CN117651024A - Method for predicting network link congestion of data center - Google Patents
Method for predicting network link congestion of data center Download PDFInfo
- Publication number
- CN117651024A CN117651024A CN202311636573.9A CN202311636573A CN117651024A CN 117651024 A CN117651024 A CN 117651024A CN 202311636573 A CN202311636573 A CN 202311636573A CN 117651024 A CN117651024 A CN 117651024A
- Authority
- CN
- China
- Prior art keywords
- data
- congestion
- training
- deep learning
- learning model
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 95
- 238000012549 training Methods 0.000 claims abstract description 126
- 238000012545 processing Methods 0.000 claims abstract description 97
- 238000013136 deep learning model Methods 0.000 claims abstract description 86
- 238000007405 data analysis Methods 0.000 claims abstract description 65
- 230000003044 adaptive effect Effects 0.000 claims abstract description 6
- 238000012795 verification Methods 0.000 claims description 55
- 238000012797 qualification Methods 0.000 claims description 33
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000006854 communication Effects 0.000 abstract description 9
- 238000004891 communication Methods 0.000 abstract description 7
- 230000005540 biological transmission Effects 0.000 description 9
- 238000012937 correction Methods 0.000 description 8
- 238000012986 modification Methods 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/50—Reducing energy consumption in communication networks in wire-line communication networks, e.g. low power modes or reduced link rate
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention relates to a method for predicting network link congestion of a data center, which relates to the technical field of network communication management and comprises the following steps: the data acquisition module acquires congestion characteristic data in the network operation process, the data processing module processes the congestion in the network operation process through a plurality of congestion control methods, and the data analysis module determines the congestion processing efficiency of each congestion control method; the training module trains a deep learning model based on congestion control to generate an adaptive congestion control model; and inputting the self-adaptive control model into a self-adaptive control module of the server, and inputting network operation data into the self-adaptive congestion control model by the self-adaptive control module when network operation is congested to perform network congestion self-adaptive control. The method is adaptive to different congestion control methods in the complex network environment, so that the control precision of congestion in the complex network environment is improved, and the network operation efficiency is further improved.
Description
Technical Field
The invention relates to the technical field of network communication management, in particular to a method for predicting network link congestion of a data center.
Background
Congestion is a common important factor affecting the transmission rate of network communication in the network communication process, congestion exists in the network, common accompanying phenomena include the situation of reduced transmission rate and packet loss, the reduced transmission rate directly affects the internet surfing experience of users, and the packet loss situation may cause economic loss of users, so that congestion management control is an essential link of network communication, the existing congestion control method is basically sufficient to solve the congestion in the network communication process, but is limited to general network communication, and when complex network communication of multiple routes, multiple servers and/or multiple nodes is encountered, a single congestion control method is almost the best.
Chinese patent publication No.: CN114866489a discloses a congestion control method and device, and a training method and device for a congestion control model, where the congestion control method includes: at the starting time of each set period, reducing the size of a congestion window of a target transmission network to a target size, and sending out a target data packet through the congestion window; determining the round trip delay of a target data packet as the minimum round trip delay of a target transmission network in a corresponding set period; in each set period, acquiring network state information of a target transmission network at intervals of set time length, and inputting the minimum round trip delay and the network state information into a congestion control model to obtain adjustment parameters; wherein the adjustment parameter is used for adjusting the size of the congestion window of the target transmission network; therefore, the congestion control method and device and the training method and device of the congestion control model have the problem that the congestion in the complex network environment cannot be adaptively controlled, so that the network operation efficiency is low.
Disclosure of Invention
Therefore, the invention provides a method for predicting network link congestion of a data center, which is used for solving the problem that the network operation efficiency is low because the congestion in a complex network environment cannot be adaptively controlled in the prior art.
To achieve the above object, the present invention provides a method for predicting congestion of a network link of a data center, including:
step S1, a data acquisition module acquires congestion characteristic data in the network operation process, a data processing module processes the congestion in the network operation process through a plurality of congestion control methods, and a data analysis module determines the congestion processing efficiency of each congestion control method;
step S2, training a deep learning model based on congestion control by a training module to generate an adaptive congestion control model;
and step S3, inputting the self-adaptive control model into a self-adaptive control module of the server, and inputting network operation data into the self-adaptive congestion control model by the self-adaptive control module to perform network congestion self-adaptive control when network operation is congested.
Further, in the step S1, when the data analysis module determines the processing efficiency of each congestion control method on congestion, the data analysis module calculates the processing efficiency P of the congestion according to the round trip delay variation D when the data processing module completes processing the congestion by the congestion control method and the data amount W when the data processing module completes processing the congestion by the congestion control method, and sets p=w/D. .
Further, in the step S1, when the data analysis unit calculates that the processing efficiency of congestion is completed, determining a standard value of processing efficiency when the data processing module processes the feature data through the congestion control method according to a comparison result of the packet loss rate U in the feature data acquired by the data acquisition module and a preset packet loss rate,
wherein the data analysis module is provided with a first preset packet loss rate U1, a second preset packet loss rate U2, a first processing efficiency standard value P1, a second processing efficiency standard value P2 and a third processing efficiency standard value P3, wherein U1 is more than U2, P1 is more than P2 and less than P3,
when U is less than or equal to U1, the data analysis module determines that the standard value of the processing efficiency is P3;
when U1 is more than or equal to U2, the data analysis module determines that the standard value of the processing efficiency is P2;
and when U is more than U2, the data analysis module determines that the processing efficiency standard value is P1.
Further, in the step S1, when the data analysis unit determines that the processing efficiency standard value of congestion is completed, the processing efficiency P is compared with the processing efficiency standard value Pi, and the feature data and the congestion control method corresponding thereto are determined as single training data according to the comparison result, wherein i=1, 2,3,
if P is less than or equal to Pi, the data processing module determines that the characteristic data and the congestion control method corresponding to the characteristic data cannot be used as training data;
if P is more than Pi, the data processing module determines that the characteristic data and the corresponding control method can be used as training data;
when the data analysis module determines that a plurality of characteristic data and a corresponding control method thereof can be used as single training data, the training data which can be used as training data are formed into a training data set, and the training data set is divided into a training data set and a verification data set according to a preset data quantity proportion.
Further, in the step S2, when the training module trains the deep learning model based on congestion control, the data analysis module determines the complexity F of the feature data in congestion, and sets
Wherein A is the data quantity of different characteristic data in the congestion data, az is the data quantity of the congestion data, G is the round trip delay of the network operation in congestion, and G0 is the preset round trip delay of the network operation in congestion.
Further, in the step S2, when the training module trains the deep learning model based on congestion control, the training module determines the iteration number when training the deep learning model based on congestion control according to the comparison result of the complexity F and the preset complexity,
wherein the training module is provided with a first preset complexity F1, a second preset complexity F2, a first iteration number W1, a second iteration number W2 and a third iteration number W3, wherein F1 is smaller than F2, W1 is smaller than W2 and smaller than W3,
when F is less than or equal to F1, the training module sets the iteration times of the deep learning model to W1;
when F1 is more than F and less than or equal to F2, the training module sets the iteration times of the deep learning model to W2;
when F > F2, the training module sets the iteration number of the deep learning model to W3.
Further, in the step S2, when training the deep learning model based on congestion control by the training module is completed, the verification data set is input to the trained deep learning model for verification, and when single verification data in the verification data set is input to the deep learning model for verification, the data analysis module determines a qualified parameter T for single verification according to the round trip delay output value Ds and the packet loss rate output value Us of the deep learning model during verification, and sets
Wherein Da is the round trip delay in the single verification data, and Ds is the packet loss rate in the single verification data.
Further, in the step S2, when the data analysis module determines that the pass parameter T is completed, the data analysis module determines whether the deep learning model is passed or not through a single verification according to a comparison result of the pass parameter T and a preset pass parameter T0,
if the absolute value T-T0 absolute value is more than Tx, the data analysis module determines that the deep learning model is qualified through single verification;
and if the absolute T-T0 is less than or equal to Tx, the data analysis module determines that the deep learning model is unqualified in a single verification.
Further, in the step S2, when the training module trains a deep learning model based on congestion control, the data analysis module counts a model qualification rate R when the verification data verifies the deep learning model, sets r=c/Cz, and the training module determines whether the deep learning is completed according to a comparison result of the model qualification rate R and a preset model qualification rate, wherein the training module is provided with a first preset model qualification rate R1 and a second preset model qualification rate R2, C is a qualification number of the verification data verifying the deep learning model, cz is a total number of the verification data verifying the deep learning model, R1 is less than R2,
if R is less than or equal to R1, the training module judges that the training of the deep learning model is not finished;
if R1 is more than R and less than or equal to R2, the training module judges that the training of the deep learning model is not completed;
and if R is more than R2, the training module judges that the training of the deep learning model is completed.
Further, in the step S2, when the training module determines that training of the deep learning model is not completed and R is less than or equal to R1, the training module adjusts the iteration number of the deep learning model; when the training module determines that the deep learning model training is not completed and R is more than R1 and less than or equal to R2, the data analysis module corrects the processing efficiency standard value.
Compared with the prior art, the method has the beneficial effects that the congestion characteristic data in the network operation process are acquired through the acquisition module, the congestion control method is determined to be selected according to the congestion characteristic data to process the congestion, the processing efficiency of the congestion control method is analyzed, the congestion control method with the processing efficiency reaching the standard and the corresponding data are used as a training data set, the self-adaptive congestion control model for the complex network environment is trained through the training data set, and the self-adaptive control model is input into the self-adaptive control module of the server after the self-adaptive control model is trained, when the congestion of the network is detected, the self-adaptive corresponding congestion control method is carried out according to the congestion characteristic, or the self-adaptive congestion control method of each route, each server and/or multiple nodes is carried out on the complex network environment of multiple routes, each server and/or multiple nodes, so that the congestion control precision of the congestion under the complex network environment is improved, and the network operation efficiency is further improved.
Further, when the data analysis module determines the processing efficiency of each congestion control method, the invention determines the processing efficiency of the congestion according to the round trip delay of the congestion and the length of time for processing the congestion according to the corresponding congestion control method, and determines the processing efficiency standard value of the congestion according to the packet loss rate when the processing efficiency is determined to be completed, and determines whether the corresponding characteristic data and the congestion control method can be used as training data according to the comparison result of the processing efficiency and the processing efficiency standard value when the processing efficiency is determined to be completed, thereby further improving the control precision of the congestion under the complex network environment and further improving the network operation efficiency.
Further, when the deep learning model is trained, the complexity of the feature data is calculated through the data quantity and the round trip delay of different feature data of the feature data, so that the iteration number of training the deep learning model is determined according to the complexity of the feature data, the control precision of congestion in a complex network environment is further improved, and the network operation efficiency is further improved.
Further, when the iteration times of the training deep learning model are determined, the iteration times are determined according to the comparison result of the calculated complexity and the preset complexity by setting the preset complexity, so that the control precision of congestion in a complex network environment is further improved, and the network operation efficiency is further improved.
Furthermore, when the deep learning model is trained, the deep learning model is verified, and when the deep learning model is verified, the qualification parameters of the single verification data when the deep learning model is verified are calculated, and whether the single verification is qualified or not is determined according to the comparison result of the qualification parameters and preset qualification parameters, so that the control precision of congestion in a complex network environment is further improved, and the network operation efficiency is further improved.
Further, when the verification of the deep learning model by the verification data set is completed, the model qualification rate of the deep learning model verification by the verification data set is calculated, whether the model is trained is determined to be completed or not is determined according to the comparison result of the model qualification rate and the preset qualification rate, and the precision of the self-adaptive congestion control model is improved, so that the control precision of congestion in a complex network environment is improved, and the network operation efficiency is further improved.
Further, when the deep learning model is not trained, the corresponding adjusting mode is determined according to the model qualification rate during verification so as to adjust the iteration times of the deep learning model or correct the processing efficiency standard value, and the accuracy of the self-adaptive congestion control model is further improved, so that the congestion control accuracy under the complex network environment is improved, and the network operation efficiency is improved.
Drawings
FIG. 1 is a logical block diagram of a system implementing a method of predicting data center network link congestion in accordance with an embodiment of the present invention;
fig. 2 is a flow chart of a method of predicting data center network link congestion in accordance with an embodiment of the present invention.
Detailed Description
In order that the objects and advantages of the invention will become more apparent, the invention will be further described with reference to the following examples; it should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are merely for explaining the technical principles of the present invention, and are not intended to limit the scope of the present invention.
It should be noted that, in the description of the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present invention can be understood by those skilled in the art according to the specific circumstances.
Referring now to fig. 1, therein is shown a system logic diagram of a method for predicting data center network link congestion in accordance with an embodiment of the present invention.
The invention provides a system for implementing a method for predicting congestion of a network link of a data center, which comprises the following steps:
the data acquisition module is used for acquiring the characteristic data of congestion in the network operation process;
the data processing module is connected with the data acquisition module and is used for processing congestion in the network operation process through a plurality of congestion control methods;
the data analysis module is connected with the data processing module and used for determining the processing efficiency of each congestion control method on congestion and verifying the model by verifying the data set when model training is finished;
the training module is connected with the data analysis module and used for training the deep learning model based on congestion control to generate an adaptive congestion control model;
the self-adaptive control module is arranged in the server, connected with the training module and used for bearing the self-adaptive congestion control model after training.
In the embodiment of the invention, the characteristic data of congestion comprise, but are not limited to, round trip delay, transmission rate, packet loss rate, congestion data and round trip delay variation.
In the embodiment of the invention, the congestion control method comprises, but is not limited to, CUBIC based on packet loss, vegas based on time delay, BBR based on detection, TCP congestion control and VCP protocol.
Referring to fig. 2, a flowchart of a method for predicting congestion of a data center network link according to an embodiment of the present invention is shown.
The method for predicting the network link congestion of the data center comprises the following steps:
step S1, a data acquisition module acquires congestion characteristic data in the network operation process, a data processing module processes the congestion in the network operation process through a plurality of congestion control methods, and a data analysis module determines the congestion processing efficiency of each congestion control method;
step S2, training a deep learning model based on congestion control by a training module to generate an adaptive congestion control model;
and step S3, inputting the self-adaptive control model into a self-adaptive control module of the server, and inputting network operation data into the self-adaptive congestion control model by the self-adaptive control module to perform network congestion self-adaptive control when network operation is congested.
Specifically, in the step S1, when the data analysis module determines the processing efficiency of each congestion control method for congestion, the data analysis module calculates the processing efficiency P of congestion according to the round trip delay variation D when the data processing module completes processing the congestion by the congestion control method and the data amount W when the data processing module completes processing the congestion by the congestion control method, and sets p=w/D.
Specifically, in the step S1, when the data analysis unit calculates that the processing efficiency of congestion is completed, determining a standard value of processing efficiency when the data processing module processes the feature data through the congestion control method according to a comparison result of the packet loss rate U in the feature data acquired by the data acquisition module and a preset packet loss rate,
wherein the data analysis module is provided with a first preset packet loss rate U1, a second preset packet loss rate U2, a first processing efficiency standard value P1, a second processing efficiency standard value P2 and a third processing efficiency standard value P3, wherein U1 is more than U2, P1 is more than P2 and less than P3,
when U is less than or equal to U1, the data analysis module determines that the standard value of the processing efficiency is P3;
when U1 is more than or equal to U2, the data analysis module determines that the standard value of the processing efficiency is P2;
and when U is more than U2, the data analysis module determines that the processing efficiency standard value is P1.
In the embodiment of the invention, a person skilled in the art can set the values of the first preset packet loss rate and the second preset packet loss rate according to actual conditions, which is not limited in the invention, preferably, the selection of the first preset packet loss rate is less than 150% of the packet loss rate under the normal throughput, and the selection of the second preset packet loss rate is less than 200% of the packet loss rate under the normal throughput.
In the embodiment of the invention, the first processing efficiency standard value, the second processing efficiency standard value and the third processing efficiency standard value respectively correspond to 85%, 75% and 65% of the historical average processing efficiency of a plurality of congestion control methods.
Specifically, in the step S1, when the data analysis unit determines that the processing efficiency standard value of congestion is completed, the processing efficiency P is compared with the processing efficiency standard value Pi, and the feature data and the congestion control method corresponding thereto are determined as single training data according to the comparison result, wherein i=1, 2,3,
if P is less than or equal to Pi, the data processing module determines that the characteristic data and the congestion control method corresponding to the characteristic data cannot be used as training data;
if P is more than Pi, the data processing module determines that the characteristic data and the corresponding control method can be used as training data;
when the data analysis module determines that a plurality of characteristic data and a corresponding control method thereof can be used as single training data, the training data which can be used as training data are formed into a training data set, and the training data set is divided into a training data set and a verification data set according to a preset data quantity proportion.
Specifically, in the step S2, when the training module trains the deep learning model based on congestion control, the data analysis module determines the complexity F of the feature data at the time of congestion, and sets
Wherein A is the data quantity of different characteristic data in the congestion data, az is the data quantity of the congestion data, G is the round trip delay of the network operation in congestion, and G0 is the preset round trip delay of the network operation in congestion.
In the embodiment of the invention, the preset round-trip delay value is the average round-trip delay of the current server for a plurality of historical congestions of the same transmission data quantity.
Specifically, in the step S2, when the training module trains the deep learning model based on congestion control, the training module determines the iteration number when training the deep learning model based on congestion control according to the comparison result of the complexity F and the preset complexity,
wherein the training module is provided with a first preset complexity F1, a second preset complexity F2, a first iteration number W1, a second iteration number W2 and a third iteration number W3, wherein F1 is smaller than F2, W1 is smaller than W2 and smaller than W3,
when F is less than or equal to F1, the training module sets the iteration times of the deep learning model to W1;
when F1 is more than F and less than or equal to F2, the training module sets the iteration times of the deep learning model to W2;
when F > F2, the training module sets the iteration number of the deep learning model to W3.
In the embodiment of the present invention, the value of the first preset complexity is 1, and the value of the second preset complexity is 1.5, and those skilled in the art can set the preset complexity and the iteration number according to the actual situation, which is not limited in the present invention.
Specifically, in the step S2, when training of the deep learning model based on congestion control by the training module is completed, the verification data set is input to the trained deep learning model for verification, and when single verification data in the verification data set is input to the deep learning model for verification, the data analysis module determines a single verification pass parameter T according to the round trip delay output value Ds and the packet loss rate output value Us of the deep learning model at the time of verification, and sets
Wherein Da is the round trip delay in the single verification data, and Ds is the packet loss rate in the single verification data.
Specifically, in the step S2, when the data analysis module determines that the pass parameter T is completed, the data analysis module determines whether the deep learning model is passed through a single verification according to a comparison result of the pass parameter T and a preset pass parameter T0,
if the absolute value T-T0 absolute value is more than Tx, the data analysis module determines that the deep learning model is qualified through single verification;
and if the absolute T-T0 is less than or equal to Tx, the data analysis module determines that the deep learning model is unqualified in a single verification.
In the embodiment of the present invention, the preset qualified parameter T0 has a value of 2, tx is a preset qualified parameter difference value, and tx=0.2 is set.
Specifically, in the step S2, when the training module trains a deep learning model based on congestion control, the data analysis module counts a model qualification rate R when the verification data verifies the deep learning model, sets r=c/Cz, the training module determines whether the deep learning is completed according to a comparison result of the model qualification rate R and a preset model qualification rate, wherein the training module is provided with a first preset model qualification rate R1 and a second preset model qualification rate R2, C is a qualification number of the verification data verifying the deep learning model, cz is a total number of the verification data verifying the deep learning model, R1 < R2,
if R is less than or equal to R1, the training module judges that the training of the deep learning model is not finished;
if R1 is more than R and less than or equal to R2, the training module judges that the training of the deep learning model is not completed;
and if R is more than R2, the training module judges that the training of the deep learning model is completed.
In the embodiment of the invention, the value of the qualification rate of the first preset model is 0.90, and the value of the qualification rate of the second preset model is 0.93.
Specifically, in the step S2, when the training module determines that the training of the deep learning model is not completed and r+.r1, the training module adjusts the iteration number of the deep learning model; when the training module determines that the deep learning model training is not completed and R is more than R1 and less than or equal to R2, the data analysis module corrects the processing efficiency standard value.
Specifically, when the training module adjusts the iteration times of the deep learning model, a first qualification rate ratio Ba of the model qualification rate R and a first preset model qualification rate is calculated, ba=R/R1 is set, a corresponding time adjustment coefficient is selected according to the comparison result of the first ratio and the preset ratio to adjust the iteration times,
wherein the training module is provided with a first preset ratio B1, a second preset ratio B2, a first time number adjustment coefficient K1, a second time number adjustment coefficient K2 and a third time number adjustment coefficient K3, wherein B1 is more than B2, K1 is more than K1 and K2 is more than K3 and less than 1.5,
when Ba is less than or equal to B1, the training module selects a first-time number adjustment coefficient K1 to adjust the iteration times;
when B1 is more than Ba and less than or equal to B2, the training module selects a second time adjustment coefficient K2 to adjust the iteration times;
when Ba is more than B2, the training module selects a third adjusting coefficient K3 to adjust the iteration times;
when the training module selects the jth time adjustment coefficient Kj to adjust the iteration times, j=1, 2,3 is set, the training module sets the adjusted iteration times to W4, and w4=wn×kj is set, wherein n=1, 2,3.
Specifically, when the data analysis module corrects the processing efficiency standard value, a second ratio Bb of the model qualification rate R and a second preset model qualification rate R2 is calculated, and corresponding correction coefficients are selected according to the comparison result of the second ratio and the preset ratio to correct the processing efficiency standard value,
wherein the data analysis module is provided with a first preset ratio B1, a second preset ratio B2, a first correction coefficient X1, a second correction coefficient X2 and a third correction coefficient X3, X1 is more than 0.5 and less than X2 and less than X3 and less than 1,
when Bb is less than or equal to B1, the data analysis module selects a first correction coefficient X1 to correct the standard value of the processing efficiency;
when B1 is more than Bb and less than or equal to B2, the data analysis module selects a second correction coefficient X2 to correct the standard value of the processing efficiency;
when Bb is more than B2, the data analysis module selects a third correction coefficient X3 to correct the standard value of the processing efficiency;
when the data analysis module selects the e correction coefficient Xe to correct the processing efficiency standard value, e=1, 2 and 3 are set, the data analysis module sets the corrected processing efficiency standard value as P4, and P4=PiXe is set.
In the embodiment of the invention, the value of the first preset ratio is 0.7, and the value of the second preset ratio is 0.8.
Thus far, the technical solution of the present invention has been described in connection with the preferred embodiments shown in the drawings, but it is easily understood by those skilled in the art that the scope of protection of the present invention is not limited to these specific embodiments. Equivalent modifications and substitutions for related technical features may be made by those skilled in the art without departing from the principles of the present invention, and such modifications and substitutions will be within the scope of the present invention.
The foregoing description is only of the preferred embodiments of the invention and is not intended to limit the invention; various modifications and variations of the present invention will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method of predicting data center network link congestion, comprising:
step S1, a data acquisition module acquires congestion characteristic data in the network operation process, a data processing module processes the congestion in the network operation process through a plurality of congestion control methods, and a data analysis module determines the congestion processing efficiency of each congestion control method;
step S2, training a deep learning model based on congestion control by a training module to generate an adaptive congestion control model;
and step S3, inputting the self-adaptive control model into a self-adaptive control module of the server, and inputting network operation data into the self-adaptive congestion control model by the self-adaptive control module to perform network congestion self-adaptive control when network operation is congested.
2. The method according to claim 1, wherein in the step S1, when the data analysis module determines the processing efficiency of each congestion control method for congestion, the data analysis module calculates the processing efficiency P of congestion according to the round trip delay variation D when the data processing module completes the processing of congestion by the congestion control method and the data processing module completes the processing of data by the congestion control method, and sets p=w/D.
3. The method for predicting network link congestion in a data center according to claim 2, wherein in said step S1, when said data analysis unit calculates that the processing efficiency of said congestion is completed, a standard value of the processing efficiency when said data processing module processes said characteristic data by said congestion control method is determined based on a comparison result of a packet loss rate U in said characteristic data acquired by said data acquisition module and a preset packet loss rate,
wherein the data analysis module is provided with a first preset packet loss rate U1, a second preset packet loss rate U2, a first processing efficiency standard value P1, a second processing efficiency standard value P2 and a third processing efficiency standard value P3, wherein U1 is more than U2, P1 is more than P2 and less than P3,
when U is less than or equal to U1, the data analysis module determines that the standard value of the processing efficiency is P3;
when U1 is more than or equal to U2, the data analysis module determines that the standard value of the processing efficiency is P2;
and when U is more than U2, the data analysis module determines that the processing efficiency standard value is P1.
4. A method of predicting congestion in a data center network according to claim 3, wherein in said step S1, when said data analysis unit determines that the processing efficiency criterion value of said congestion is complete, said processing efficiency P is compared with the processing efficiency criterion value Pi, and said characteristic data and its corresponding congestion control method are determined as a single training data based on the comparison result, wherein i = 1,2,3,
if P is less than or equal to Pi, the data processing module determines that the characteristic data and the congestion control method corresponding to the characteristic data cannot be used as training data;
if P is more than Pi, the data processing module determines that the characteristic data and the corresponding control method can be used as training data;
when the data analysis module determines that a plurality of characteristic data and a corresponding control method thereof can be used as single training data, the training data which can be used as training data are formed into a training data set, and the training data set is divided into a training data set and a verification data set according to a preset data quantity proportion.
5. The method according to claim 4, wherein in the step S2, when the training module trains a deep learning model based on congestion control, the data analysis module determines the complexity F of the feature data in congestion, and sets
Wherein A is the data quantity of different characteristic data in the congestion data, az is the data quantity of the congestion data, G is the round trip delay of the network operation in congestion, and G0 is the preset round trip delay of the network operation in congestion.
6. The method according to claim 5, wherein in the step S2, when the training module trains the deep learning model based on congestion control, the training module determines the iteration number when training the deep learning model based on congestion control according to the comparison result of the complexity F and a preset complexity,
wherein the training module is provided with a first preset complexity F1, a second preset complexity F2, a first iteration number W1, a second iteration number W2 and a third iteration number W3, wherein F1 is smaller than F2, W1 is smaller than W2 and smaller than W3,
when F is less than or equal to F1, the training module sets the iteration times of the deep learning model to W1;
when F1 is more than F and less than or equal to F2, the training module sets the iteration times of the deep learning model to W2;
when F > F2, the training module sets the iteration number of the deep learning model to W3.
7. The method according to claim 6, wherein in the step S2, when training of the deep learning model based on congestion control by the training module is completed, the verification data set is input to the deep learning model whose training is completed for verification, and when single verification data in the verification data set is input to the deep learning model for verification, the data analysis module determines a qualification parameter T for one-time verification based on the round trip delay output value Ds and the packet loss rate output value Us of the deep learning model at the time of verification, and sets
Wherein Da is the round trip delay in the single verification data, and Ds is the packet loss rate in the single verification data.
8. The method for predicting network link congestion in a data center of claim 7, wherein in said step S2, when said data analysis module determines that said pass parameter T is complete, said data analysis module determines whether said deep learning model is passed for a single verification based on a comparison of said pass parameter T and a preset pass parameter T0,
if the absolute value T-T0 absolute value is more than Tx, the data analysis module determines that the deep learning model is qualified through single verification;
and if the absolute T-T0 is less than or equal to Tx, the data analysis module determines that the deep learning model is unqualified in a single verification.
9. The method according to claim 8, wherein in the step S2, when the training module trains a deep learning model based on congestion control, the data analysis module counts a model qualification rate R when the verification data verifies the deep learning model, sets r=c/Cz, and the training module determines whether the deep learning is completed according to a comparison result of the model qualification rate R and a preset model qualification rate, wherein the training module is provided with a first preset model qualification rate R1 and a second preset model qualification rate R2, C is a qualification number of times the verification data verifies the deep learning model, cz is a total number of times the verification data verifies the deep learning model, R1 < R2,
if R is less than or equal to R1, the training module judges that the training of the deep learning model is not finished;
if R1 is more than R and less than or equal to R2, the training module judges that the training of the deep learning model is not completed;
and if R is more than R2, the training module judges that the training of the deep learning model is completed.
10. The method of predicting data center network link congestion as set forth in claim 9, wherein in said step S2, when said training module determines that said deep learning model training is incomplete and r+.r1, said training module adjusts the number of iterations of said deep learning model; when the training module determines that the deep learning model training is not completed and R is more than R1 and less than or equal to R2, the data analysis module corrects the processing efficiency standard value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311636573.9A CN117651024A (en) | 2023-12-01 | 2023-12-01 | Method for predicting network link congestion of data center |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311636573.9A CN117651024A (en) | 2023-12-01 | 2023-12-01 | Method for predicting network link congestion of data center |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117651024A true CN117651024A (en) | 2024-03-05 |
Family
ID=90049099
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311636573.9A Pending CN117651024A (en) | 2023-12-01 | 2023-12-01 | Method for predicting network link congestion of data center |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117651024A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110581808A (en) * | 2019-08-22 | 2019-12-17 | 武汉大学 | Congestion control method and system based on deep reinforcement learning |
| CN113595923A (en) * | 2021-08-11 | 2021-11-02 | 国网信息通信产业集团有限公司 | Network congestion control method and device |
| CN113992599A (en) * | 2021-11-18 | 2022-01-28 | 北京达佳互联信息技术有限公司 | Training method and device of time delay prediction model and congestion control method and device |
| WO2023226284A1 (en) * | 2022-05-26 | 2023-11-30 | 鹏城实验室 | Deep learning model training method and apparatus, device and storage medium |
-
2023
- 2023-12-01 CN CN202311636573.9A patent/CN117651024A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110581808A (en) * | 2019-08-22 | 2019-12-17 | 武汉大学 | Congestion control method and system based on deep reinforcement learning |
| CN113595923A (en) * | 2021-08-11 | 2021-11-02 | 国网信息通信产业集团有限公司 | Network congestion control method and device |
| CN113992599A (en) * | 2021-11-18 | 2022-01-28 | 北京达佳互联信息技术有限公司 | Training method and device of time delay prediction model and congestion control method and device |
| WO2023226284A1 (en) * | 2022-05-26 | 2023-11-30 | 鹏城实验室 | Deep learning model training method and apparatus, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Foh et al. | Comments on IEEE 802.11 saturation throughput analysis with freezing of backoff counters | |
| Haßlinger et al. | The Gilbert-Elliott model for packet loss in real time services on the Internet | |
| Konrad et al. | A Markov-based channel model algorithm for wireless networks | |
| US10027567B2 (en) | Time measurement in a packet-switched communication network | |
| US10050909B2 (en) | Method and device for determining transmission buffer size | |
| Hooghiemstra et al. | Delay distributions on fixed internet paths | |
| US10637788B2 (en) | Stability of delay-based congestion control in a computer network using an alpha-beta filter and round-trip-time predictor | |
| CN103259696A (en) | Network bandwidth detecting method and device and network devices | |
| CN106899387A (en) | Speed self-adaption method based on LoRa wireless networks | |
| US7469203B2 (en) | Wireless network hybrid simulation | |
| EP2066048B1 (en) | A method and device of determining the instant of consideration of a modification of at least one reception condition for signals in a communication system | |
| CN117651024A (en) | Method for predicting network link congestion of data center | |
| Kalman et al. | Modeling the delays of successively-transmitted Internet packets | |
| US20080225746A1 (en) | Method and system for generating packet delay variation with a uniform distribution | |
| Fischer | Optimal sequence-based control of networked linear systems | |
| Mitzenmacher et al. | Towards more complete models of tcp latency and throughput | |
| Li et al. | Packet transport for maritime communications: A streaming coded UDP approach | |
| Altman et al. | TCP modeling in the presence of nonlinear window growth | |
| CN114613372A (en) | Error concealment technical method for preventing packet loss in audio transmission | |
| Xie et al. | An adaptive tuning algorithm for IEEE 802.15. 4-based network control system | |
| US7697635B2 (en) | Method and apparatus for controlling the decision point of a receiver for digital communication signals | |
| CN101753367B (en) | Congestion packet loss membership function construction method based on potential function | |
| Martorell et al. | A refined 3D Markov model for non-saturated IEEE 802.11 DCF netwoks | |
| CN117792991B (en) | Automatic switching method for router links and multi-link router | |
| EP4080825A1 (en) | Threshold optimization method and apparatus applied to communications system, and computer readable medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |