Queue management method based on dynamic Hurst parameter and queue growth rate
By combining dynamic Hurst parameters and queue growth rate, the packet loss strategy is adaptively adjusted, and the problem of congestion judgment in the existing technology is solved, and stable management and performance improvement of network traffic is achieved.
Patent Information
- Application Number
- CN202411179270.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The existing active queue management algorithm only judges network congestion based on a single angle, ignoring the impact of self-similarity of network traffic on queue performance, resulting in unstable queue length and increasing cache overflow rate and delay.
The queue management method based on dynamic Hurst parameters and queue growth rate is adopted, and the Hurst parameters are calculated in real time through R/S analysis method, combined with the trend of changing the average queue length, and adaptively adjust the packet loss strategy to achieve reliable guarantees for network traffic service quality.
In a complex dynamic network environment, the stability, packet loss rate and throughput of the average queue length are significantly improved, and network transmission performance is improved.
Smart Images

Figure CN118694723B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of information engineering, and in particular to a queue management method based on dynamic Hurst parameters and queue growth rate. Background Art
[0002] Network traffic has the characteristics of many services and large traffic, which makes the buffer of the forwarding router easy to be filled, resulting in network congestion. The queue management mechanism is an effective method to achieve network congestion control. When the network is congested, the routing node adjusts the occupancy of the buffer by dropping packets, and reduces the length of the cache queue by packet discarding strategy to reduce the probability of cache overflow. Different queue management mechanisms directly affect the efficiency of congestion control of network nodes and the service quality of the network. The queue management mechanism can be divided into two categories: passive queue management mechanism (Passive Queue Management, PQM) and active queue management mechanism (Active Queue Management, AQM), the latter of which is widely used. According to the different congestion detection methods, active queue management algorithms can be divided into methods based on queue length: such as adaptive RED, weighted queue dynamic AQM, etc.; rate-based methods, such as adaptive virtual queue AVQ, etc.; delay-based methods, such as adaptive control delay ACoDel-TIT algorithm with target and interval adjustment, AQM based on deep Q network, etc.; methods based on the combination of queue length and rate, such as AC-RED with average queue length and rate, etc.; methods based on the combination of queue length and delay, such as fuzzy logic RED, etc.
[0003] However, most active queue management algorithms only judge the degree of network congestion from a single perspective, ignoring the impact of the characteristics of network traffic on queue performance. Self-similarity is an important characteristic of network traffic. It manifests itself as burstiness, resulting in unstable queue length and increasing the probability of congestion. The higher the degree of self-similarity, the greater the change in queue length, and the greater the impact on queue performance, resulting in a high buffer overflow rate and long delay. At present, most active queue management algorithms based on traffic self-similarity simply introduce the Hurst parameter directly into the calculation of packet loss probability, and do not take into account the average queue length change trend directly affected by the change in data transmission rate. It cannot stabilize the queue length well and thus improve the transmission performance of the network. Summary of the invention
[0004] The purpose of the present invention is to provide a queue management method based on dynamic Hurst parameters and queue growth rate. The method judges the traffic situation from two perspectives, not only taking into account the impact of network traffic self-similarity on queue performance, but also taking into account the change in average queue growth rate due to link variable rate characteristics. On this basis, the Hurst parameter and the average queue growth rate trend are modeled, and a queue management (HC-AQM) method based on dynamic Hurst parameters and queue growth rate is proposed. The R / S analysis method is used to calculate the Hurst parameter in real time according to the average queue length. Combined with the average queue length change trend, an active queue control algorithm that adaptively adjusts the packet loss strategy is used to achieve reliable protection of network traffic service quality.
[0005] To achieve the above functions, the present invention designs a queue management method based on dynamic Hurst parameters and queue growth rate. The router buffer has multiple queues composed of data packets. The following steps S1 to S7 are executed to complete the dynamic management of the queue:
[0006] Step S1: monitor the queue status in the target router buffer in real time and wait for new data packets to arrive;
[0007] Step S2: after the new data packet arrives at the target router buffer, it is determined whether all the data in the data packet needs to be discarded according to the queue status of the target router buffer at the current moment, including the average queue length and the average queue growth rate at the current moment. If yes, it returns to step S1, otherwise it jumps to step S3;
[0008] Step S3: Put the new data packet into the target router buffer, and calculate the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment, including the average queue length and average queue growth rate at the next moment, according to the queue state of the target router buffer at the current moment;
[0009] Step S4: Calculate the drop probability based on the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment;
[0010] Step S5: according to the discard probability, the data in the new data packet is placed into the queue of the target router buffer, and the remaining data in the new data packet is discarded from the target router buffer;
[0011] Step S6: Update the queue state of the target router buffer and calculate the Hurst parameter and queue state of the queue at the next moment;
[0012] Step S7: Wait for a new data packet to arrive, and jump to step S2.
[0013] Beneficial effects: Compared with the prior art, the advantages of the present invention include:
[0014] 1. The present invention takes into account the impact of network traffic self-similarity on queue performance and the change in average queue length caused by link variable rate characteristics, and combines the average queue length with the average queue growth rate. On this basis, the Hurst parameter and the average queue length change trend are modeled, and the R / S analysis method is used to calculate the Hurst parameter in real time according to the average queue length. Combined with the average queue length change trend, the active queue control algorithm adaptively adjusts the packet loss strategy to achieve reliable protection of network traffic service quality. In a complex and dynamic network environment, the method of the present invention significantly improves the stability of the average queue length, packet loss rate, and throughput.
[0015] 2. The process of the present invention is simple, operability is strong, and application range is wide. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 is a flow chart of a queue management method based on dynamic Hurst parameters and queue growth rate provided in an embodiment of the present invention;
[0017] Figure 2 is a schematic diagram of a module provided according to an embodiment of the present invention;
[0018] Figure 3 is an experimental topology diagram provided according to an embodiment of the present invention;
[0019] Figure 4 is a queue length diagram of a RED algorithm under load one provided by an embodiment of the present invention;
[0020] Figure 5 is a queue length diagram of the URED algorithm under load one provided by an embodiment of the present invention;
[0021] Figure 6 is a queue length diagram of a TRED algorithm under load one provided by an embodiment of the present invention;
[0022] Figure 7 is a queue length diagram of the AgRED algorithm under load one provided by an embodiment of the present invention;
[0023] Figure 8 is a queue length diagram of the HC-AQM algorithm under load one provided by an embodiment of the present invention;
[0024] Fig. 9 is a queue length diagram of the RED algorithm under load conditions provided by an embodiment of the present invention;
[0025] Fig.10is a queue length diagram of the URED algorithm under load 2 provided by an embodiment of the present invention;
[0026] Fig.11 is a queue length diagram of the TRED algorithm under load 2 provided by an embodiment of the present invention;
[0027] Fig.12 is a queue length diagram of the AgRED algorithm under load 2 provided by an embodiment of the present invention;
[0028] Fig.13 is a queue length diagram of the HC-AQM algorithm under load 2 provided by an embodiment of the present invention;
[0029] Fig.14 is a comparison chart of throughputs of five algorithms provided according to an embodiment of the present invention under changes in the number of two sending sources;
[0030] Fig.15 is a comparison chart of packet loss rates of five algorithms provided by an embodiment of the present invention under two changes in the number of sending sources;
[0031] Fig.16 is a comparison diagram of the delays of five algorithms provided by an embodiment of the present invention under changes in the number of two sending sources;
[0032] Fig.17 It is a comparison chart of jitters of five algorithms provided according to an embodiment of the present invention under changes in the number of two sending sources. DETAILED DESCRIPTION
[0033] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.
[0034] The queue management method based on dynamic Hurst parameters and queue growth rate provided in the embodiment of the present invention comprises a plurality of queues composed of data packets in the router buffer. Figure 1 , execute the following steps S1 to S7 to complete the dynamic management of the queue:
[0035] Step S1: monitor the queue status in the target router buffer in real time and wait for new data packets to arrive;
[0036] Step S2: after the new data packet arrives at the target router buffer, it is determined whether all the data in the data packet needs to be discarded according to the queue status of the target router buffer at the current moment, including the average queue length and the average queue growth rate at the current moment. If yes, it returns to step S1, otherwise it jumps to step S3;
[0037] If the current moment t Average queue length The maximum threshold of the average queue length is exceeded. , then all the data in the data group will be discarded;
[0038] Step S3: Put the new data packet into the target router buffer, and calculate the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment, including the average queue length and average queue growth rate at the next moment, according to the queue state of the target router buffer at the current moment;
[0039] In step S3, the R / S analysis method is used to calculate the Hurst parameter of the queue at the current moment.
[0040] The method for calculating the queue state of the target router buffer at the next moment in step S3 is as follows:
[0041] The expression for calculating the average queue length at the next moment is as follows:
[0042] ;
[0043] In the formula, express t The average queue length at time +1, express t The average queue length at time, express t The instantaneous queue length at time +1, represents the weight coefficient;
[0044] The expression for calculating the average queue growth rate at the next moment is as follows:
[0045] ;
[0046] In the formula, and Respectively t +1, t The average queue growth rate at time When , the average queue length increases significantly; when When , the average queue length decreases significantly; when When , the average queue length is relatively stable; ε Represents a constant close to 0; express t The instantaneous queue length at time.
[0047] Step S4: Calculate the drop probability based on the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment;
[0048] If the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment meet one of the following six conditions, the corresponding drop probability is calculated, otherwise the execution jumps to step S7, where ε It can be adjusted adaptively according to the current queue status:
[0049] Determine whether the current network traffic state satisfies self-similarity. If it does not satisfy self-similarity, , at this time, the average queue length sequence is random and has no regularity; then judge the average queue growth rate, if it satisfies , the speed at which the average queue length reaches the maximum threshold slows down, and the probability that the router node is in an idle state increases; in order to better utilize the router buffer, a conservative packet loss strategy is adopted, and the corresponding drop probability is P1;
[0050] The discard probability P1 is as follows:
[0051] ;
[0052] In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length; H Indicates the Hurst parameter of the queue at the current moment. ave is the average queue length, express t The average queue growth rate at time +1, is the middle threshold of the average queue length, is the maximum drop probability.
[0053] like ,and At this time, the network traffic change trend is relatively gentle, the speed at which the average queue length reaches the maximum threshold is relatively stable, and the congestion probability of the router node increases slightly; in this case, the HC-AQM algorithm The packet loss probability function is constructed using a sine function curve within the range. In order to more reasonably reflect the changes in network traffic, the corresponding drop probability is P2;
[0054] The discard probability P2 is as follows:
[0055] ;
[0056] In the formula, is the maximum threshold of the average queue length, The minimum threshold for the average queue length.
[0057] like ,and , the network traffic changes greatly, the average queue length will quickly reach the maximum threshold, and the router node is likely to be congested. In this case, the HC-AQM algorithm will use a more aggressive packet loss strategy. The packet loss probability function is constructed using the tangent function curve within the range. When , the discard probability directly becomes 1, and the corresponding discard probability is P3;
[0058] The discard probability P3 is as follows:
[0059] ;
[0060] In the formula, is the maximum threshold of the average queue length, The minimum threshold for the average queue length.
[0061] If the current network traffic state satisfies self-similarity, that is, At this time, the average queue length sequence has a certain regularity, and the Hurst parameter is positively correlated with the average queue length. The Hurst parameter is combined with the average queue change rate and the corresponding packet loss strategy is adaptively adjusted; then the average queue growth rate is determined. If it satisfies , the average queue length is and In the range of , the quadratic function and the half-linear function are used to form the packet loss probability function. , appropriately increase The interval length makes the packet loss probability more conservative, and the corresponding drop probability is P4;
[0062] The discard probability P4 is as follows:
[0063] ;
[0064] In the formula, is the maximum threshold of the average queue length, The minimum threshold for the average queue length.
[0065] like ,and , the average queue length is In the range of , a one-third function is used to form a packet loss probability function, which can accurately handle network congestion, and the corresponding drop probability is P5;
[0066] The discard probability P5 is as follows:
[0067] ;
[0068] In the formula, is the maximum threshold of the average queue length, The minimum threshold for the average queue length.
[0069] like ,and , the average queue length is The packet loss probability function is constructed using a fifth function within the range. When , the discard probability directly becomes 1; at this time , as the Hurst parameter increases, it decreases appropriately The interval length makes the packet loss probability more aggressive and avoids congestion. The corresponding drop probability is P6;
[0070] The discard probability P6 is as follows:
[0071] ;
[0072] In the formula, is the maximum threshold of the average queue length, The minimum threshold for the average queue length.
[0073] Step S5: according to the discard probability, the data in the new data packet is placed into the queue of the target router buffer, and the remaining data in the new data packet is discarded from the target router buffer;
[0074] Step S6: Update the queue state of the target router buffer and calculate the Hurst parameter and queue state of the queue at the next moment;
[0075] Step S7: Wait for a new data packet to arrive, and jump to step S2.
[0076] The following is an application example of the present invention:
[0077] This embodiment implements a queue management method (HC-AQM) based on dynamic Hurst parameters and queue growth rate on NS2 (Network Simulator, Version 2) network simulation software, and tests its performance in detail. NS2 is a network simulation software widely used in academia, and the research results obtained through this platform are generally recognized by the academic community.
[0078] This embodiment uses five modules to implement the method proposed in this patent. Figure 2, including: an entry module, a calculation module, a discard module, an update module and a dequeue module. The main function of the entry module is to monitor the average queue length and average queue growth rate of the router buffer at the current moment, corresponding to step S2; the calculation module is one of the core modules of the embodiment, and its main function is to calculate the average queue length and average queue growth rate at the next moment according to the data of the entry module, and to build a queue state judgment model, corresponding to step S3; the main function of the discard module is to discard the data packets in the router buffer according to the queue state judgment criterion obtained in the above steps, corresponding to steps S4-S5; the update module is the second core module of this embodiment, and its main function is to update the average queue length and average queue growth rate of the router buffer at this time after the data packet is discarded, and calculate the queue state at the next moment, corresponding to step S6; the dequeue module corresponds to step S7.
[0079] This embodiment adopts a typical dumbbell topology. Figure 3 , the aggregation nodes R1 and R2 form a bottleneck link, the bandwidth is set to 20Mbps, and the propagation delay is set to 20ms; S1 to S n represents n transmission sources, which are connected to R1 to form access links. The bandwidth of the access links is set to 8 Mbps and the propagation delay is set to 10 ms. n Represents n receiving sources, which are connected to R2 to form an output link. The bandwidth of the output link is set to 8Mbps, and the propagation delay is set to 10ms. AQM algorithms are executed on the bottleneck link: RED algorithm, URED algorithm, TRED algorithm, AgRED algorithm, HC-AQM algorithm, where , , , , based on experience, the router buffer size is set to 150 packets.
[0080] For a static network environment, this embodiment may not adjust concurrent connections during the simulation process to keep the network topology unchanged, thereby maintaining the stability of the entire network. For the embodiment of a dynamic network environment during simulation, concurrent connections may be adjusted during the simulation process. The heavier the load, the more serious the network congestion tends to be. In order to better verify the performance of this method in a network environment, two situations are simulated: a high load situation with 75 sending sources and a traffic burst situation with [75,115] sending sources, which are respectively recorded as load one and load two. The traffic burst situation is specifically as follows: 75 sending sources are started at the beginning of the simulation; 40 sending sources are added at 35 seconds, and the number of sending sources becomes 115 at this time; 40 sending sources are reduced at 65 seconds, and the number of sending sources becomes 75 at this time. The simulation process lasts for 100 seconds.
[0081] In this embodiment, HC-AQM is compared with existing AQM methods such as RED algorithm, URED algorithm, TRED algorithm, and AgRED algorithm. The average queue length of the RED algorithm changes over time under high load conditions. Figure 4 As shown in the figure, the average queue length of the URED algorithm changes over time. Figure 5 As shown in the figure, the average queue length of the TRED algorithm changes over time. Figure 6 As shown in the figure, the average queue length of the AgRED algorithm changes over time. Figure 7 As shown; the average queue length of the HC-AQM algorithm changes over time as shown Figure 8 The average queue length of the RED algorithm changes over time under traffic burst conditions. Fig. 9 As shown in the figure, the average queue length of the URED algorithm changes over time. Fig.10 As shown in the figure, the average queue length of the TRED algorithm changes over time. Fig.11 As shown in the figure, the average queue length of the AgRED algorithm changes over time. Fig.12 As shown; the average queue length of the HC-AQM algorithm changes over time as shown Fig.13 shown.
[0082] Figure 4~Figure 13 It reflects the overall situation of the queue length of the five algorithms under two loads. Comparative analysis shows that the average queue length of the RED algorithm, URED algorithm, TRED algorithm and AgRED algorithm is basically stable at Compared with the other four algorithms, the average queue length of the HC-AQM algorithm is generally maintained within the ideal range under the two load conditions, and the overall length and fluctuation range are smaller than the other four algorithms, so that the router buffer has enough space to handle burst traffic and effectively improve the packet forwarding efficiency.
[0083] The comparison of the throughput of the five algorithms as the number of sending sources changes is shown in the figure below: Fig.14 As shown in the figure, from the comparative analysis, it can be seen that since the HC-AQM algorithm reduces unnecessary packet loss when the average queue length is short and its change rate is small, more traffic is forwarded, which improves the link utilization. The average throughput of the HC-AQM algorithm is the highest compared with the RED algorithm, URED algorithm, TRED algorithm and AgRED algorithm regardless of the load condition, and the bandwidth of the bottleneck link is fully utilized.
[0084] The comparison of packet loss rate of five algorithms with the number of sending sources is as follows: Fig.15As shown in the figure, the comparative analysis shows that no matter what load condition it is in, since the HC-AQM algorithm uses two indicators, the Hurst parameter and the average queue growth rate, to reflect the changes in network traffic, it is applied to the adjustment of the packet drop probability, and adopts a reasonable packet drop strategy for the packets arriving at the routing node according to the detected changes to reduce the packet drop events arriving at the routing node. Therefore, the average packet loss rate of this method is smaller than that of the RED algorithm, the URED algorithm, the TRED algorithm, and the AgRED algorithm.
[0085] The comparison of the delay of the five algorithms with the number of sending sources is as follows: Fig.16 As shown in the figure, the comparison of jitter with the number of transmission sources is as follows: Fig.17 As shown in the figure, the comparative analysis shows that this method adjusts the drop probability more timely according to the average queue length and the queue length change rate. Under load 1, the end-to-end average delay in the HC-AQM algorithm is reduced by 5.00%, 5.34%, and 5.07% compared with the RED algorithm, URED algorithm, and TRED algorithm, respectively, and is 0.41% higher than the AgRED algorithm; under load 2, the delay under the HC-AQM algorithm is reduced by 4.28%, 4.50%, and 4.26% compared with the RED algorithm, URED algorithm, and TRED algorithm, respectively, and is 1.92% higher than the AgRED algorithm. Therefore, regardless of the load condition, the end-to-end average delay of the HC-AQM algorithm is smaller than that of the RED algorithm, URED algorithm, and TRED algorithm, but slightly larger than that of the AgRED algorithm. This is because the packet loss rate of the HC-AQM algorithm is very low, so in this case, the delay will inevitably increase. Therefore, the performance of the HC-AQM algorithm in terms of delay still needs to be further improved.
[0086] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.
Claims
1. A queue management method based on dynamic Hurst parameters and queue growth rate, characterized in that: The router buffer has multiple queues composed of data packets. The following steps S1 to S7 are executed to complete the dynamic management of the queues: Step S1: monitor the queue status in the target router buffer in real time and wait for new data packets to arrive; Step S2: after the new data packet arrives at the target router buffer, it is determined whether all the data in the data packet needs to be discarded according to the queue status of the target router buffer at the current moment, including the average queue length and the average queue growth rate at the current moment. If yes, it returns to step S1, otherwise it jumps to step S3; Step S3: Put the new data packet into the target router buffer, and calculate the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment, including the average queue length and average queue growth rate at the next moment, according to the queue state of the target router buffer at the current moment; The method for calculating the queue state of the target router buffer at the next moment in step S3 is as follows: The expression for calculating the average queue length at the next moment is as follows: ; In the formula, express t The average queue length at time +1, express t The average queue length at time, express t The instantaneous queue length at time +1, represents the weight coefficient; The expression for calculating the average queue growth rate at the next moment is as follows: ; In the formula, and Respectively t +1, t The average queue growth rate at time, express t The instantaneous queue length at time; Step S4: Calculate the drop probability based on the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment; In step S4, if the Hurst parameter of the queue at the current moment and the queue state of the target router buffer at the next moment meet one of the following six conditions, the corresponding discard probability is calculated, otherwise the process jumps to step S7: like ,and , then the discard probability is P1; in, H Indicates the Hurst parameter of the queue at the current moment. express t Average queue growth rate at time +1; The discard probability P1 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; like ,and , ε is a constant, then the discard probability is P2; The discard probability P2 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; like ,and , then the discard probability is P3; The discard probability P3 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; like ,and , then the discard probability is P4; The discard probability P4 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; like ,and , then the discard probability is P5; The discard probability P5 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; like ,and , then the discard probability is P6; The discard probability P6 is as follows: ; In the formula, is the maximum threshold of the average queue length, is the minimum threshold of the average queue length, ave is the average queue length, is the middle threshold of the average queue length, is the maximum drop probability; Step S5: according to the discard probability, the data in the new data packet is placed into the queue of the target router buffer, and the remaining data in the new data packet is discarded from the target router buffer; Step S6: Update the queue state of the target router buffer and calculate the Hurst parameter and queue state of the queue at the next moment; Step S7: Wait for a new data packet to arrive, and jump to step S2.
2. The queue management method based on dynamic Hurst parameter and queue growth rate according to claim 1, characterized in that: In step S3, the R / S analysis method is used to calculate the Hurst parameter of the queue at the current moment.
