RU2777035C1 - Method for probabilistic weighted fair queue maintenance and a device implementing it - Google Patents

Abstract

FIELD: communication technology.

SUBSTANCE: invention relates to a method for fair queue maintenance in routers of a transport communication network. The absolute weight for each queue, the rate of receipt of priority data traffic streams and the average packet size for each priority data traffic stream are set. All options (combinations) of active queues are formed. For each option (combination) of active queues, a vector of service intensities (WM) is calculated. A vector of service intensity intervals (IntWM) is formed for each vector of service intensities. Streams of priority data traffic are received to the input ports of the device. Priority data traffic flows are classified into classes according to the ToS Precedence field of the IP packet header, or according to the DSCP field of the IP packet header. The priority data traffic flows are routed to the corresponding output ports of the router based on the routing table. Data traffic packets of the appropriate class are queued in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, the current status of the output port of the device is checked. The presence of packets in queues is checked, if the queues are not empty, a vector of active queues is formed, a vector of service intensity intervals taking into account the vector of active queues is selected. A random number evenly distributed in the range from 0 to 1 is generated. The number of the active queue is selected taking into account a random number and the vector of service intensity intervals (IntWM). The packet is transmitted to the output port of the device, the current state of the output port of the device is updated.

EFFECT: reduction in the complexity of calculations with the support of a large number of classes or service flows.

3 cl, 9 dwg

Description

Technical field

This invention relates to the field of telecommunications and, in particular, to a method of fair queue service in routers of a transport communication network.

State of the art

At present, with the development of the global Internet and the provision of a wide range of multimedia services to users, an important issue is to ensure the quality of service for network traffic (Quality of Service, QoS). In providing quality of service in IP-based transport networks, multiple data traffic flows have different quality of service requirements and accordingly share the resources of a transport network router. Today, the global Internet uses a differentiated service mechanism that allows you to "fairly" distribute the resources of the router of the transport communication network between individual classes of traffic and provide priority service to some of it. Differentiated services divides all traffic into multiple classes based on the service requirements of each class.

A modern packet-switched transport communication network includes many nodes (routers, switches, hosts, etc.) connected by communication channels. Each router has from 1 to n input ports to which incoming communication channels are connected, and from 1 to n output ports to which outgoing communication channels are connected. In this case, the router is a common resource of the transport communication network. There is a known issue with the allocation of a limited amount of shared resources (buffer memory and egress port bandwidth) to competing users, applications, and classes of service. The queue scheduling discipline allows you to control access to a fixed amount of output port bandwidth by choosing the next packet that is transmitted on the port. There are many different queuing disciplines, each trying to strike the right balance between complexity, control, and fairness. The queuing discipline supported by router vendors is implementation dependent, which means there are no real industry standards. However, a vendor sometimes refers to their own implementation as a standard. The difficulty in accepting any implementation as a standard is that each vendor combines features from a number of known common queuing disciplines to provide the common functionality that their customers need to manage egress port congestion.

Difficulties arise when packets arrive at the output port faster than they can be transmitted. This means that the router interface is considered to be overloaded. The delay caused by congestion can range from the time it takes to transmit the last bit of a previously arrived packet to infinity, where the packet is discarded due to lack of space in the memory buffer (queue). It is important to minimize the number of congested links in a network, as congestion reduces network throughput, increases end-to-end packet delays, causes jitter, and can result in packet loss if there is not enough buffer memory in the buffer to hold all the packets waiting to be transmitted. The router module that performs the queuing procedure is called the scheduler.

The first routers used a first-in-first-out (FIFO) scheduler on each egress port. The FIFO maintains one queue per output port. When a new packet arrives, it is placed at the end of the queue, and as long as the queue is not empty, the router forwards the packets from the queue, accepting the packet from the front of the queue (i.e., accepting the "oldest" packet). Thus, FIFO provides a simple best effort service for all applications. FIFO has a number of serious disadvantages. First, a "greedy source" that sends packets at an extremely high bit rate can preempt other sources and gain more bandwidth than other sources. Second, if several shorter packets are behind very long packets, then the FIFO scheme results in a larger average delay (per packet) than if the shorter packets were transmitted before the longer packet. Thirdly, according to the FIFO scheme, there are no mechanisms to ensure the quality of service of data traffic from high-priority sources, such as delay-sensitive (multimedia traffic).

A priority queuing (PQ) scheduler can be used to provide delay guarantees. In a classic PQ scheduler, packets are first classified by the system and then placed into different priority queues. Packets are only taken out of the queue for service if all higher priority queues are empty. In each priority queue, packets are served in FIFO order. The advantage of this scheduler is the ease of implementation and the ability to provide low latency for high priority data traffic. The disadvantage is the inability to exclude the influence of threads on each other. If high-priority data traffic is constantly entering the high-density queue, then data traffic entering other queues will be blocked.

To eliminate the shortcomings of the FIFO and PQ schedulers, a queuing discipline was developed - weighted fair queue service (Weighted Fair Queuing, WFQ). WFQ maintains fair bandwidth allocation for variable length packets by using an approximation of Generalized Processor Sharing (GPS). GPS is an ideal scheduler that provides each stream with a guaranteed bit rate (bandwidth) and fairly distributes excess bandwidth between streams according to their relative bandwidth weights (see E.A. Kucheryavy, Traffic Management and Network Quality of Service Internet, St. Petersburg Science and Technology, 2004, 152 p.). As a result, GPS can provide end-to-end delays and fairness guarantees for priority data traffic flows as long as the flows do not exceed the allocated share of the uplink bandwidth. Unfortunately, the GPS discipline is not practical because it services bits at a time (clock). The real scheduler has to finish servicing the entire packet from the queue before it can move on to the next queue.

A serious problem with WFQ is that GPS simulations can predict that many packets may end up with the same virtual end time. Updating the virtual time function between two successive packet arrivals can be computationally intensive. In particular, if N active sessions (data traffic streams) are simultaneously transmitted in the communication channel, then updating the virtual time may entail computing O(N) sessions or queues. The number of active threads can be very large. For example, there may be tens of thousands or even hundreds of thousands of queues using a single communication channel. This results in a proportionately large number of calculations. This problem is called iterative deletion.

Due to the high complexity associated with modeling a GPS system, various solutions to this problem have been proposed recently. Many methods have been proposed to simplify the calculation of virtual time. Some of these methods are Self-Clocked Fair Queuing (SCFQ), Virtual Clock (VC), Start-time Fair Queuing (SFQ), Frame-Based Fair Queuing. Fair Queuing, FFQ), Minimum-Delay Self-Clocked Fair Queuing (MD-SCFQ), and Discrete-Rate (DR) scheduling. In general, such simplifying approaches lead to a decrease in fairness (flow isolation), or to an increase in the delay margin, or do not solve the problem of repeated deletion.

A device "Weighted Fair Service Scheduler" is known (patent US 7461159 B2 dated December 2, 2008), which uses GPS modeling to determine the order of servicing objects, uses a new dynamic data structure with a complex but simple mechanism for updating pointers. The preferred scheduler options perform a fixed amount of work per scheduling event. The scheduling event can either be a calculation of a new virtual completion timestamp on new arrivals to the scheduler, or a determination of which objects should leave the GPS system because their completion timestamp has expired. Such a scheduler can be used when scheduling packets in a packet processing device such as a router, scheduling access of software processes to a computer processor, or the like. The scheduler may implement a weighted fair queuing (WFQ).

The known method "Fair queue service using dynamic weights (DWFQ)" (patent US 7023866 B2 dated 4.04.2006) includes the following steps: requesting incoming packets in the queues associated with the respective classes of service; transmitting outgoing packets from said queues in accordance with a schedule determined by weights associated with service classes used by the queue scheduler; and real-time dynamic weighting that causes the queue size for each service class to move towards a target value that satisfies the transmission delay associated with the service class.

It is known the device "Shaper of a common weighted fair queue (WFQ)" (patent US 8638664 B2 dated January 28, 2014), which includes a port, a buffer, a flow control module and a service differentiation module. The port is configured to send and receive a packet, and the port is connected to a network entity. The buffer is configured to hold the packet. The flow control module is configured to control the transmission of a packet within the network device. The service differentiation module is connected to the buffer and the flow control module. The service differentiation module is configured to regulate the storage of the packet in the buffer and to regulate the transmission of the packet from the network device to the network entity. The service differentiation module is also configured to determine the excess bandwidth available within the network device and allocate the excess bandwidth to transmit the packet to the network entity.

The closest in technical essence to the claimed method and selected as a prototype is the "Method and device for performing weighted queue scheduling using a set of fairness coefficients" (patent US 10291540 B2 dated May 14, 2019), which consists in receiving packet data from multiple source ports; classifying the packet data based on multiple source ports and their associated destination ports; sorting packet data into multiple queues in a buffer; scheduling each of the plurality of queues for transmission based on the time weight of each of the plurality of queues, which is calculated based on a set of fairness factors, the set of fairness factors including: a validity bit of the corresponding queue, a priority weight of the corresponding queue, a first position identification indicating the position of the corresponding queue, and identifying a second position indicating the position of the last scheduled queue with the highest priority from the previous round, scheduling further includes selecting from at least two queues with the same maximum priority weight, with an additional amount added to the time weight of the corresponding queue if the identification the first position indicates a lower position than the identification of the second position, and while the additional amount is equal to the number of queues selected from at least two queues with the same maximum priority weight; and updating the static state of the respective queue of the plurality of queues responsive to the data packet being input to or output from the respective queue; sending the batch output dequeued from the scheduler to the destination.

The closest in technical essence to the claimed device and selected as a prototype is "A system for performing weighted scheduling of queues using a set of fairness coefficients" (patent US 10291540 B2 dated May 14, 2019), containing: at least one processor; at least one tangible non-transitive computer-readable storage medium storing instructions that, when executed by at least one processor, cause the system to perform a method of using a scheduler to process requests, including: receiving packet data from multiple source ports; classifying packet data based on multiple source ports and destination ports associated with them; sorting packet data into multiple queues in a buffer; scheduling each of the plurality of queues for transmission based on the time weight of each of the plurality of queues, which is calculated based on a set of fairness factors, the set of fairness factors including: a validity bit of the corresponding queue, a priority weight of the corresponding queue, a first position identification indicating the position of the corresponding queue, and identifying a second position indicating the position of the last scheduled queue with the highest priority from the previous round, scheduling further includes selecting from at least two queues with the same maximum priority weight, and wherein the time weight of the corresponding queue is increased by an additional amount if the identification the first position indicates a lower position than the identification of the second position, and while the additional amount is equal to the number of queues selected from at least two queues with the same maximum priority weight; and updating the static state of the respective queue of the plurality of queues responsive to the data packet being input to or output from the respective queue; and sending the output packet data dequeued from the scheduler to the destination.

A technical problem is the high computational complexity when supporting a large number of classes or service flows. Another technical problem is the low accuracy of the Generalized Processor Sharing System (GPS) approximation.

The technical result is to reduce the complexity of calculations while supporting a large number of classes of service due to the fact that from 1 to n modules of probabilistic weighted fair queuing are additionally included, each including a module for calculating the intensity vector, a module for calculating the vector of intervals, a package selection module, a random number generator , a module for generating combinations of active queues.

Another technical result is that the present invention allows a more accurate approximation of the Generalized Processor Sharing System (GPS). This technical result is achieved due to the fact that at the first stage, the absolute weight for each queue, the arrival rate of priority data traffic flows and the average packet size for each priority data traffic flow are set, all options (combinations) of active queues are formed, calculated for each option ( combinations) of active queues, the service intensity vector (WM), form the service intensity interval vector (IntWM) for each service intensity vector, at the second stage, check the current state of the output port of the device, check the presence of packets in the queues, if the queues are not empty, then form the vector of active queues, select the vector of service intensity intervals taking into account the vector of active queues, generate a random number uniformly distributed in the interval from 0 to 1, select the number of the active queue, taking into account the random number and the vector of service intensity intervals (IntWM), eating a packet to the output port of the device, updating the current state of the output port of the device.

The creation of a method for probabilistic weighted fair queuing and a device that implements it are aimed at solving this technical problem, allowing to reduce the complexity of calculations while supporting a large number of classes of service and more accurately approximate the generalized processor sharing system (GPS).

Disclosure of invention

In the claimed method, this technical problem is solved by the fact that in the method of weighted probabilistic fair queuing, which consists in the fact that at the first stage, the absolute weight for each queue, the arrival rate of priority data traffic flows and the average packet size for each priority data traffic flow, all options (combinations) of active queues are formed, the service intensity vector (WM) is calculated for each option (combination) of active queues, the service intensity interval vector (IntWM) is formed for each service intensity vector, at the second stage priority data traffic flows are received at the input ports devices, classify priority data traffic flows into classes in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, route priority data traffic flows to the corresponding output ports of the device on the OS routing table, queue data traffic packets of the corresponding class in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, check the current state of the device's output port, check the presence of packets in the queues, if the queues are not empty, then a vector of active queues is formed, a vector of service intensity intervals is selected taking into account the vector of active queues, a random number uniformly distributed in the interval from 0 to 1 is generated, an active queue number is selected taking into account the random number and the vector of service intensity intervals (IntWM), the packet is transmitted to the output port of the device, update the current state of the output port of the device.

A new set of essential features allows to achieve the specified technical result due to the fact that at the first stage, the absolute weight is set for each queue, the rate of arrival of priority data traffic flows and the average packet size for each priority data traffic flow, all options (combinations) of active queues are formed, for each option (combination) of active queues, the service intensity vector (WM) is calculated, the service intensity interval vector (IntWM) is formed for each service intensity vector, at the second stage, the current state of the device output port is checked, and the presence of packets in the queues is checked if the queues are not empty , then a vector of active queues is formed, a vector of service intensity intervals is selected taking into account the vector of active queues, a random number is generated uniformly distributed in the interval from 0 to 1, the active queue number is selected taking into account the random number and vector service rate intervals (IntWM), send a packet to the output port of the device, update the current state of the output port of the device.

In order to achieve the above object, in accordance with another aspect of the present invention, a probabilistic weighted fair queuing apparatus is provided.

In the claimed device, this problem is solved by the fact that a device for probabilistic priority queuing, containing from 1 to n input ports for receiving priority data traffic flows, a classifier configured to classify data traffic flows in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP header of the packet, a routing module configured to forward the packet to the output port in accordance with the routing table, from 1 to n queue modules for placing packets in the queue of the corresponding class in accordance with the ToS Precedence field of the IP header of the packet, or according to the DSCP field of the IP packet header; measuring the rate of data traffic arrival and the average length of packets of the corresponding class, from 1 to n output ports for transmitting priority data traffic flows to the communication channel, characterized in that from 1 to n modules of probabilistic weighted fair queuing are additionally included, moreover, priority data traffic flows arrive at the input ports, then from the input ports the priority data traffic flows are transmitted to the input of the classifier, the output of the classifier is connected to the input of the routing module, from the output of the routing module the flows of priority data traffic arrive at the input of the corresponding queue module, from the output of the queue module the priority data traffic flows are transmitted to the corresponding probabilistic weighted fair queuing module, from the output of the probabilistic weighted fair queuing module, priority data traffic flows arrive at the output ports.

According to one particular implementation, the probabilistic weighted fair queuing module comprises a rate vector calculation module configured to calculate a service rate vector (WM), configure parameters of at least an absolute weight for each queue, an arrival rate of priority data traffic flows, and an average size packets for each priority data traffic flow, an interval vector calculation module configured to generate a service rate interval vector (IntWM) for each service rate vector, a packet selection module configured to forward packets from a plurality of queues of the corresponding class to an output port, check and updating the current state of the router's output port, checking for the presence of packets in queues, selecting a vector of service intervals based on the vector of active queues, selecting an active queue number based on m of a random number and a vector of service intensity intervals (IntWM), a random number generator configured to generate a random number uniformly distributed in the interval from 0 to 1, a module for generating combinations of active queues, configured to generate all combinations of active queues, while all modules in the device are connected via a common internal AXI4 bus, which acts as a means of information exchange between modules.

All modules can be implemented according to a well-known scheme, for example, the Eltex ME 5000 router (see https://eltex-co.ru/catalog/provider_edge_router/marshrutizator_me5000).

A new set of essential features allows you to achieve the specified technical result due to additionally introduced from 1 to n modules of probabilistic weighted fair queuing, while each module of probabilistic weighted fair queuing contains a module for calculating the intensity vector, a module for calculating the vector of intervals, a package selection module, a random generator numbers, a module for generating combinations of active queues.

The analysis of the prior art made it possible to establish that there are no analogues characterized by a set of features that are identical to all the features of the claimed method of probabilistic priority queue service and a device that implements it. Therefore, each of the claimed inventions meets the condition of patentability "novelty".

The results of the search for known solutions in this and related fields of technology in order to identify features that match the distinguishing features of the prototype of the claimed object showed that they do not follow explicitly from the prior art. From the prior art, the influence of the transformations provided for by the essential features of the claimed invention on the achievement of the specified technical result has not been revealed either. Therefore, each of the claimed inventions meets the condition of patentability "inventive step".

"Industrial applicability" is due to the presence of an elementary base, on the basis of which devices can be made that implement this method and device with the achievement of the result specified in the invention.

To more clearly illustrate the technical solutions according to the embodiments of the present invention, a brief description of the accompanying drawings is given below.

In FIG. 1 is a diagram of the operation of the probabilistic weighted fair queuing method;

in fig. 2 - device for probabilistic weighted fair queuing;

in fig. 3 is a diagram of a module for probabilistic weighted fair queuing;

in fig. 4 - a fragment of the transport communication network in the environment of the network simulator OMNet++;

in fig. 5 shows simulation results, average packet latency in WFQ, GPS schedulers, and the present invention;

in fig. 6 shows simulation results, average latency of first priority packets in WFQ, GPS schedulers, and the present invention;

in fig. 7 shows simulation results, average latency of second priority packets in WFQ, GPS schedulers, and the present invention;

in fig. 8 shows simulation results, average latency of third priority packets in WFQ, GPS schedulers, and the present invention.

The technical solutions for the embodiments of the present invention will be fully and clearly described below with reference to the accompanying drawings. It should be obvious that the embodiments described below are only a part of the present invention and not all possible embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention and not using creativity fall within the protection scope of the present invention.

The implementation of the claimed method is as follows (Fig. 1).

First stage.

101. Set the absolute weight for each queue, the arrival rate of the priority data traffic streams, and the average packet size for each priority data traffic stream.

102. Form all variants (combinations) of active queues.

103. For each option (combination) of active queues, a service intensity vector (WM) is calculated.

104. A service rate interval vector (IntWM) is generated for each service rate vector.

Second phase.

105. Receive streams of priority data traffic to the input ports of the router.

106. Classify priority data traffic flows into classes in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header.

107. Route priority data traffic flows to the corresponding output ports of the router based on the routing table.

108. The data traffic packets of the corresponding class are queued in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header.

109. Check the current state of the output port of the router.

110. Check the presence of packets in the queues, if the queues are not empty, then form a vector of active queues.

111. A vector of intervals of service rates is selected taking into account the vector of active queues.

112. Generate a random number uniformly distributed in the range from 0 to 1.

113. The number of the active queue is selected, taking into account a random number and a vector of service intensity intervals (IntWM).

114. Send the packet to the output port of the router.

115. Update the current state of the output port of the router.

The claimed method of probabilistic weighted fair queuing allows to achieve the specified technical result by eliminating discrimination of low-priority data traffic, as well as reducing the average network delay of packets for low-priority traffic due to the fact that at the first stage, the absolute weight for each queue, the rate of arrival of priority data traffic flows and the average packet size for each flow of priority data traffic, all options (combinations) of active queues are formed, the service intensity vector (WM) is calculated for each option (combination) of active queues, the service intensity interval vector (IntWM) is formed for each service intensity vector, on the second stage, they check the current state of the output port of the router, check the presence of packets in the queues, if the queues are not empty, then form the vector of active queues, select the vector of service intensity intervals based on the vector of active queues, generate a random number uniformly distributed in the range from 0 to 1, choose the number of the active queue based on the random number and the vector of service intervals (IntWM), send the packet to the router output port, update the current state of the router output port .

The probabilistic weighted fair queuing device (FIG. 2) consists of 1 to n input ports, a classifier 21, a routing module 22, 1 to n queue modules 23, 1 to n probabilistic weighted fair queuing modules 24, from 1 to n output ports, while the priority data traffic streams arrive at the input ports, then from the input ports the priority data traffic streams are transmitted to the input of the classifier, the output of the classifier is connected to the input of the routing module, from the output of the routing module the streams of priority data traffic arrive at the input of the corresponding queue module , from the output of the queuing module, the priority data traffic flows are transferred to the corresponding module of the probabilistic weighted fair queuing, from the output of the probabilistic weighted fair queuing module, the priority data traffic flows arrive at the output ports.

Input ports from 1 to n are configured to receive streams of priority data traffic.

Classifier 21 is configured to classify data traffic flows according to the ToS Precedence field of the IP packet header, or according to the DSCP field of the IP packet header.

Routing module 22 configured to forward the packet to the output port in accordance with the routing table.

Queue modules 23 (from 1 to n) are configured to place packets in a queue of the appropriate class in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header; measuring the rate of data traffic arrival and the average length of packets of the corresponding class.

Output ports from 1 to n are configured to transfer priority data traffic streams to the communication channel.

Probabilistic weighted fair queuing modules 24 (from 1 to n), each of which contains a module for calculating the intensity vector 24.1, a module for calculating the vector of intervals 24.2, a packet selection module 24.3, a random number generator 24.4, a module for generating combinations of active queues 24.5, while all modules in the device are connected via a common internal bus AXI4, which acts as a means of information exchange between modules.

A rate vector calculation module 24.1 configured to calculate a service rate vector (WM), parameter configuration of at least an absolute weight for each queue, an arrival rate of priority data traffic streams, and an average packet size for each priority data traffic stream.

An interval vector calculation module 24.2, configured to generate a service rate interval vector (IntWM) for each service rate vector.

A packet selection module configured to forward packets from a plurality of queues of the corresponding class to the output port, check and update the current state of the router's output port, check for the presence of packets in queues, select the vector of service intensity intervals taking into account the vector of active queues, select the number of the active queue taking into account a random number and a vector of service intensity intervals (IntWM).

Random number generator 24.4 configured to generate a random number uniformly distributed between 0 and 1.

A module for generating combinations of active queues, configured to generate all combinations of active queues.

Priority recovery module 27.7, configured to restore the original priority in the DSCP field of the IP packet header, or in the ToS Precedence field of the IP packet header, taking into account the lookup table.

All modules can be implemented according to a well-known scheme, for example, the Eltex ME 5000 router (see https://eltex-co.ru/catalog/provider_edge_router/marshrutizator_me5000).

The device works as follows.

At the first stage, the probabilistic weighted fair queuing module 24 is configured. In the intensity vector calculation module 24.1, the absolute weight for each queue, the arrival rate of priority data traffic streams, and the average packet size for each priority data traffic stream are set. Based on these data, all variants (combinations) of active queues are formed in the module for generating combinations of active queues. In the rate vector calculation module 24.1, a service rate vector (WM) is calculated for each option (combination) of active queues. In the interval vector calculation module 24.2, a service rate interval vector (IntWM) is generated for each service rate vector.

At the second stage, streams of priority data traffic are received at the input ports of the router and then transferred to the classifier. Classifier 21 classifies priority data traffic flows into classes according to the ToS Precedence field of the IP packet header, or according to the DSCP field of the IP packet header. After classifier 21, priority data traffic flows enter the routing module, where packets are promoted in accordance with the routing table to the corresponding queue module. In the queue module 23, the data traffic packets of the corresponding class are queued in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header. The packet selection module 24.3 checks the current state of the output port of the router and checks for the presence of packets in the queues of the corresponding queue module 23; active queues. At the same time, a random number evenly distributed in the range from 0 to 1 is generated in the random number generator 24.4. Taking into account the random number and the service intensity interval vector (IntWM), the active queue number of the corresponding queue module 23 is selected in the packet selection module and the packet is transmitted to the output router port. After that, in the package selection module 24.3, the current state of the output port of the router is updated.

The interactive process of the device operation should be considered with an example. Suppose a device receives three streams of priority data traffic on different input ports that need to be transferred to one output port. The packet arrival rate from the first source is 30 Kbps, from the second source - 21 Kbps, from the third source - 14.7 Kbps. The average packet size from all sources is 500 bytes. The output port has a bandwidth of 31.4 Kbps.

, вес второго потока трафика данных

, вес третьего потока трафика данных

. На основе абсолютных весов для каждой очереди, скорости поступления потоков приоритетного трафика данных и среднего размера пакетов для каждого потока приоритетного трафика данных в модуле формирования активных очередей формируют варианты (комбинации) активных очередей. Для представленных потоков приоритетного трафика данных получаются следующие варианты (комбинации) активных очередей -

At the first stage, the module of probabilistic weighted fair queuing 24 is configured in the device at the output port.

, the weight of the second data traffic flow

, the weight of the third data traffic flow

. Based on the absolute weights for each queue, the arrival rate of priority data traffic streams, and the average packet size for each priority data traffic stream, variants (combinations) of active queues are formed in the active queuing module. For the presented flows of priority data traffic, the following variants (combinations) of active queues are obtained -

In the intensity vector calculation module 24.1, a service intensity vector (WM) is calculated for each option (combination) of active queues, which is presented in Table 1.

Table 1 - Service rate vector (WM)

Variants (combinations) of active queues Service rate vector (WM)

In the interval vector calculation module 24.2, for each service intensity vector, a service intensity interval vector (IntWM) is generated, shown in Table 2.

Table 2 - Service rate interval vector (IntWM)

Service rate vector (WM) Service rate interval vector (IntWM)

After the probabilistic weighted fair queuing module 24 is configured, 3 streams of priority data traffic arrive at the input ports of the device. The classifier assigns data traffic flows to the corresponding classes based on the value of the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header. In the routing module 22, all data traffic flows are redirected to the output port, which is connected by a communication channel to the opposite router. From the routing module 22, priority data traffic packets enter the queue module 23, which places the packets in the queue of the appropriate class based on the values in the ToS Precedence field of the IP packet header, or in the DSCP field of the IP packet header. The package selection module 24.3 checks for the presence of packets in the queues of the queue module 23, if the queues are not empty, then a vector of active queues is formed in the module for generating combinations of active queues 24.5. Next, in the package selection module 24.3, a vector of intervals of service rates is selected taking into account the vector of active queues (table 2). At the same time, a random number is generated in the random number generator 24.4, uniformly distributed in the range from 0 to 1.

In the packet selection module 24.3, the active queue number is selected, taking into account a random number and a vector of service intervals, and the packet is transmitted to the output port of the router. After serving the packet, the packet selection module 24.3 updates the current state of the output port of the device.

Thanks to a new set of essential features, the specified technical result is achieved by additionally introduced from 1 to n modules of probabilistic weighted fair queuing 24, each module of probabilistic weighted fair queuing 24 contains a module for calculating the intensity vector 24.1, a module for calculating the vector of intervals 24.2, a selection module package 24.3, random number generator 24.4, module for generating combinations of active queues 24.5.

The validity of the theoretical premises was tested on a fragment of the transport communication network in the OMNet++ network simulator environment with the following initial data (Fig. 4):

- number of routers in the transport communication network

- number of information sources

Кбит/с;- packet arrival rate of the first data traffic flow

Kbps;

Кбит/с;- the rate of arrival of packets of the second data traffic flow

Kbps;

Кбит/с;- packet arrival rate of the third data traffic stream

Kbps;

бит;- average packet length of all data traffic flows

bit;

Кбит/с.- bandwidth of the communication channel

kbps.

Analysis of the simulation results (Fig. 5-8) shows that the average waiting time for packets in the GPS scheduler queue and in the proposed group of inventions practically do not differ in the entire range of the communication channel utilization factor. The values of the average waiting time of packets in the queue of the WFQ scheduler are lower than the values of the average waiting time of packets in the queue of the GPS scheduler in the range of the communication channel utilization factor from 0.6 to 1. Thus, the proposed group of inventions allows to more accurately approximate the generalized processor sharing system (GPS) .

The computational complexity of the WFQ scheduler is O(N), where N is the number of queues (see E. A. Kucheryavy, traffic control and quality of service on the Internet. - St. Petersburg. Science and Technology, 2004. - 187 p.). In the proposed group of inventions, the computational complexity is reduced due to the fact that, at the first stage, the absolute weight for each queue, the arrival rate of priority data traffic flows and the average packet size for each priority data traffic flow are set, all options (combinations) of active queues are formed, calculated for of each option (combination) of active queues, the service intensity vector (WM), form the service intensity interval vector (IntWM) for each service intensity vector, at the second stage receive priority data traffic flows to the input ports of the device, classify the priority data traffic flows into classes according to with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, route priority data traffic flows to the corresponding output ports of the router based on the routing table, queue data traffic packets x of the corresponding class in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, check the current state of the output port of the router, check the presence of packets in the queues, if the queues are not empty, then form a vector of active queues, select the vector intervals of service rates, taking into account the vector of active queues, generate a random number uniformly distributed in the interval from 0 to 1, select the number of the active queue, taking into account the random number and the vector of service rate intervals (IntWM), transmit the packet to the output port of the device, update the current state of the output device port.

то есть в трех очередях находятся пакеты соответствующего класса, вектор интенсивностей обслуживания WM равен

и вектор интервалов интенсивностей обслуживания IntWM равен

В случае, если выходной порт устройства свободен, то выбирают номер активной очереди с учетом случайного числа и вектора интервалов интенсивностей обслуживания (IntWM). Если случайное число попало в диапазон от 0 до 0,51, то передают пакет в выходной порт устройства из очереди 1 класса. Если случайное число попало в диапазон от 0,51 до 0,816, то передают пакет в выходной порт устройства из очереди 2 класса. Если случайное число попало в диапазон от 0,816 до 1, то передают пакет в выходной порт устройства из очереди 3 класса.Figure 9A) shows that for a combination of active queues

that is, there are packets of the corresponding class in three queues, the service rate vector WM is equal to

and the service rate interval vector IntWM is equal to

If the output port of the device is free, then the number of the active queue is selected taking into account a random number and a vector of service intensity intervals (IntWM). If the random number is in the range from 0 to 0.51, then the packet is transmitted to the output port of the device from the class 1 queue. If the random number is in the range from 0.51 to 0.816, then the packet is transmitted to the output port of the device from the class 2 queue. If the random number is in the range from 0.816 to 1, then the packet is transmitted to the output port of the device from the class 3 queue.

то есть в трех очередях находятся пакеты соответствующего класса, вектор интенсивностей обслуживания WM равен

и вектор интервалов интенсивностей обслуживания IntWM равен

В случае, если выходной порт устройства свободен, то выбирают номер активной очереди с учетом случайного числа и вектора интервалов интенсивностей обслуживания (IntWM). Если случайное число попало в диапазон от 0 до 1, то передают пакет в выходной порт устройства из очереди 1 класса. Ввиду того, что диапазон попадания случайного числа для очередей 2 и 3 класса совпадает с диапазоном попадания случайного числа для очереди 1 класса, то они не рассматриваются.Figure 9 B) shows that for a combination of active queues

that is, there are packets of the corresponding class in three queues, the service rate vector WM is equal to

and the service rate interval vector IntWM is equal to

If the output port of the device is free, then the number of the active queue is selected taking into account a random number and a vector of service intensity intervals (IntWM). If the random number is in the range from 0 to 1, then the packet is transmitted to the output port of the device from the class 1 queue. Since the hit range of a random number for class 2 and 3 queues is the same as the hit range of a random number for class 1 queue, they are not considered.

то есть в трех очередях находятся пакеты соответствующего класса, вектор интенсивностей обслуживания WM равен

и вектор интервалов интенсивностей обслуживания IntWM равен

В случае, если выходной порт устройства свободен, то выбирают номер активной очереди с учетом случайного числа и вектора интервалов интенсивностей обслуживания (IntWM). Если случайное число попало в диапазон от 0 до 0,735, то передают пакет в выходной порт устройства из очереди 1 класса. Если случайное число попало в диапазон от 0,735 до 1, то передают пакет в выходной порт устройства из очереди 3 класса. Ввиду того, что для очереди 2 класса диапазон попадания случайного числа совпадает с диапазоном попадания случайного числа для очереди 1 класса, то данная очередь не рассматриваются.Figure 9B) shows that for a combination of active queues

that is, there are packets of the corresponding class in three queues, the service rate vector WM is equal to

and the service rate interval vector IntWM is equal to

If the output port of the device is free, then the number of the active queue is selected taking into account a random number and a vector of service intensity intervals (IntWM). If the random number is in the range from 0 to 0.735, then the packet is transmitted to the output port of the device from the class 1 queue. If the random number is in the range from 0.735 to 1, then the packet is sent to the output port of the device from the class 3 queue. Due to the fact that for the class 2 queue, the hit range of the random number coincides with the hit range of the random number for the class 1 queue, this queue is not considered.

за счет того, что необходимо рассматривать не все очереди, а только часть из них в зависимости от номера активной очереди, случайного числа и вектора интервалов интенсивностей обслуживания (IntWM).Therefore, for the considered group of inventions, the computational complexity decreases and is equal to

due to the fact that it is necessary to consider not all queues, but only a part of them, depending on the number of the active queue, a random number and the vector of service intensity intervals (IntWM).

The claimed method of probabilistic weighted fair queuing and the device that implements it provide a reduction in the complexity of calculations while supporting a large number of classes or service flows, and also allows you to more accurately approximate the generalized processor sharing system (GPS) due to the fact that at the first stage the absolute weight is set for each queue, the rate of arrival of priority data traffic flows and the average packet size for each priority data traffic flow in the intensity vector calculation module 24.1, all options (combinations) of active queues are formed in the module for generating combinations of active queues 24.5, calculated for each option (combination) active queues service rate vector (WM) in the rate vector calculation module 24.1, form a service rate interval vector (IntWM) for each service rate vector in the vector calculation module intervals 24.2, at the second stage they receive priority data traffic flows to the input ports of the router, classify priority data traffic flows into classes in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header in classifier 21, route the priority traffic flows data to the corresponding output ports of the router based on the routing table in the routing module 22, queue the data traffic packets of the corresponding class in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header in the queue module 23, check the current state output port of the router, check the presence of packets in the queues in the packet selection module 24.3, if the queues are not empty, then form the vector of active queues in the module for generating combinations of active queues 24.5, select the vector of service intensity intervals taking into account the active vector of queues in the queue module 23, generate a random number uniformly distributed in the interval from 0 to 1, in the random number generator 24.4, select the number of the active queue in the package selection module 24.3, taking into account the random number and the vector of service intensity intervals (IntWM), transmit the packet to the output port of the router from the package selection module 24.3, update the current state of the output port of the router in the package selection module 24.3.

Claims (3)

1. A probabilistic weighted fair queuing device containing from 1 to n input ports for receiving priority data traffic streams, a classifier configured to classify data traffic streams into classes in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field header of the IP packet, a routing module configured to forward the packet to the output port in accordance with the routing table, from 1 to n queue modules for placing packets in the queue of the appropriate class in accordance with the ToS Precedence field of the IP header of the packet, or in accordance with the DSCP field IP packet header measuring the rate of data traffic arrival and the average length of packets of the corresponding class, from 1 to n output ports for transmitting priority data traffic flows to the communication channel, characterized in that from 1 to n modules of probabilistic weighted fair queuing are additionally included, moreover, priority data traffic flows arrive at the input ports, then from the input ports the priority data traffic flows are transmitted to the input of the classifier, the output of the classifier is connected to the input of the routing module, from the output of the routing module the flows of priority data traffic arrive at the input of the corresponding queue module, from the output of the queue module the priority data traffic flows are transmitted to the corresponding module of the probabilistic weighted fair queuing, from the output of the module of the probabilistic weighted fair queuing, priority data traffic flows arrive at the output ports. 2. The device according to claim 1, in which the probabilistic weighted fair queuing module contains a rate vector calculation module configured to calculate the service rate vector (WM), parameter configuration, at least the absolute weight for each queue, the rate of arrival of priority traffic flows data and average packet size for each priority data traffic flow, an interval vector calculation module configured to generate a service rate interval vector (IntWM) for each service rate vector, a packet selection module configured to promote packets from a plurality of queues of the corresponding class to the output port, checking and updating the current state of the output port of the router, checking the presence of packets in queues, choosing a vector of service intensity intervals, taking into account the vector of active queues, choosing an active queue number, taking into account a random number service intensity interval vector (IntWM), a random number generator configured to generate a random number uniformly distributed in the interval from 0 to 1, a module for generating combinations of active queues, configured to generate all combinations of active queues, while all modules in device are connected via a common internal AXI4 bus, which acts as a means of information exchange between modules. 3. The method of probabilistic weighted fair queuing performed by the device according to claim 1, which consists in setting the absolute weight for each queue, the arrival rate of priority data traffic flows and the average packet size for each priority data traffic flow at the first stage, form all variants (combinations) of active queues, for each variant (combination) of active queues, a service intensity vector (WM) is calculated, a service intensity interval vector (IntWM) is formed for each service intensity vector, at the second stage priority data traffic flows are received at the input ports of the device, classify the priority data traffic flows into classes in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, route the priority data traffic flows to the corresponding output ports of the router based on the routing table tations, queue data traffic packets of the corresponding class in accordance with the ToS Precedence field of the IP packet header, or in accordance with the DSCP field of the IP packet header, check the current state of the device output port, check the presence of packets in the queues, if the queues are not empty, then form the vector of active queues, select the vector of service intensity intervals taking into account the vector of active queues, generate a random number uniformly distributed in the interval from 0 to 1, select the number of the active queue taking into account the random number and the vector of service intensity intervals (IntWM), transmit the packet to the output device port update the current state of the device output port.

Images