Penguin
Recent Changes
Annotated edit history of tc-htb(8) version 4, including all changes. View license author blame.
Rev Author # Line
1 perry 1 HTB
2 !!!HTB
3 NAME
4 SYNOPSIS
5 DESCRIPTION
6 SHAPING ALGORITHM
7 CLASSIFICATION
8 LINK SHARING ALGORITHM
9 QDISC
10 CLASSES
11 NOTES
12 BUGS
13 SEE ALSO
14 AUTHOR
15 ----
16 !!NAME
17
18
19 HTB - Hierarchy Token Bucket
20 !!SYNOPSIS
21
22
23 __tc qdisc ... dev__ dev __( parent__ classid __|
24 root) [[ handle__ major: __] htb [[ default__ minor-id
25 __]__
26
27
28 __tc class ... dev__ dev __parent__ major:[[minor] __[[
29 classid__ major:minor __] htb rate__ rate __[[ ceil__
30 rate __] burst__ bytes __[[ cburst__ bytes __] [[
31 prio__ priority __]__
32 !!DESCRIPTION
33
34
35 HTB is meant as a more understandable and intuitive
36 replacement for the CBQ qdisc in Linux. Both CBQ and HTB
37 help you to control the use of the outbound bandwidth on a
38 given link. Both allow you to use one physical link to
39 simulate several slower links and to send different kinds of
40 traffic on different simulated links. In both cases, you
41 have to specify how to divide the physical link into
42 simulated links and how to decide which simulated link to
43 use for a given packet to be sent.
44
45
46 Unlike CBQ, HTB shapes traffic based on the Token Bucket
47 Filter algorithm which does not depend on interface
48 characteristics and so does not need to know the underlying
49 bandwidth of the outgoing interface.
50 !!SHAPING ALGORITHM
51
52
53 Shaping works as documented in __tc-tbf
54 (8).__
55 !!CLASSIFICATION
56
57
58 Within the one HRB instance many classes may exist. Each of
59 these classes contains another qdisc, by default
4 perry 60 tc-pfifo(8).
1 perry 61
62
63 When enqueueing a packet, HTB starts at the root and uses
64 various methods to determine which class should receive the
65 data.
66
67
68 In the absence of uncommon configuration options, the
69 process is rather easy. At each node we look for an
70 instruction, and then go to the class the instruction refers
71 us to. If the class found is a barren leaf-node (without
72 children), we enqueue the packet there. If it is not yet a
73 leaf node, we do the whole thing over again starting from
74 that node.
75
76
77 The following actions are performed, in order at each node
78 we visit, until one sends us to another node, or terminates
79 the process.
80
81
82 (i)
83
84
85 Consult filters attached to the class. If sent to a
86 leafnode, we are done. Otherwise, restart.
87
88
89 (ii)
90
91
92 If none of the above returned with an instruction, enqueue
93 at this node.
94
95
96 This algorithm makes sure that a packet always ends up
97 somewhere, even while you are busy building your
98 configuration.
99 !!LINK SHARING ALGORITHM
100
101
102 FIXME
103 !!QDISC
104
105
106 The root of a CBQ qdisc class tree has the following
107 parameters:
108
109
110 parent major:minor | root
111
112
113 This mandatory parameter determines the place of the CBQ
114 instance, either at the __root__ of an interface or
115 within an existing class.
116
117
118 handle major:
119
120
121 Like all other qdiscs, the CBQ can be assigned a handle.
122 Should consist only of a major number, followed by a colon.
123 Optional, but very useful if classes will be generated
124 within this qdisc.
125
126
127 default minor-id
128
129
130 Unclassified traffic gets sent to the class with this
131 minor-id.
132 !!CLASSES
133
134
135 Classes have a host of parameters to configure their
136 operation.
137
138
139 parent major:minor
140
141
142 Place of this class within the hierarchy. If attached
143 directly to a qdisc and not to another class, minor can be
144 omitted. Mandatory.
145
146
147 classid major:minor
148
149
150 Like qdiscs, classes can be named. The major number must be
151 equal to the major number of the qdisc to which it belongs.
152 Optional, but needed if this class is going to have
153 children.
154
155
156 prio priority
157
158
159 In the round-robin process, classes with the lowest priority
160 field are tried for packets first. Mandatory.
161
162
163 rate rate
164
165
166 Maximum rate this class and all its children are guaranteed.
167 Mandatory.
168
169
170 ceil rate
171
172
173 Maximum rate at which a class can send, if its parent has
174 bandwidth to spare. Defaults to the configured rate, which
175 implies no borrowing
176
177
178 burst bytes
179
180
181 Amount of bytes that can be burst at __ceil__ speed, in
182 excess of the configured __rate.__ Should be at least as
183 high as the highest burst of all children.
184
185
186 cburst bytes
187
188
189 Amount of bytes that can be burst at 'infinite' speed, in
190 other words, as fast as the interface can transmit them. For
191 perfect evening out, should be equal to at most one average
192 packet. Should be at least as high as the highest cburst of
193 all children.
194 !!NOTES
195
196
197 Due to Unix timing constraints, the maximum ceil rate is not
198 infinite and may in fact be quite low. On Intel, there are
199 100 timer events per second, the maximum rate is that rate
200 at which 'burst' bytes are sent each timer tick. From this,
201 the mininum burst size for a specified rate can be
202 calculated. For i386, a 10mbit rate requires a 12 kilobyte
203 burst as 100*12kb*8 equals 10mbit.
204 !!BUGS
205
206
207 Not in the stock kernel yet.
208 !!SEE ALSO
209
210
211 tc(8)
212
213
214 HTB website:
215 http://luxik.cdi.cz/~devik/qos/htb/
216 !!AUTHOR
217
218
219 Martin Devera
220 ----
This page is a man page (or other imported legacy content). We are unable to automatically determine the license status of this page.