and recommended value is 5ms. |
| 73 |
|
| 74 |
.SS interval |
| 75 |
is used to ensure that the measured minimum delay does not become too stale. The |
| 76 |
minimum delay must be experienced in the last epoch of length |
| 77 |
.B interval. |
| 78 |
It should be set on the order of the worst-case RTT through the bottleneck to |
| 79 |
give endpoints sufficient time to react. Default value is 100ms. |
| 80 |
|
| 81 |
.SS ecn | noecn |
| 82 |
can be used to mark packets instead of dropping them. If |
| 83 |
.B ecn |
| 84 |
has been enabled, |
| 85 |
.B noecn |
| 86 |
can be used to turn it off and vice-a-versa. By default, |
| 87 |
.B ecn |
| 88 |
is turned off. |
| 89 |
|
| 90 |
.SS ce_threshold |
| 91 |
sets a threshold above which all packets are marked with ECN Congestion |
| 92 |
Experienced. This is useful for DCTCP-style congestion control algorithms that |
| 93 |
require marking at very shallow queueing thresholds. |
| 94 |
|
| 95 |
|
| 96 |
.SH EXAMPLES |
| 97 |
# tc qdisc add dev eth0 root codel |
| 98 |
# tc -s qdisc show |
| 99 |
qdisc codel 801b: dev eth0 root refcnt 2 limit 1000p target 5.0ms |
| 100 |
interval 100.0ms |
| 101 |
Sent 245801662 bytes 275853 pkt (dropped 0, overlimits 0 requeues 24) |
| 102 |
backlog 0b 0p requeues 24 |
| 103 |
count 0 lastcount 0 ldelay 2us drop_next 0us |
| 104 |
maxpacket 7306 ecn_mark 0 drop_overlimit 0 |
| 105 |
|
| 106 |
# tc qdisc add dev eth0 root codel limit 100 target 4ms interval 30ms ecn |
| 107 |
# tc -s qdisc show |
| 108 |
qdisc codel 801c: dev eth0 root refcnt 2 limit 100p target 4.0ms |
| 109 |
interval 30.0ms ecn |
| 110 |
Sent 237573074 bytes 268561 pkt (dropped 0, overlimits 0 requeues 5) |
| 111 |
backlog 0b 0p requeues 5 |
| 112 |
count 0 lastcount 0 ldelay 76us drop_next 0us |
| 113 |
maxpacket 2962 ecn_mark 0 drop_overlimit 0 |
| 114 |
|
| 115 |
|
| 116 |
.SH SEE ALSO |
| 117 |
.BR tc (8), |
| 118 |
.BR tc-red (8) |
| 119 |
|
| 120 |
.SH SOURCES |
| 121 |
Kathleen Nichols and Van Jacobson, "Controlling Queue Delay", ACM Queue, |
| 122 |
http://queue.acm.org/detail.cfm?id=2209336 |
| 123 |
|
| 124 |
.SH AUTHORS |
| 125 |
CoDel was implemented by Eric Dumazet and David Taht. This manpage was written |
| 126 |
by Vijay Subramanian. Please reports corrections to the Linux Networking |
| 127 |
mailing list <[email protected]>. |