| 1 |
.TH PIE 8 "16 January 2014" "iproute2" "Linux" |
| 2 |
.SH NAME |
| 3 |
PIE \- Proportional Integral controller-Enhanced AQM algorithm |
| 4 |
.SH SYNOPSIS |
| 5 |
.B tc qdisc ... pie |
| 6 |
[ |
| 7 |
.B limit |
| 8 |
PACKETS ] [ |
| 9 |
.B target |
| 10 |
TIME ] [ |
| 11 |
.B tupdate |
| 12 |
TIME ] [ |
| 13 |
.B alpha |
| 14 |
int ] [ |
| 15 |
.B beta |
| 16 |
int ] [ |
| 17 |
.B ecn |
| 18 |
| |
| 19 |
.B noecn |
| 20 |
] [ |
| 21 |
.B bytemode |
| 22 |
| |
| 23 |
.B nobytemode |
| 24 |
] [ |
| 25 |
.B dq_rate_estimator |
| 26 |
| |
| 27 |
.B no_dq_rate_estimator |
| 28 |
] |
| 29 |
|
| 30 |
.SH DESCRIPTION |
| 31 |
Proportional Integral controller-Enhanced (PIE) is a control theoretic active |
| 32 |
queue management scheme. It is based on the proportional integral controller but |
| 33 |
aims to control delay. The main design goals are |
| 34 |
o Low latency control |
| 35 |
o High link utilization |
| 36 |
o Simple implementation |
| 37 |
o Guaranteed stability and fast responsiveness |
| 38 |
|
| 39 |
.SH ALGORITHM |
| 40 |
PIE is designed to control delay effectively. First, an average dequeue rate is |
| 41 |
estimated based on the standing queue. The rate is used to calculate the current |
| 42 |
delay. Then, on a periodic basis, the delay is used to calculate the dropping |
| 43 |
probability. Finally, on arrival, a packet is dropped (or marked) based on this |
| 44 |
probability. |
| 45 |
|
| 46 |
PIE makes adjustments to the probability based on the trend of the delay i.e. |
| 47 |
whether it is going up or down.The delay converges quickly to the target value |
| 48 |
specified. |
| 49 |
|
| 50 |
alpha and beta are statically chosen parameters chosen to control the drop probability |
| 51 |
growth and are determined through control theoretic approaches. alpha determines how |
| 52 |
the deviation between the current and target latency changes probability. beta exerts |
| 53 |
additional adjustments depending on the latency trend. |
| 54 |
|
| 55 |
The drop probability is used to mark packets in ecn mode. However, as in RED, |
| 56 |
beyond 10% packets are dropped based on this probability. The bytemode is used |
| 57 |
to drop packets proportional to the packet size. |
| 58 |
|
| 59 |
Additional details can be found in the paper cited below. |
| 60 |
|
| 61 |
.SH PARAMETERS |
| 62 |
.SS limit |
| 63 |
limit on the queue size in packets. Incoming packets are dropped when this limit |
| 64 |
is reached. Default is 1000 packets. |
| 65 |
|
| 66 |
.SS target |
| 67 |
is the expected queue delay. The default target delay is 15ms. |
| 68 |
|
| 69 |
.SS tupdate |
| 70 |
is the frequency at which the system drop probability is calculated. The default is 15ms. |
| 71 |
|
| 72 |
.SS alpha |
| 73 |
.SS beta |
| 74 |
alpha and beta are parameters chosen to control the drop probability. These |
| 75 |
should be in the range between 0 and 32. |
| 76 |
|
| 77 |
.SS ecn | noecn |
| 78 |
is used to mark packets instead of dropping. |
| 79 |
.B ecn |
| 80 |
to turn on ecn mode, |
| 81 |
.B noecn |
| 82 |
to turn off ecn mode. By default, |
| 83 |
.B ecn |
| 84 |
is turned off. |
| 85 |
|
| 86 |
.SS bytemode | nobytemode |
| 87 |
is used to scale drop probability proportional to packet size. |
| 88 |
.B bytemode |
| 89 |
to turn on bytemode, |
| 90 |
.B nobytemode |
| 91 |
to turn off bytemode. By default, |
| 92 |
.B bytemode |
| 93 |
is turned off. |
| 94 |
|
| 95 |
.SS dq_rate_estimator | no_dq_rate_estimator |
| 96 |
is used to calculate delay using Little's law. |
| 97 |
.B dq_rate_estimator |
| 98 |
to turn on dq_rate_estimator, |
| 99 |
.B no_dq_rate_estimator |
| 100 |
to turn off no_dq_rate_estimator. By default, |
| 101 |
.B dq_rate_estimator |
| 102 |
is turned off. |
| 103 |
|
| 104 |
.SH EXAMPLES |
| 105 |
# tc qdisc add dev eth0 root pie |
| 106 |
# tc -s qdisc show |
| 107 |
qdisc pie 8036: dev eth0 root refcnt 2 limit 1000p target 15.0ms tupdate 16.0ms alpha 2 beta 20 |
| 108 |
Sent 31216108 bytes 20800 pkt (dropped 80, overlimits 0 requeues 0) |
| 109 |
backlog 16654b 11p requeues 0 |
| 110 |
prob 0.006161 delay 15666us |
| 111 |
pkts_in 20811 overlimit 0 dropped 80 maxq 50 ecn_mark 0 |
| 112 |
|
| 113 |
# tc qdisc add dev eth0 root pie dq_rate_estimator |
| 114 |
# tc -s qdisc show |
| 115 |
qdisc pie 8036: dev eth0 root refcnt 2 limit 1000p target 15.0ms tupdate 16.0ms alpha 2 beta 20 |
| 116 |
Sent 63947420 bytes 42414 pkt (dropped 41, overlimits 0 requeues 0) |
| 117 |
backlog 271006b 179p requeues 0 |
| 118 |
prob 0.000092 delay 22200us avg_dq_rate 12145996 |
| 119 |
pkts_in 41 overlimit 343 dropped 0 maxq 50 ecn_mark 0 |
| 120 |
|
| 121 |
# tc qdisc add dev eth0 root pie limit 100 target 20ms tupdate 30ms ecn |
| 122 |
# tc -s qdisc show |
| 123 |
qdisc pie 8036: dev eth0 root refcnt 2 limit 100p target 20.0ms tupdate 32.0ms alpha 2 beta 20 ecn |
| 124 |
Sent 6591724 bytes 4442 pkt (dropped 27, overlimits 0 requeues 0) |
| 125 |
backlog 18168b 12p requeues 0 |
| 126 |
prob 0.008845 delay 11348us |
| 127 |
pkts_in 4454 overlimit 0 dropped 27 maxq 65 ecn_mark 0 |
| 128 |
|
| 129 |
# tc qdisc add dev eth0 root pie limit 100 target 50ms tupdate 30ms bytemode |
| 130 |
# tc -s qdisc show |
| 131 |
qdisc pie 8036: dev eth0 root refcnt 2 limit 100p target 50.0ms tupdate 32.0ms alpha 2 beta 20 bytemode |
| 132 |
Sent 1616274 bytes 1137 pkt (dropped 0, overlimits 0 requeues 0) |
| 133 |
backlog 13626b 9p requeues 0 |
| 134 |
prob 0.000000 delay 0us |
| 135 |
pkts_in 1146 overlimit 0 dropped 0 maxq 23 ecn_mark 0 |
| 136 |
|
| 137 |
.SH SEE ALSO |
| 138 |
.BR tc (8), |
| 139 |
.BR tc-codel (8) |
| 140 |
.BR tc-red (8) |
| 141 |
|
| 142 |
.SH SOURCES |
| 143 |
o RFC 8033: https://tools.ietf.org/html/rfc8033 |
| 144 |
|
| 145 |
.SH AUTHORS |
| 146 |
PIE was implemented by Vijay Subramanian and Mythili Prabhu, also the authors of |
| 147 |
this man page. Please report bugs and corrections to the Linux networking |
| 148 |
development mailing list at <[email protected]>. |