A method for dynamic bandwidth allocation in a multi-user multi-service passive optical network

Abstract

The application discloses a kind of multi-user multi-service passive optical network dynamic bandwidth allocation method, belong to optical access network and industrial internet field.The application is based on the passive optical network dynamic bandwidth allocation method of traffic classification, by improving the architecture and bandwidth allocation method of traditional time division multiplexing passive optical network, provide different allocation method for the different type of traffic of user, realize high bandwidth utilization and the bandwidth resource allocation of architecture high flexibility under the scene of optical access network for industrial internet.The application improves flexibility and scalability by dividing two logically isolated virtual OLTs and two slices;By collecting historical traffic data and inputting into the joint neural network model, accurate traffic classification is performed to achieve differentiated resource allocation;By traffic classification, user data traffic is divided into specific network slices, and different bandwidth allocation methods are used in different slices to improve the utilization of overall network bandwidth resources.

Description

TECHNICAL FIELD

[0001] The application relates to a multi-user multi-service passive optical network dynamic bandwidth allocation method and belongs to the fields of optical access networks and industrial Internet. BACKGROUND

[0002] With the rapid development of industrial Internet and network function virtualization technology, various new network services including mobile front-haul and industrial Internet of Things will deploy a large number of access points, and industrial passive optical networks are required to develop towards lower transmission delay and higher throughput.

[0003] Network slicing technology is a method of logically dividing network resources, allowing different applications or services to share resources on the same physical network infrastructure while ensuring isolation and performance requirements. Through network slicing technology, the entire network is divided into multiple independent virtual network slices, each slice having its own network functions, topology, bandwidth and security policies.

[0004] Network traffic generated by users is classified, and different types of network traffic are allocated to different network slices to achieve more targeted resource allocation. Neural networks accurately classify different types of traffic through learning and pattern matching, and the Long Short-Term Memory (LSTM) neural network is a variant of recurrent neural networks suitable for processing time series data, which can capture long-term dependencies of time series and extract and capture periodic features of time series. The Gated Recurrent Unit (GRU) neural network is another variant of recurrent neural networks, which reduces the number of gating units and simplifies the calculation process to improve training and inference efficiency.

[0005] To solve the problems of high service coupling degree, insufficient utilization of bandwidth resources and fixed loss in optical access networks, the application provides a multi-service multi-user dynamic resource allocation method to reduce the latency and packet loss rate of optical access networks. SUMMARY

[0006] To solve the problems of low uplink bandwidth resource utilization and inflexible architecture in optical access networks for industrial Internet, the application aims to provide a multi-user multi-service passive optical network dynamic bandwidth allocation method based on traffic classification, which realizes high bandwidth utilization and flexible bandwidth resource allocation in the scenario of optical access networks for industrial Internet.

[0007] The application is achieved by the following technical solutions:

[0008] The application discloses a multi-user multi-service passive optical network dynamic bandwidth allocation method.

[0009] Step 1, based on network function virtualization technology, logically divide the physical OLT of the TDM-PON into two virtual OLTs, and divide the ONU connected to the OLT into two slices, use the SDN controller to collect the historical traffic generated by the user, input the historical traffic of the user into the joint neural network model of LSTM and GRU in the bandwidth allocation period, and the neural network model predicts the future traffic of the user, including the following sub-steps:

[0010] Step 1.1, based on network function virtualization technology, build two virtual OLTs, vOLT1 and vOLT2, in the physical OLT of the traditional TDM-PON architecture: vOLT1 is used for bandwidth allocation in the periodic traffic slice, and vOLT2 is used for bandwidth allocation in the non-periodic traffic slice;

[0011] Step 1.2, divide the ONU connected to the physical OLT of the TDM-PON into two slices, which are used to receive the periodic traffic and the non-periodic traffic of the user, and the user equipment is connected to the ONU through wired transmission or wireless transmission;

[0012] Step 1.3, set an SDN controller on the side of the physical OLT of the TDM-PON, collect the historical traffic of the user in the previous N periods through the SDN controller, and set a joint neural network of LSTM and GRU in the SDN controller, and predict the traffic of the user in the next two periods according to the historical traffic of the user;

[0013] Step 2, according to the traffic prediction result of step 1, compare the actual traffic of the next period of the user, complete classification of the traffic of the user, divide the traffic into periodic traffic and non-periodic traffic, and the SDN controller will divide the network traffic of different categories into different ONU slices and finally upload to the respective vOLTs, so as to complete the differentiated bandwidth allocation, including the following sub-steps:

[0014] Step 2.1, the traffic prediction module of the SDN controller receives the historical traffic of the user, sets the parameters of the LSTM neural network, and trains the GRU model by using the maximum likelihood estimation method;

[0015] Step 2.2, after the parameters of the joint prediction model of LSTM and GRU in the traffic prediction module are set, input the historical traffic of the user into the model as input, and obtain the traffic condition of the user in the next two periods, as shown in formula (1):

[0016] Predict final = w1*RES LSTM + w2*RES GRU (1)

[0017] wherein Predict final is the final prediction result of the joint neural network, w1 and w2 are weight coefficients, RES LSTM and RES GRU are prediction results of the LSTM model and the GRU model respectively;

[0018] Step 2.3, according to the comparison between the joint prediction result and the actual value of the next period, the incoming traffic is classified into different categories: the traffic close to the actual value and meeting the expectation is classified as periodic traffic, and the traffic not meeting the expectation is classified as aperiodic traffic;

[0019] According to the predicted value and the actual value, the mean absolute percentage error MAPE Periodic is determined, and the traffic with MAPE Periodic less than a set threshold η is classified as periodic traffic;

[0020] Step 3, according to the classification of the traffic in step 2, the bandwidth of the entire network is first allocated to different ONU slices, in the current period, the bandwidth of the periodic traffic slice is allocated according to the prediction result of the traffic prediction module, and the aperiodic traffic slice uses the remaining bandwidth of the total bandwidth; in the ONU slice of the periodic traffic, the vOLT directly allocates the bandwidth to each ONU according to the prediction result of the SDN controller in a polling manner; in the ONU slice of the aperiodic traffic, a fixed bandwidth is allocated to each ONU to avoid the problem that user data in the ONU cannot be transmitted for a long time, and the resources allocated to the aperiodic ONU slice in addition to the fixed bandwidth are allocated to each ONU in a polling manner as common bandwidth;

[0021] Step 3.1, according to the traffic classification result of step 2, the SDN controller uploads the classification result to the OLT for inter-slice bandwidth allocation in the next period, and the inter-slice bandwidth allocation is shown in formula (2) and formula (3):

[0022] B P = B total *n p (2)

[0023] B a = B total *n a (3)

[0024] wherein B P is the allocated bandwidth of the periodic traffic slice, and B ais the allocated bandwidth of non-periodic traffic slice, B total is the total upstream bandwidth of PON network, n p and n a is the normalized coefficient of statistical traffic, as shown in equation (4) and equation (5) respectively:

[0025]

[0026] n a =1-n p (5)

[0027] wherein, Predict p is the prediction result of periodic traffic, Predict a is the prediction result of non-periodic traffic;

[0028] Step 3.2, according to the bandwidth allocation between slices in step 3.1, the bandwidth allocation within the slice is carried out;

[0029] In the non-periodic traffic slice, each ONU is allocated guaranteed bandwidth to avoid long time transmission of low priority traffic, resulting in a large number of packet loss and reducing network throughput, the remaining bandwidth is allocated to different ONUs according to priority, the remaining bandwidth B remaina is shown in equation (6):

[0030] B remaina =B a -m*B f (6)

[0031] wherein, B f is the guaranteed bandwidth allocated to each non-periodic ONU, and m is the number of non-periodic ONUs;

[0032] The allocated bandwidth B allocai of the ONUs in the i-th non-periodic group is shown in equation (7):

[0033]

[0034] wherein, p i is the priority of data packet;

[0035] In the periodic traffic slice, the priority is not distinguished, the vOLT polls the ONUs, and the bandwidth is allocated to each ONU according to the bandwidth request of the ONU;

[0036] It also includes step 4: according to the bandwidth allocation method of step 3, the information transmission method of public network is analyzed and designed, which can improve the traffic load balancing, resource fair allocation, priority management and elastic adaptation in public network, and improve the robustness and user experience of public network.

[0037] Advantages

[0038] 1. The multi-user multi-service passive optical network dynamic bandwidth allocation method of the application divides the physical OLT of the traditional TDM-PON architecture into two logically isolated virtual OLTs based on network function virtualization technology and network slicing technology, and manages the resource allocation of periodic and aperiodic traffic; the ONUs connected to the physical OLT are divided into two logically isolated slices, which are used to receive periodic and aperiodic traffic of users, respectively, and users access the OLT through wired or wireless means, thereby improving the flexibility and scalability of the traditional TDM-PON architecture.

[0039] 2. The multi-user multi-service passive optical network dynamic bandwidth allocation method of the application collects historical traffic data of users and inputs the traffic data into an LSTM and GRU joint neural network model as input to accurately classify the traffic and realize differentiated resource allocation.

[0040] 3. The multi-user multi-service passive optical network dynamic bandwidth allocation method of the application divides user data traffic into specific network slices through traffic classification, and uses different bandwidth allocation methods in different slices in uplink transmission, thereby improving the utilization rate of the entire network bandwidth resource. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 For Figure 1 is a flow chart of the multi-user multi-service passive optical network dynamic bandwidth allocation method of the application;

[0042] Figure 2 For Figure 2 is a TDM-PON architecture diagram based on the improved network slicing technology of the application;

[0043] Figure 3 is a performance result graph of the LSTM and GRU neural network model used in the embodiment to predict periodic traffic;

[0044] Figure 4 is a performance result comparison graph of the method of the application compared with the traditional method in the embodiment. DETAILED DESCRIPTION

[0045] In order to better illustrate the purpose and advantages of the application, the content of the application is further described below in combination with the drawings and examples.

[0046] Example 1:

[0047] In the embodiment, the OLT of the TDM-PON is connected with 32 ONUs, the length of the optical fiber from the ONU to the OLT is 20 kilometers, the round-trip delay is 150 microseconds, there are 160 user terminals in the IPON network, each ONU is connected with five user terminals, the user terminals are connected to different slices through wireless or wired networks, the random data packet generated by each user terminal has a size of 64 bytes to 1518 bytes, the aperiodic traffic generated by each user terminal follows a generalized Pareto distribution and has a priority label, the periodic traffic of the IIoT is generated by users with fixed cycles and sizes, and the dynamic bandwidth allocation period of the network is 2 milliseconds.

[0048] In the above scenario, the application of the multi-user multi-service passive optical network dynamic bandwidth allocation method provides a differentiated allocation method for different types of traffic of users, improves the bandwidth resource utilization of the entire optical access network, and includes the following steps:

[0049] Step 1, based on the network function virtualization technology, logically divide the physical OLT of the TDM-PON into two virtual OLTs, and divide the ONUs connected by the OLT into two slices, each slice has 16 ONUs, use an SDN controller to collect historical traffic generated by users, and input the historical traffic of the users into a joint neural network model of LSTM and GRU within a bandwidth allocation period, and the neural network model predicts the future traffic of the users, including the following sub-steps:

[0050] Step 1.1, based on the network function virtualization technology, build two virtual OLTs, vOLT1 and vOLT2, in the physical OLT of the traditional TDM-PON architecture, vOLT1 is used for bandwidth allocation in the periodic traffic slice, and vOLT2 is used for bandwidth allocation in the aperiodic traffic slice;

[0051] Step 1.2, divide the ONUs connected by the physical OLT of the TDM-PON into two slices for receiving periodic traffic and aperiodic traffic of users, and the user equipment is connected to the ONU through wired transmission or wireless transmission;

[0052] Step 1.3, set an SDN controller on the side of the physical OLT of the TDM-PON, the improved TDM-PON network architecture is as shown in Figure 2 The SDN controller collects the historical traffic of the users in the previous N periods, the SDN controller is internally provided with a joint neural network of LSTM and GRU, and the future traffic of the users within two periods is predicted according to the historical traffic of the users;

[0053] Step 2, according to the flow prediction result of step 1, compare the actual flow of the next period of the user, complete the classification of the user flow, divide it into periodic flow and non-periodic flow, and the SDN controller will divide the network flow of different categories into different ONU slices according to the classification result, and finally upload it to the respective vOLT to complete the differentiated bandwidth allocation, including the following sub-steps:

[0054] Step 2.1, the flow prediction module of the SDN controller receives the historical flow of the user, sets the parameters of the LSTM neural network, the batch size of the LSTM neural network is 1000, the learning rate is 0.01, the training period is 100, and the maximum likelihood estimation method is used to train the GRU model, the seasonal period is 1 millisecond, and the lag order is 1000;

[0055] Step 2.2, after the LSTM and GRU joint prediction model parameters in the flow prediction module are set, the historical flow of the user is input into the model as input to obtain the flow condition of the user in the next two periods, as shown in formula (1):

[0056] Predict final =w1*RES LSTM +W2*RES GRU (1)

[0057] Among them, Predict final is the final prediction result of the joint neural network, w1 and w2 are weight coefficients, RES LSTM and RES GRU are the prediction results of the LSTM model and the GRU model respectively;

[0058] Step 2.3, according to the comparison between the joint prediction result and the actual value of the next period, the incoming flow is classified into different categories: the flow close to the actual value and meeting the expectation is classified as periodic flow, and the flow not meeting the expectation is classified as non-periodic flow;

[0059] According to the predicted value and the actual value, the mean absolute percentage error MAPE Periodic is determined, and the flow with MAPE Periodic less than the set threshold η is classified as periodic flow;

[0060] In the embodiment, the result of predicting periodic flow by using the LSTM and GRU joint neural network is shown in Figure 3 , and the MAPE result of the periodic flow is 0.08%, which can accurately classify the flow category;

[0061] Step 3, according to the classification of the flow in step 2, the bandwidth of the entire network is allocated to different ONU slices, in this period, the bandwidth of the periodic flow slice is allocated according to the prediction result of the flow prediction module, and the non-periodic flow slice uses the remaining bandwidth of the total bandwidth; in the ONU slice of the periodic flow, the vOLT directly allocates the bandwidth to each ONU according to the prediction result of the SDN controller in a polling manner; in the ONU slice of the non-periodic flow, a fixed bandwidth is allocated to each ONU to avoid the problem that user data in the ONU cannot be transmitted for a long time, and the resources allocated to the non-periodic ONU slice in addition to the fixed bandwidth are allocated to each ONU as common bandwidth in a polling manner;

[0062] Step 3.1, according to the classification result of the flow in step 2, the SDN controller uploads the classification result to the OLT to perform inter-slice bandwidth allocation for the next period, and the inter-slice bandwidth allocation is shown in formula (2) and formula (3):

[0063] B P = B total *n p (2)

[0064] B a = B total *n a (3)

[0065] Wherein, B P is the allocated bandwidth of the periodic flow slice, B a is the allocated bandwidth of the non-periodic flow slice, B total is the total upstream bandwidth of the PON network, n p and n a are the normalized coefficients of the statistical flow, which are shown in formula (4) and formula (5) respectively:

[0066]

[0067] n a = 1-n p (5)

[0068] Wherein, Predict p is the prediction result of the periodic flow, Predict a is the prediction result of the non-periodic flow;

[0069] Step 3.2, according to the inter-slice bandwidth allocation in step 3.1, intra-slice bandwidth allocation is performed;

[0070] In the non-periodic traffic slice, each ONU is allocated guaranteed bandwidth to avoid long time transmission of low priority traffic, resulting in a large number of packet loss and reducing network throughput, and the remaining bandwidth is allocated to different ONUs according to priority, and the remaining bandwidth B remaina As shown in formula (6):

[0071] B remaina = B a -m*B f (6)

[0072] Wherein, B f is the guaranteed bandwidth allocated to each non-periodic ONU, and m is the number of non-periodic ONUs;

[0073] The allocated bandwidth B allocai As shown in formula (7):

[0074]

[0075] Wherein, p i is the priority of the data packet;

[0076] In the periodic traffic slice, there is no priority differentiation, and the vOLT polls the ONUs and allocates bandwidth to each ONU according to the bandwidth request of the ONU;

[0077] In the improved TDM-PON architecture described in step 1, the method of the present application is simulated, and the simulation results are shown in Figure 4 Under the condition of fixed load, the method of the present application is superior to the conventional IPACT algorithm in network delay;

[0078] It also includes step 4: according to the allocation bandwidth method of step 3, the information transmission method of public network is analyzed and designed, which can improve the traffic load balancing, resource fair allocation, priority management and elastic adaptation in public network, and improve the robustness and user experience of public network.

[0079] The above specific description further describes the purpose, technical scheme and beneficial effects of the application. It should be understood that the above description is only a specific embodiment of the application and does not limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. A method for dynamic bandwidth allocation in a multi-user multi-service passive optical network, characterized in that: The application improves the architecture and bandwidth allocation method of traditional time division multiplexing passive optical network, provides differentiated allocation method for different types of traffic of users, and improves the bandwidth resource utilization of the whole optical access network, including the following steps: Step 1, based on network function virtualization technology, logically divide the physical OLT of TDM-PON into two virtual OLTs, and divide the ONU connected to the OLT into two slices, use the SDN controller to collect the historical traffic generated by the user, input the historical traffic of the user into the joint neural network model of LSTM and GRU in the bandwidth allocation period, and the neural network model predicts the future traffic of the user; Step 2, according to the traffic prediction result of step 1, compare the actual traffic of the next period of the user, complete the classification of the user traffic, divide the traffic into periodic traffic and non-periodic traffic, and the SDN controller will divide the network traffic of different categories into different ONU slices according to the classification result, and finally upload to the respective vOLT to complete the differentiated bandwidth allocation; Step 3, according to the classification of the traffic in step 2, first allocate the bandwidth of the whole network to different ONU slices, in this period, the bandwidth of the periodic traffic slice is allocated according to the prediction result of the traffic prediction module, and the non-periodic traffic slice uses the remaining bandwidth; in the ONU slice of periodic traffic, the vOLT directly allocates the bandwidth to each ONU according to the prediction result of the SDN controller by using the polling method; in the ONU slice of non-periodic traffic, a fixed bandwidth is allocated to each ONU to avoid the problem that the user data in the ONU cannot be transmitted for a long time, and the resources allocated to the non-periodic ONU slice except the fixed bandwidth are used as common bandwidth and allocated to each ONU by using the polling method.

2. The method of claim 1, wherein the method is a dynamic bandwidth allocation method for a multi-user multi-service passive optical network. Further comprising step 4: according to the bandwidth allocation method of step 3, guide the analysis and design of information transmission method of public network, which can improve the traffic load balancing, resource fair allocation, priority management and elastic adaptation in public network, improve the robustness and user experience of public network.

3. The method of claim 1, wherein the method further comprises: determining a bandwidth request for each of the plurality of users; and allocating the bandwidth to the plurality of users based on the determined bandwidth request. Step 1 includes the following sub-steps: Step 1.1, based on network function virtualization technology, build two virtual OLTs in the physical OLT of traditional TDM-PON architecture: vOLT1 and vOLT2, vOLT1 is used for bandwidth allocation in periodic traffic slice, and vOLT2 is used for bandwidth allocation in non-periodic traffic slice; Step 1.2, divide the ONU connected to the physical OLT of TDM-PON into two slices, which are used to receive periodic traffic and non-periodic traffic of users respectively, and the user equipment is connected to the ONU through wired transmission or wireless transmission; Step 1.3, set the SDN controller on the side of the physical OLT of TDM-PON, collect the historical traffic of the user in the previous N periods through the SDN controller, and set the joint neural network of LSTM and GRU in the SDN controller, and predict the traffic of the user in the next two periods according to the historical traffic of the user.

4. The method of claim 1, wherein the method is a dynamic bandwidth allocation method for a multi-user multi-service passive optical network. Step 2 includes the following sub-steps: Step 2.1, the traffic prediction module of the SDN controller receives the user's historical traffic, sets the parameters of the LSTM neural network, and trains the GRU model using the maximum likelihood estimation method; Step 2.2, after the LSTM and GRU joint prediction model parameters in the traffic prediction module are set, the user's historical traffic is input into the model as input to obtain the user's traffic status in the next two periods, as shown in equation (1): Predict final = w1*RES LSTM + w2*RES GRU (1) wherein Predict final is the final prediction result of the joint neural network, w1 and w2 are weight coefficients, RES LSTM and RES GRU are the prediction results of the LSTM model and the GRU model, respectively; Step 2.3, according to the comparison between the joint prediction result and the actual value of the next period, the incoming traffic is classified into different categories: the traffic close to the actual value and meeting the expectation is classified as periodic traffic, and the traffic not meeting the expectation is classified as non-periodic traffic; determining mean absolute percentage error MAPE from predicted and actual values Periodic and classifying the traffic as periodic traffic if the MAPE Periodic is less than a set threshold η.

5. The method of claim 1, wherein the method further comprises: determining a bandwidth request for each of the plurality of users; and allocating the bandwidth to the plurality of users based on the determined bandwidth request. Step 3 includes the following sub-steps: Step 3.1, according to the traffic classification result of step 2, the SDN controller uploads the classification result to the OLT for inter-slice bandwidth allocation in the next period, and the inter-slice bandwidth allocation is shown in equations (2) and (3): B P = B total *n p (2) B a = B total *n a (3) wherein B P is the allocated bandwidth of the periodic traffic slice, Ba is the allocated bandwidth of the aperiodic traffic slice, B total is the total upstream bandwidth of the PON network, n p and n a are the normalized coefficients of the statistical traffic, as shown in equations (4) and (5), respectively: n a = 1 - n p (5) wherein Predict p is the prediction result of the periodic flow, Predict a is the prediction result of the aperiodic flow; Step 3.2, according to the inter-slice bandwidth allocation of step 3.1, the intra-slice bandwidth allocation is performed; In the non-periodic traffic slice, each ONU is allocated guaranteed bandwidth, avoiding long-time transmission of low-priority traffic, resulting in a large number of packet loss and reducing network throughput. The remaining bandwidth is allocated to different ONUs in a round-robin manner according to priority, and the remaining bandwidth B remaina As shown in formula (6): B remaina = B a - m*B f (6) where B f is the guaranteed bandwidth allocated to each aperiodic ONU, and m is the number of aperiodic ONUs. The allocated bandwidth B of the ONUs in the ith aperiodic community allocai As shown in equation (7): wherein p i is the priority of the data packet; In the periodic traffic slice, there is no priority, the vOLT polls the ONU, and the bandwidth is allocated to each ONU according to the bandwidth request of the ONU.

Images