CN1297098C - System flow controlling method - Google Patents
System flow controlling method Download PDFInfo
- Publication number
- CN1297098C CN1297098C CNB031251102A CN03125110A CN1297098C CN 1297098 C CN1297098 C CN 1297098C CN B031251102 A CNB031251102 A CN B031251102A CN 03125110 A CN03125110 A CN 03125110A CN 1297098 C CN1297098 C CN 1297098C
- Authority
- CN
- China
- Prior art keywords
- message
- flow control
- idle queues
- drop probability
- probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The technology belongs to the technical field of data communication service quality, and relates to a system flow controlling method. The technology comprises processing which is carried out for initializing the system and processing carried out for a message arriving at a device at time t, wherein the processing carried out for initializing the system comprises that discretized probability tables are calculated according to different minimum discarding idle queue thresholds, maximum discarding idle queue thresholds and maximum discarding probability, and the value of the maximum discarding probability can be adjusted according to the type of the message; the processing carried out for a message arriving at a device at time t comprises that the rate at which messages arrive at the device in a time interval between a previous time t'and the time t, the average queue length of the idle queue at the time t and the integrated average idle queue length are calculated, the table entry index values of a discarding probability table (DPT table) are calculated, the probability P(t) of transmitting a message in the time period is obtained and is compared with a random number P which is generated currently, and whether the message is discarded or forwarded or not is determined. The present invention can distinguish message type to carry out different flow control curve configuration, and a more optimized flow control effect is achieved.
Description
Technical field
Present technique belongs to data communication QoS (Quality of Service, service quality) technical field, specially refers to the realization technology of the flow system flow control among the Diffserv (Differentiated Service, differential service).
Background technology
In current data network, along with the network evolution trend of unification of three nets (voice, video, data) is further deepened, and the establishment of the de facto standard of TCP/IP network technology, on the original data network of doing one's best, provide service quality more and more to become operator to the active demand of IP network equipment.IP network equipment is wanted to distinguish voice, video, data service, and at the characteristics of these different business, provides different transfer qualities just to become the important subject of IP QoS.Because Internet resources are limited, therefore in the network planning and operation management, the management and use of Internet resources become the most important thing of QoS research topic again simultaneously.Internet resources (buffer memory, bandwidth) are limited, and data network service be at random with burst, so that network inevitably can occur is congested.The congestion management of IPQoS (Congestion Avoidance and congested control) just requires the network equipment that a kind of mechanism can be provided, under this mechanism control, can there be higher throughput (resource utilization) in system, and system can differentiated service under because of the situation of resource overload classification, message to different priority levels carries out different actions, thereby guarantees that other professional forwarding of high priority is unaffected.WRED (Weighted Random Early Discard: Weighted random early abandons) method is exactly a kind of congestion management means commonly used.
WRED (Weighted Random Early Discard: Weighted random early abandons) abandons the curve packet loss according to linearity: in the control plane system regularly, carry out the weighting statistics of idling-resource (generally being the buffer queue resource) such as every t cycle in the time interval:
queue(t)=queue(t)+(1-w)queue(t′)
Wherein queue (t) represents t idle queues length constantly, and queue (t ') represents the idle queues length of previous moment, and W is the weighted factor between [0,1].
Judge the Congestion Level SPCC of this system according to this weighting statistical value time period, and determine the drop probability transmitted in this time period T just to refresh DPT (Discard PossibilityTable: the drop probability table) according to this Congestion Level SPCC.The Flow Control curve that the WRED method derives is for as shown in Figure 1, and among the figure, ordinate is a drop probability, span [0,1], and the maximum drop probability value of the drop probability oblique line on it is maxp; Abscissa is an idle queues length, on it to minimum idle queues thresholding (minq) value and maximum idle quene threshold (maxq) value that should oblique line.
At Forwarding plane, be to transmit or abandon according to current drop probability decision to the message that reaches.
The effect that the General System realization is regulated congested control by the fine setting of time interval t and weighting introduction w, simultaneously at Diffserv (Differentiated Service, differential service) in the QoS model, the Flow Control curve difference of general different classs of service, just under identical formation Congestion Level SPCC, little to other packet drop probability of high priority, big to other packet drop probability of low priority, guarantee different service class with this.
Realized basic WRED method in the existing equipment mostly, i.e. 0,1 model shown in Fig. 1, during less than lower limit, 100% abandons in idle queues length, 100% transmits during greater than higher limit in idle queues.This model by the configuration of w and t and idle queues lower limit higher limit, can well be worked on low-speed device.But for high-speed network appliance, such as GSR (Gigabit Switch Route gigabit route-exchanging device), TSR (Trilbit Switch RouteT bit exchange routing device), because interface link speed improves greatly, but buffer memory limited amount, this 0/1 model and regulate by w/t and can not well carry out the congestion management of system, because when bigger burst sustained overload, use this model, system cache will be just depleted in very short time, and system will reach 100% very soon and abandon a little; Because exchange and forwarding rate improve, and buffer memory is by can very fast emptying, system reaches 100% forwarding point again very soon again, so repeatedly, transmit often acutely shake between 0 and 1 of probability, thereby cause the formation utilance that bigger shake is arranged, the accumulation throughput of link is also lower.
Simultaneously at Diffserv (Differentiated Service, differential service) in the QoS model, though can distinguish different classs of service, design different Flow Control curves, but not all type of message is responsive to abandoning all, transport layer protocol to TCP (Transport Control Protocol) the built-in Flow Control of this class, can make a response to packet loss, and then reach the effect of back-pressure, connectionless transport layer protocols such as UDP are not reacted packet loss, and therefore above-mentioned model scope has certain limitation to type of message.
Summary of the invention
The objective of the invention is has certain limitation for overcoming prior art to type of message, flow system flow is controlled unfavorable weak point, a kind of flow system flow control method is proposed, can be in high-end devices under the discharge model, throughput and the fluid control effect of raising system when overload; Simultaneously can distinguish type of message and carry out different Flow Control curved configuration, reach more optimal flow control effect.
A kind of flow system flow control method that the present invention proposes, the processing when comprising system initialization reaches processing two parts of carving at a time the message that reaches equipment; Wherein,
Treatment step during described system initialization is: according to the flow control parameters of user configured different priority levels and different type of messages, the drop probability table that carries out discretization calculates;
The processing of carving at a time the message that reaches equipment may further comprise the steps:
Calculate final comprehensive average idle queues length;
Search described drop probability table according to described final comprehensive average idle queues length and type of message, obtain the transmission probability of this this message of time interval;
The transmission probability of described message and the random number of current generation are compared, if described transmission probability less than described random number, then abandons this message; Otherwise transmit this message.
The flow control parameters of described dissimilar messages can comprise: minimum abandons the idle queues thresholding, maximum abandons idle queues thresholding, maximum drop probability.
Treatment step during described system initialization also can comprise: adjust the flow control parameters of different priority levels and dissimilar messages, and the discretization progression of drop probability table.
The final comprehensive average idle queues length of described calculating can may further comprise the steps:
Calculate the speed that at interval interior sometime message arrives;
Early abandon the average queue length that formula calculates described moment idle queues by Weighted random;
Average queue length, message arrival rate, the regulatory factor that comprehensive average idle queues length of the described moment is described idle queues is long-pending;
Described final comprehensive average idle queues length can be the poor of the average queue length of idle queues and comprehensive average idle queues length.
The speed that the at interval interior sometime message of described calculating arrives can may further comprise the steps:
The counting that reaches with message in the described time interval is divided by the described time interval;
Turn to several intervals according to the capacity of system with described result of calculation value is discrete then;
Described discretization interval is mapped in [0,1] interval.
The three-dimensional array that described drop probability table can be made of message grade, type, final comprehensive idle queues length, the step that described discretization drop probability table calculates can be:
Abandon by the flow control parameters minimum of message that idle queues thresholding, maximum abandon the idle queues thresholding, maximum drop probability obtains corresponding drop probability curve;
The minimum of drop probability curve is abandoned the idle queues thresholding abandon the idle queues thresholding to maximum and divide by discretization progression, the mean value of corresponding drop probability promptly is the drop probability of idle queues length message in this is interval the time between each dividing regions.
Characteristics of the present invention:
(1) the present invention reaches the message of stream speed (Packet Arrival Rate) and carries out the calculating of drop probability as a governing factor.Introduce this factor, can originally determine on the basis of drop probability according to average queue according to the input rate of stream, the dynamic drop probability of transmitting of regulating reaches better flow system flow control effect.
(2) existing WRED method is when average queue depth reaches certain thresholding, will carry out 100% abandon.And for the high-end devices of surface speed forwarding, when system not congested (surface speed forwarding), often only need a spot of system cache, and in case exceed the surface speed forwarding capacity, system will the congested rapidly minimum threshold that reaches system cache, the formation utilance that shows of system has bigger shake like this, and the accumulation throughput of link is also lower.For this reason, the present invention is directed to the message of different Flow Control ranks and discard level, the maximum drop probability during to its network congestion is optimized calculating, and make it<1, can make system that throughput is preferably arranged when overload.
(3) the present invention carries out different flow control parameters (grade, classification and granularity rank) configuration to four types message [TCP SYN message, TCP message, UDP message, other IP message], can obtain different Flow Control curves, to reach better fluid control effect.
(4) the inventive method is through experimental simulation and actual motion, all shows in the network acute variation or has at network under the situation of bigger overload, and the system that adopts this method is than adopting existent method that higher formation utilance and system's aggregated throughput are arranged.
Description of drawings
Fig. 1 early abandons the Flow Control curve chart for existing Weighted random.
Fig. 2 is for early abandoning the discretization principle schematic of Flow Control curve for Weighted random.
Fig. 3 is that the Weighted random after the optimization of employing the inventive method early abandons the Flow Control curve chart.
Embodiment
A kind of flow system flow control method embodiment that the present invention proposes is described in detail as follows:
Processing when present embodiment comprises system initialization and at moment t processing two parts to the message that reaches equipment; Wherein,
Treatment step during system initialization comprises:
(1) according to the different priority levels of user's configuration (or system default), the flow control parameters of dissimilar message (TCP SYN, other IP, UDP, TCP): minimum abandons the idle queues thresholding, maximum abandons flow control parameters such as idle queues thresholding, maximum drop probability, and the DPT table that carries out discretization calculates; Because the drop probability curve of WRED is an oblique line (as shown in Figure 1), in digital system, need as required its discretization to be represented, principle is seen Fig. 2, the minimum threshold that is about to the idle queues of oblique line is divided into a plurality of intervals to max threshold, and the progression of discretization can be selected to be provided with according to system resource condition.With TCP SYN grade is that 1 message and discretization progression are 128 to be example, discretization is calculated as follows carries out: the minimum threshold with idle queues arrives max threshold interval [minq earlier, maxq] be divided into 128 equal portions, obtain the mean value of the little oblique line line of this section then on each minizone, the result is the drop probability of message when dropping in this interval when the idle queues value just.
(2) priority level of Adjustment System configuration according to actual needs and the numerical value of type of message in actual applications, and the granularity progression of the discretization of drop probability table, the three-dimensional array P[class that present embodiment is made up of message grade class, type type, idle queues length queue_depth] [type] [queue_depth] formation DPT table, wherein the class grade is exactly the message forwarding grade, as an input, the loss priority of its expression message, the high more message of priority level is not easy to abandon more, and General System all is designed to 4 or 6 priority; The type type of message is exactly that the protocol type that carries of user data generally has TCP SYN, TCP, UDP, four types of Other IP as mentioned above; The value of last idle queues length queue_depth and the discretization progression of idle queues length are functional relation to every kind of CLASS, the WRED parameter of TYPE, such as minimum abandon idle queues thresholding MINTH, maximum abandons idle queues thresholding MAXTH, maximum drop probability MAXP and is disposed by the user, system can calculate the value of all three-dimensional array under a proportional relationship.。
(3) also can adjust the value of its maximum drop probability according to dissimilar messages, in the present embodiment, the EF the highest (expedite forwarding) message to priority level, maximum drop probability is adjusted into 0.1 to 0.3; For the lower message of priority level, maximum drop probability is adjusted into 0.7 to 0.9.
At the moment t message processing that reaches equipment be may further comprise the steps:
(1) calculate the speed O (t) that previous moment t ' arrives to interior message of the t time interval, this value can reach counting count (t)-count (t ') by message and obtain divided by the time interval, i.e. O (t)=f (count (t)-count (t ')/(t-t ')); General capacity according to system reaches speed with message and carries out discretization, such as if the possible message arrival rate of system be 0bps (bit per sencond bits per second)~3Gbps (1G=10^9), can it be 8 parts with this interval division, [0,375M] be interval 0, and the like ... [2.625M, 3G] is interval 7.O (t) gets the function f () in this discretization interval then, and function model is 2^ (x-A) in the present embodiment, and x is interval index, and A is that the interval number of discretization also is a discretization progression, as above routine desirable A=8.
(2) calculate this average queue length queue (t) of idle queues constantly, this value is calculated by formula queue (t)=queue (t)+(1-w) queue (t ');
(3) calculate the constantly comprehensive average idle queues length queue (t ') of t with and the discretization interval of value.Method is as follows: queue (t)=queue (t) * O (t) * M, and wherein M is a regulatory factor, and the effect of regulatory factor M mainly is that the constraint message reaches the influence that speed O (t) calculates final average queue length, and the discretization interval is [0,1];
(4) calculate final comprehensive average idle queues length queue_depth, (t)-queue ' (t) for queue_depth=queue, wherein queue (t) and queue ' (t) are calculated by step (2), (3) respectively, obtain the DPT table index value of the correspondence of idle queues then, the index value computational methods of present embodiment illustrate: if the idle queues length of system is the individual buffer unit of 128K (1k=10^3), it discrete is turned to 128 intervals, if calculate queue_depth=4k, then the correspondence of drop probability is that DPT table array indexing value is 4;
(5) according to the grade class of message, type of message type and (4) calculate comprehensive average idle queues discrete length value queue_depth as input, remove to search the DPT table, be array P[class described above] [type] [queue_depth], thereby obtain the transmission probability P (t) of this message, the relation of it and drop probability is: transmission probability=1-drop probability;
(6) transmission probability of the message that (5) are obtained and the random number P of current generation compare, if p (t)<p then abandons this message; If p (t)>=p then transmits this message.
The optimization Flow Control curve of the different messages that the method for present embodiment obtains as shown in Figure 3.
Wherein: curve 4 is former WRED curve; All the other from left to right (curve 1,2,3 and 5) be followed successively by: the improvement WRED Flow Control curve of TCP SYN, other IP, UDP, TCP.Ordinate is a drop probability, span [0,1], the different value of the maximum drop probability (maxp) of each curve of correspondence on it; Abscissa is an idle queues length, minimum idle queues thresholding (minq) value of each curve of correspondence on it and maximum idle quene threshold (maxq) value.
Claims (9)
1, a kind of flow system flow control method is characterized in that, comprising:
Steps A, according to the flow control parameters of user configured different priority levels and different type of messages, calculate corresponding drop probability curve; The described drop probability curve of discretization, the drop probability table of the discretization that to generate with message grade, type of message and final comprehensive idle queues length be variable;
The long-pending comprehensively average idle queues length that obtains of the speed that the at interval interior sometime message of step B, calculating arrives, the average queue length of this moment idle queues and regulatory factor, and by the average queue length of idle queues and the final comprehensive on average idle queues length of difference calculating of comprehensive average idle queues length;
Step C, search described drop probability table, obtain the transmission probability of this this message of time interval according to described final comprehensive average idle queues length and type of message;
The transmission probability of described message and the random number of current generation are compared, if described transmission probability less than described random number, then abandons this message; Otherwise transmit this message.
2, the system as claimed in claim 1 flow control methods is characterized in that, the flow control parameters of described dissimilar messages comprises: minimum abandons the idle queues thresholding, maximum abandons idle queues thresholding, maximum drop probability.
3, flow system flow control method as claimed in claim 1 or 2 is characterized in that, described steps A also comprises: adjust different priority levels, the flow control parameters of dissimilar messages and the discretization progression of drop probability table.
4, flow system flow control method as claimed in claim 3 is characterized in that, the flow control parameters of the dissimilar messages of described adjustment comprises the adjustment maximum drop probability: the message the highest to priority level, maximum drop probability are adjusted into 0.1 to 0.3; For the lower message of priority level, maximum drop probability is adjusted into 0.7 to 0.9.
5, the system as claimed in claim 1 flow control methods is characterized in that, described step B further may further comprise the steps:
Calculate the speed that at interval interior sometime message arrives;
Early abandon the average queue length that formula calculates described moment idle queues by Weighted random.
6, flow system flow control method as claimed in claim 5 is characterized in that, the speed that the at interval interior sometime message of described calculating arrives may further comprise the steps:
The counting that reaches with message in the described time interval is divided by the described time interval;
Turn to several intervals according to the capacity of system with described result of calculation value is discrete then;
Described discretization interval is mapped in [0,1] interval.
7, flow system flow control method as claimed in claim 3 is characterized in that, described steps A specifically comprises:
Abandon by the flow control parameters minimum of message that idle queues thresholding, maximum abandon the idle queues thresholding, maximum drop probability obtains corresponding drop probability curve;
The minimum of drop probability curve is abandoned the idle queues thresholding abandon the idle queues thresholding to maximum and divide by discretization progression, the mean value of corresponding drop probability promptly is the drop probability of idle queues length message in this is interval the time between each dividing regions.
8, the system as claimed in claim 1 flow control methods is characterized in that: described transmission probability deducts drop probability by 1 and obtains.
9, the system as claimed in claim 1 flow control methods, it is characterized in that described dissimilar messages comprise: sync message, other IP agreement bearing message, User Datagram Protocol bearing message, transmission control protocol bearing message are set up in the connection of transmission control protocol.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNB031251102A CN1297098C (en) | 2003-05-09 | 2003-05-09 | System flow controlling method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNB031251102A CN1297098C (en) | 2003-05-09 | 2003-05-09 | System flow controlling method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN1549508A CN1549508A (en) | 2004-11-24 |
CN1297098C true CN1297098C (en) | 2007-01-24 |
Family
ID=34321832
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNB031251102A Expired - Fee Related CN1297098C (en) | 2003-05-09 | 2003-05-09 | System flow controlling method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1297098C (en) |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100450081C (en) * | 2005-06-10 | 2009-01-07 | 华为技术有限公司 | Method and system for carrying out flow-control |
CN101277253B (en) * | 2007-03-30 | 2011-06-08 | 迈普通信技术股份有限公司 | Method for controlling packet channel of master-slave mode multiprocessor system |
CN101350765B (en) * | 2007-07-20 | 2012-07-04 | 中国科学院声学研究所 | Network flow detection method |
US8325687B2 (en) * | 2008-05-09 | 2012-12-04 | Qualcomm Incorporated | Techniques for enhanced backhaul flow control |
CN102571973B (en) * | 2012-02-02 | 2015-01-07 | 瑞斯康达科技发展股份有限公司 | Network control method and device |
CN103379146B (en) * | 2012-04-19 | 2016-12-28 | 深圳市云帆世纪科技有限公司 | The method for limiting speed of a kind of data transmission and device |
CN103929372B (en) * | 2013-01-11 | 2017-10-10 | 华为技术有限公司 | Active queue management method and apparatus |
CN103733581B (en) * | 2013-06-19 | 2017-01-25 | 华为技术有限公司 | Message processing method and base station |
CN105791140A (en) * | 2014-12-19 | 2016-07-20 | 中兴通讯股份有限公司 | Message sending processing method and apparatus and network equipment |
CN104580008B (en) * | 2015-01-07 | 2019-02-22 | 盛科网络(苏州)有限公司 | The method and device of more queue random drop message precision is improved based on hardware |
US9864716B2 (en) * | 2015-05-20 | 2018-01-09 | International Business Machines Corporation | Receiving buffer credits by a plurality of channels of one or more host computational devices for transmitting data to a control unit |
CN106789722A (en) * | 2015-11-23 | 2017-05-31 | 深圳市中兴微电子技术有限公司 | A kind of method for controlling network congestion and device |
CN106437520B (en) * | 2016-10-17 | 2018-07-31 | 华北理工大学 | A kind of broken coal hole punched device of automatic screw type high pressure water |
CN111385667A (en) * | 2018-12-27 | 2020-07-07 | 中国电信股份有限公司 | Video data processing method, device and computer readable storage medium |
CN115277591B (en) * | 2022-08-04 | 2023-11-07 | 深圳云豹智能有限公司 | Message processing circuit, method, chip and computer equipment |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002007381A2 (en) * | 2000-07-18 | 2002-01-24 | Alcatel | Domain based congestion management |
CN1380771A (en) * | 2001-04-17 | 2002-11-20 | 陈常嘉 | Method for implementing hierarchical direction to randomly and early discard queue management mechanism and circuit |
-
2003
- 2003-05-09 CN CNB031251102A patent/CN1297098C/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2002007381A2 (en) * | 2000-07-18 | 2002-01-24 | Alcatel | Domain based congestion management |
CN1380771A (en) * | 2001-04-17 | 2002-11-20 | 陈常嘉 | Method for implementing hierarchical direction to randomly and early discard queue management mechanism and circuit |
Also Published As
Publication number | Publication date |
---|---|
CN1549508A (en) | 2004-11-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN1297098C (en) | System flow controlling method | |
KR101075724B1 (en) | Apparatus and method for limiting packet transmission rate in a communication system | |
CN109120544B (en) | Transmission control method based on host end flow scheduling in data center network | |
US7447152B2 (en) | Controlling traffic congestion | |
CN113767600A (en) | System and method for adaptive routing in the presence of a continuous flow | |
US20070280277A1 (en) | Method and system for adaptive queue and buffer control based on monitoring in a packet network switch | |
US20040081090A1 (en) | Congestion controller for ethernet switch | |
CN1716914A (en) | Fair WRED for TCP and UDP traffic mix | |
JP2003504900A (en) | A method for improving system performance in data networks by queue management based on incoming rate monitoring | |
KR20050088874A (en) | Congestion avoidance method for video service bandwidth | |
CN1801778A (en) | Multidimensional queue dispatching and managing system for network data stream | |
CN112995048B (en) | Blocking control and scheduling fusion method of data center network and terminal equipment | |
US9350659B1 (en) | Congestion avoidance for network traffic | |
US8203956B1 (en) | Method and apparatus providing a precedence drop quality of service (PDQoS) | |
TWI763261B (en) | Data flow classification device | |
Xue et al. | Afcd: An approximated-fair and controlled-delay queuing for high speed networks | |
US20130117553A1 (en) | Method and Apparatus for Increasing the Output of a Cryptographic System | |
Zhang et al. | An improved adaptive active queue management algorithm based on nonlinear smoothing | |
JP2020022023A (en) | Packet transfer device, method, and program | |
Rao et al. | Analysis of sfqCoDel for active queue management | |
WO2019029220A1 (en) | Network device | |
Jindal et al. | Congestion Control Framework in Ad-Hoc Wireless using Neural Networks in QoS | |
Kozačinski et al. | Configuration of quality of service parameters in communication networks | |
Jiang et al. | Self adjustable CHOKe: an active queue management algorithm for congestion control and fair bandwidth allocation | |
CN114629847A (en) | Coupling multi-flow TCP congestion control method based on available bandwidth allocation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20070124 Termination date: 20170509 |
|
CF01 | Termination of patent right due to non-payment of annual fee |