1 .TH CBQ 8 "16 December 2001" "iproute2" "Linux"
3 CBQ \- Class Based Queueing
53 .B ] [ bounded isolated ] [ split
62 Class Based Queueing is a classful qdisc that implements a rich
63 linksharing hierarchy of classes. It contains shaping elements as
64 well as prioritizing capabilities. Shaping is performed using link
65 idle time calculations based on the timing of dequeue events and
66 underlying link bandwidth.
69 When shaping a 10mbit/s connection to 1mbit/s, the link will
70 be idle 90% of the time. If it isn't, it needs to be throttled so that it
71 IS idle 90% of the time.
73 During operations, the effective idletime is measured using an
74 exponential weighted moving average (EWMA), which considers recent
75 packets to be exponentially more important than past ones. The Unix
76 loadaverage is calculated in the same way.
78 The calculated idle time is subtracted from the EWMA measured one,
79 the resulting number is called 'avgidle'. A perfectly loaded link has
80 an avgidle of zero: packets arrive exactly at the calculated
83 An overloaded link has a negative avgidle and if it gets too negative,
84 CBQ throttles and is then 'overlimit'.
86 Conversely, an idle link might amass a huge avgidle, which would then
87 allow infinite bandwidths after a few hours of silence. To prevent
88 this, avgidle is capped at
91 If overlimit, in theory, the CBQ could throttle itself for exactly the
92 amount of time that was calculated to pass between packets, and then
93 pass one packet, and throttle again. Due to timer resolution constraints,
94 this may not be feasible, see the
99 Within the one CBQ instance many classes may exist. Each of these classes
100 contains another qdisc, by default
103 When enqueueing a packet, CBQ starts at the root and uses various methods to
104 determine which class should receive the data.
106 In the absence of uncommon configuration options, the process is rather easy.
107 At each node we look for an instruction, and then go to the class the
108 instruction refers us to. If the class found is a barren leaf-node (without
109 children), we enqueue the packet there. If it is not yet a leaf node, we do
110 the whole thing over again starting from that node.
112 The following actions are performed, in order at each node we visit, until one
113 sends us to another node, or terminates the process.
116 Consult filters attached to the class. If sent to a leafnode, we are done.
120 Consult the defmap for the priority assigned to this packet, which depends
121 on the TOS bits. Check if the referral is leafless, otherwise restart.
124 Ask the defmap for instructions for the 'best effort' priority. Check the
125 answer for leafness, otherwise restart.
128 If none of the above returned with an instruction, enqueue at this node.
130 This algorithm makes sure that a packet always ends up somewhere, even while
131 you are busy building your configuration.
133 For more details, see
134 .BR tc-cbq-details(8).
136 .SH LINK SHARING ALGORITHM
137 When dequeuing for sending to the network device, CBQ decides which of its
138 classes will be allowed to send. It does so with a Weighted Round Robin process
139 in which each class with packets gets a chance to send in turn. The WRR process
140 starts by asking the highest priority classes (lowest numerically -
141 highest semantically) for packets, and will continue to do so until they
142 have no more data to offer, in which case the process repeats for lower
145 Classes by default borrow bandwidth from their siblings. A class can be
146 prevented from doing so by declaring it 'bounded'. A class can also indicate
147 its unwillingness to lend out bandwidth by being 'isolated'.
150 The root of a CBQ qdisc class tree has the following parameters:
153 parent major:minor | root
154 This mandatory parameter determines the place of the CBQ instance, either at the
156 of an interface or within an existing class.
159 Like all other qdiscs, the CBQ can be assigned a handle. Should consist only
160 of a major number, followed by a colon. Optional, but very useful if classes
161 will be generated within this qdisc.
164 This allotment is the 'chunkiness' of link sharing and is used for determining packet
165 transmission time tables. The qdisc allot differs slightly from the class allot discussed
166 below. Optional. Defaults to a reasonable value, related to avpkt.
169 The average size of a packet is needed for calculating maxidle, and is also used
170 for making sure 'allot' has a safe value. Mandatory.
173 To determine the idle time, CBQ must know the bandwidth of your underlying
174 physical interface, or parent qdisc. This is a vital parameter, more about it
178 The cell size determines he granularity of packet transmission time calculations. Has a sensible default.
181 A zero sized packet may still take time to transmit. This value is the lower
182 cap for packet transmission time calculations - packets smaller than this value
183 are still deemed to have this size. Defaults to zero.
186 When CBQ needs to measure the average idle time, it does so using an
187 Exponentially Weighted Moving Average which smoothes out measurements into
188 a moving average. The EWMA LOG determines how much smoothing occurs. Lower
189 values imply greater sensitivity. Must be between 0 and 31. Defaults
192 A CBQ qdisc does not shape out of its own accord. It only needs to know certain
193 parameters about the underlying link. Actual shaping is done in classes.
196 Classes have a host of parameters to configure their operation.
200 Place of this class within the hierarchy. If attached directly to a qdisc
201 and not to another class, minor can be omitted. Mandatory.
204 Like qdiscs, classes can be named. The major number must be equal to the
205 major number of the qdisc to which it belongs. Optional, but needed if this
206 class is going to have children.
209 When dequeuing to the interface, classes are tried for traffic in a
210 round-robin fashion. Classes with a higher configured qdisc will generally
211 have more traffic to offer during each round, so it makes sense to allow
212 it to dequeue more traffic. All weights under a class are normalized, so
213 only the ratios matter. Defaults to the configured rate, unless the priority
214 of this class is maximal, in which case it is set to 1.
217 Allot specifies how many bytes a qdisc can dequeue
218 during each round of the process. This parameter is weighted using the
219 renormalized class weight described above. Silently capped at a minimum of
220 3/2 avpkt. Mandatory.
224 In the round-robin process, classes with the lowest priority field are tried
225 for packets first. Mandatory.
229 See the QDISC section.
233 Maximum rate this class and all its children combined can send at. Mandatory.
237 This is different from the bandwidth specified when creating a CBQ disc! Only
238 used to determine maxidle and offtime, which are only calculated when
239 specifying maxburst or minburst. Mandatory if specifying maxburst or minburst.
243 This number of packets is used to calculate maxidle so that when
244 avgidle is at maxidle, this number of average packets can be burst
245 before avgidle drops to 0. Set it higher to be more tolerant of
246 bursts. You can't set maxidle directly, only via this parameter.
250 As mentioned before, CBQ needs to throttle in case of
251 overlimit. The ideal solution is to do so for exactly the calculated
252 idle time, and pass 1 packet. However, Unix kernels generally have a
253 hard time scheduling events shorter than 10ms, so it is better to
254 throttle for a longer period, and then pass minburst packets in one
255 go, and then sleep minburst times longer.
257 The time to wait is called the offtime. Higher values of minburst lead
258 to more accurate shaping in the long term, but to bigger bursts at
259 millisecond timescales. Optional.
263 If avgidle is below 0, we are overlimits and need to wait until
264 avgidle will be big enough to send one packet. To prevent a sudden
265 burst from shutting down the link for a prolonged period of time,
266 avgidle is reset to minidle if it gets too low.
268 Minidle is specified in negative microseconds, so 10 means that
269 avgidle is capped at -10us. Optional.
273 Signifies that this class will not borrow bandwidth from its siblings.
276 Means that this class will not borrow bandwidth to its siblings
279 split major:minor & defmap bitmap[/bitmap]
280 If consulting filters attached to a class did not give a verdict,
281 CBQ can also classify based on the packet's priority. There are 16
282 priorities available, numbered from 0 to 15.
284 The defmap specifies which priorities this class wants to receive,
285 specified as a bitmap. The Least Significant Bit corresponds to priority
288 parameter tells CBQ at which class the decision must be made, which should
289 be a (grand)parent of the class you are adding.
291 As an example, 'tc class add ... classid 10:1 cbq .. split 10:0 defmap c0'
292 configures class 10:0 to send packets with priorities 6 and 7 to 10:1.
294 The complimentary configuration would then
295 be: 'tc class add ... classid 10:2 cbq ... split 10:0 defmap 3f'
296 Which would send all packets 0, 1, 2, 3, 4 and 5 to 10:1.
298 estimator interval timeconstant
299 CBQ can measure how much bandwidth each class is using, which tc filters
300 can use to classify packets with. In order to determine the bandwidth
301 it uses a very simple estimator that measures once every
303 microseconds how much traffic has passed. This again is a EWMA, for which
304 the time constant can be specified, also in microseconds. The
306 corresponds to the sluggishness of the measurement or, conversely, to the
307 sensitivity of the average to short bursts. Higher values mean less
311 The actual bandwidth of the underlying link may not be known, for example
312 in the case of PPoE or PPTP connections which in fact may send over a
313 pipe, instead of over a physical device. CBQ is quite resilient to major
314 errors in the configured bandwidth, probably a the cost of coarser shaping.
316 Default kernels rely on coarse timing information for making decisions. These
317 may make shaping precise in the long term, but inaccurate on second long scales.
320 .BR tc-cbq-details(8)
321 for hints on how to improve this.
326 Sally Floyd and Van Jacobson, "Link-sharing and Resource
327 Management Models for Packet Networks",
328 IEEE/ACM Transactions on Networking, Vol.3, No.4, 1995
332 Sally Floyd, "Notes on CBQ and Guaranteed Service", 1995
336 Sally Floyd, "Notes on Class-Based Queueing: Setting
341 Sally Floyd and Michael Speer, "Experimental Results
342 for Class-Based Queueing", 1998, not published.
350 Alexey N. Kuznetsov, <kuznet@ms2.inr.ac.ru>. This manpage maintained by
351 bert hubert <ahu@ds9a.nl>
git://git.onelab.eu / iproute2.git / blob