1FQ(8) Linux FQ(8)
2
3
4
6 FQ - Fair Queue traffic policing
7
9 tc qdisc ... fq [ limit PACKETS ] [ flow_limit PACKETS ] [ quantum
10 BYTES ] [ initial_quantum BYTES ] [ maxrate RATE ] [ buckets NUMBER ] [
11 orphan_mask NUMBER ] [ pacing | nopacing ] [ ce_threshold TIME ]
12
13
15 FQ (Fair Queue) is a classless packet scheduler meant to be mostly used
16 for locally generated traffic. It is designed to achieve per flow pac‐
17 ing. FQ does flow separation, and is able to respect pacing require‐
18 ments set by TCP stack. All packets belonging to a socket are consid‐
19 ered as a 'flow'. For non local packets (router workload), packet hash
20 is used as fallback.
21
22 An application can specify a maximum pacing rate using the SO_MAX_PAC‐
23 ING_RATE setsockopt call. This packet scheduler adds delay between
24 packets to respect rate limitation set on each socket. Note that after
25 linux-4.20, linux adopted EDT (Earliest Departure Time) and TCP
26 directly sets the appropriate Departure Time for each skb.
27
28 Dequeueing happens in a round-robin fashion. A special FIFO queue is
29 reserved for high priority packets ( TC_PRIO_CONTROL priority), such
30 packets are always dequeued first.
31
32 FQ is non-work-conserving.
33
34 TCP pacing is good for flows having idle times, as the congestion win‐
35 dow permits TCP stack to queue a possibly large number of packets.
36 This removes the 'slow start after idle' choice, badly hitting large
37 BDP flows and applications delivering chunks of data such as video
38 streams.
39
40
42 limit
43 Hard limit on the real queue size. When this limit is reached, new
44 packets are dropped. If the value is lowered, packets are dropped so
45 that the new limit is met. Default is 10000 packets.
46
47 flow_limit
48 Hard limit on the maximum number of packets queued per flow. Default
49 value is 100.
50
51 quantum
52 The credit per dequeue RR round, i.e. the amount of bytes a flow is
53 allowed to dequeue at once. A larger value means a longer time period
54 before the next flow will be served. Default is 2 * interface MTU
55 bytes.
56
57 initial_quantum
58 The initial sending rate credit, i.e. the amount of bytes a new flow is
59 allowed to dequeue initially. This is specifically meant to allow
60 using IW10 without added delay. Default is 10 * interface MTU, i.e.
61 15140 for 'standard' ethernet.
62
63 maxrate
64 Maximum sending rate of a flow. Default is unlimited. Application
65 specific setting via SO_MAX_PACING_RATE is ignored only if it is larger
66 than this value.
67
68 buckets
69 The size of the hash table used for flow lookups. Each bucket is
70 assigned a red-black tree for efficient collision sorting. Default:
71 1024.
72
73 orphan_mask
74 For packets not owned by a socket, fq is able to mask a part of
75 skb->hash and reduce number of buckets associated with the traffic.
76 This is a DDOS prevention mechanism, and the default is 1023 (meaning
77 no more than 1024 flows are allocated for these packets)
78
79 [no]pacing
80 Enable or disable flow pacing. Default is enabled.
81
82 ce_threshold
83 sets a threshold above which all packets are marked with ECN Congestion
84 Experienced. This is useful for DCTCP-style congestion control algo‐
85 rithms that require marking at very shallow queueing thresholds.
86
87
89 #tc qdisc add dev eth0 root est 1sec 4sec fq ce_threshold 4ms
90 #tc -s -d qdisc sh dev eth0
91 qdisc fq 800e: root refcnt 9 limit 10000p flow_limit 1000p buckets 1024
92 orphan_mask 1023 quantum 3028 initial_quantum 15140 low_rate_threshold
93 550Kbit refill_delay 40.0ms ce_threshold 4.0ms
94 Sent 533368436185 bytes 352296695 pkt (dropped 0, overlimits 0
95 requeues 1339864)
96 rate 39220Mbit 3238202pps backlog 12417828b 358p requeues 1339864
97 1052 flows (852 inactive, 0 throttled)
98 112 gc, 0 highprio, 212 throttled, 21501 ns latency, 470241 ce_mark
99
101 tc(8), socket(7)
102
104 FQ was written by Eric Dumazet.
105
106
107
108iproute2 10 Sept 2015 FQ(8)