A deterministic network congestion control method based on priority queues

Abstract

A deterministic network congestion control method based on priority queues divides nodes in the network topology into two groups: non-preemptible nodes and preemptible nodes. Each node is assigned a priority category. A first-in-first-out (FIFO) priority queue is established for each priority category, and data packets are placed into the corresponding queue according to their priority category. When a data packet arrives at the server, it is checked whether it belongs to a non-preemptible or preemptible node. If the data packet belongs to a non-preemptible node, it is placed in queue Q1; if it belongs to a preemptible node, it is placed in queue Q2. The data packet is then pushed out through the link server, thereby preventing node buffer overflow. This invention implements different priority queues for data packets in different priority groups to distinguish traffic types and prevent header blocking. By implementing cautious preemption / non-preemption rules, low-priority data packets are protected from interruption by high-priority data packets during non-preemption periods.

Description

Technical Field

[0001] This invention belongs to the field of network information transmission technology, specifically congestion control, and relates to a deterministic network congestion control method based on priority queues. Background Technology

[0002] With the development of the internet, the challenge of ever-increasing network traffic and user demands is inevitable. With the widespread adoption of the internet and the rapid development of various network applications, network congestion has become an extremely pressing issue. Network congestion typically refers to excessive data traffic on network links, leading to degraded network performance, increased latency, and even service unavailability. This situation not only inconveniences end users but also poses significant challenges to network service providers and operators.

[0003] To address this problem, researchers and engineers have been working to develop various network congestion control algorithms and technologies. Over the past few decades, many classic congestion control methods have emerged, such as the TCP (Transmission Control Protocol) congestion control mechanism. However, with the continuous increase in network size and complexity, traditional congestion control methods have revealed some limitations and cannot adequately meet the needs of modern networks. Therefore, researchers have begun to focus on deterministic network congestion control, aiming to better address network congestion problems and improve network performance and user experience through precise scheduling and control mechanisms.

[0004] Against this backdrop, this invention aims to delve into the relevant technologies of deterministic network congestion control and explore their potential advantages and applicability in addressing modern network challenges. Through a comprehensive review and analysis of existing technologies, it is undeniable that most specific networks share the following characteristics: (1) they are platforms hosting different types of applications on the Internet, each with different real-time requirements; and (2) due to their varying real-time requirements, different traffic types should have different priorities. Therefore, most studies have implemented different priority models / methods / algorithms to assign different priorities to different traffic types to prevent / improve collisions / congestion. However, a major drawback of the aforementioned review is that, during congestion, each node transmits in a selfish manner, without any rules describing how data packets are transmitted. Furthermore, no priority queue algorithm has yet implemented a priority queue scheduling rule with preemptive / non-preemptive properties to address network congestion. Summary of the Invention

[0005] The purpose of this invention is to overcome the above-mentioned shortcomings of the prior art and to provide a deterministic network congestion control method based on priority queues to address the problem of deterministic network congestion control.

[0006] The technical solution for implementing this invention is: to control congestion in the node buffer using a priority-based preemptible / non-preemptible queue model, including the following specific steps:

[0007] Step 1: First, divide the nodes in the network topology into two groups: non-preemptible nodes NPN and preemptible nodes PN. Non-preemptible nodes have applications with hard real-time or soft real-time requirements, while preemptible nodes have applications with soft real-time or no real-time requirements.

[0008] Step 2: Assign a priority class to each node, where non-preemptible nodes have the highest priority, and preemptible nodes have different priority classes. The lower the priority class, the easier it is for data packets to be interrupted.

[0009] Step 3: Establish a first-in-first-out priority queue for each priority category, and place data packets into the corresponding queue according to their priority category;

[0010] Step 4: When a data packet arrives at the server, check whether it belongs to a non-preemptible node or a preemptible node; if the data packet belongs to a non-preemptible node, put it into queue Q1, push it to the front of the queue, and remove it from the queue by the server; if the data packet belongs to a preemptible node PN... g If a packet is not found, it is placed in queue Q2 and checked whether it can leave the server without interference from packets from other priority categories of non-preemptible or preemptible nodes. If it can, it is pushed to the front of the queue and removed by the server. If not, the discretionary rule is applied to decide whether to interrupt the currently transmitted preemptible node packet based on the service time of the preemptible node.

[0011] Step 5: If an interruption occurs, the transmission of the currently transmitting preemptible node data packet is paused and pushed to the front of queue Q2. The newly arriving preemptible node data packet is pushed to the front of the queue and ejected by the server. If there is no interruption, the newly arriving preemptible node data packet is placed at the rear of queue Q2 and waits for the currently transmitting preemptible node data packet to be completely ejected by the server before being ejected by the server.

[0012] The delay of each PN data packet in this invention can be divided into two parts: (1) initial delay, which is the time required to serve the initial data packet; and (2) delay busy period, which is the time spent serving high priority data packets before considering PN data packets.

[0013] PN based on high-priority data packets gDue to the impact of packet queuing time, nodes are further divided into three categories: Type A, Type B, and Type C. Type A nodes include all NPNs and PNs with a priority higher than g-1 (g-1). Type C nodes include PNs with lower priorities g+1 to M. Type B nodes are all PNs with priority g; assuming packet arrival follows a Poisson distribution, the Type A delay period begins when an NPN packet or a Type A packet arrives at the server. The Type B delay period begins when a Type B packet arrives at the server. Finally, the Type C delay period begins when a Type C packet arrives at the server.

[0014] A typical delay period ends when the packet initiating the delay leaves the server and the server no longer receives type A and type B packets. Generally, a typical busy delay period can be viewed as a series of mutually exclusive delay periods (i.e., type A delay periods, type B delay periods, or possibly multiple type C delay periods). PN g Data packet occupancy time R g It can be considered as the sum of N interruptions, plus the preempted service time S. pg and a successful service time S sg However, during the successful service time S sg Unpreemptible period S Bg There may be packets from high-priority nodes waiting in the queue. Therefore, PN g Data packet completion time C g Including the time R g and by S bg The delay caused by high-priority data packets waiting in the queue during the busy period Y g . Pay attention to Y g The independent failure time D is generated by high-priority data packets. g The combination of . Furthermore, each failure time D g The length of PN is uniformly distributed for each interruption. g The moment a data packet arrives at the server, the server is empty; this can be represented as the busy period of Bg. Therefore, the Laplace transform of Bg can be expressed as:

[0015]

[0016] λ g This represents the average rate at which a node (which could be an intermediate or leaf node) generates data packets, measured in data packets per second. Because D g Triggered by a type A packet (in this case, a PN with priority g-1), its length is equal to that of type B. g-1However, if the failure time is caused by an NPN packet (i.e., a node with non-preemptive priority), then the initial delay period D is initiated. g-1 Furthermore, in the initial delay period D g-1 During this period, each type A data packet waiting in the server generates a B. g-1 The sub-busy cycle. The probability of each failure time occurring is λ. g-1 / Λ g-1 and Λ g-1 -λ g-1 / Λ g-1 Therefore, D g The Laplace transform can be expressed recursively as:

[0017]

[0018] Where g≥2, D1(s)=1,

[0019] Subsequently, PN g The number of interruptions a data packet encounters before it completely leaves the server is equal to the number of interruptions encountered in S. Ag The number of high-priority packets arriving during the (preemptible period) and the completion time can be considered as a delay period plus S. g (S Ag With S Bg The initial delay (sum of) during which high-priority packets waiting in the queue generate a D g The sub-busy cycle. Therefore, the Laplace transform of the completion time can be expressed as:

[0020]

[0021] With the server in state j (where j∈{0, A, B, C}, using the previously described delay period, and assuming that packet arrivals follow a Poisson distribution), PN g Steady-state probability π of data packet arrival g It can be represented as:

[0022] π0=1-ρ,π A =ρ A (1-ρ) / (1-ρ B -ρ A ), π B =ρ B (1-ρ) / 1-ρ B -ρ A ), π C =ρ C / (1-ρ B -ρ A )

[0023] Here, the steady-state condition is utilized, where ρ represents the utilization rate of PNs, defined as ρ = λ. g E[S eg ], where S eg It is PN g The effective service time of the data packet.

[0024] Furthermore, assume that the server is in state 0 when it is empty. Therefore, the Laplace transform of the queuing time can be expressed as:

[0025]

[0026] in It is PN g The conditional waiting time for a data packet while the server is in state j (i.e., a type A delay period, a type B delay period, or multiple type C delay periods). Therefore, if ψ g / j Indicates the initial delay, C g Let j represent the service time for a delay period of type j, where j∈{A, B, C}. Then the conditional waiting time can be obtained by the following formula:

[0027]

[0028] in, The initial delay for type A and type B delay periods only needs to be estimated. You can then get the queue time. In order to get... Consider type C delay periods. In this case... This can be viewed as a type C busy period triggered by type h packets (where h ∈ {C}). This type C busy period ends when a type h packet leaves the server and the server is no longer in a type A or type B delay period. Therefore, This can be considered as the failure time caused by packets from type C nodes, with priorities ranging from g+1 to h-1. When packets arrive at the link server, they are immediately classified according to priority, with high-priority packets moving directly to the front of the queue. Furthermore, if the judgment rules are met, high-priority packets are pushed off the link server, while low-priority packets wait in the queue. Clearly, this results in a lower average latency for high-priority packets.

[0029] This invention first organizes network nodes according to preemptible and non-preemptible service categories, and pushes data packets out through a link server to prevent node buffer overflow, employing a preemptive / non-preemptive principle. This invention also provides a queuing model based on discretionary rules, used to determine whether a data packet can leave the server without interference from data packets of other priority categories, based on the data packet's service time. This invention further provides a method for implementing different first-in-first-out priority queues for data packets of different priority groups, used to distinguish traffic types and prevent header blocking.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By implementing a priority queue model, preemption and non-preemption are performed on data packets of each priority group to distinguish different traffic types and prevent data packet collisions and congestion; 2. By implementing a priority queue model, interrupted and uninterrupted periods are divided for data packets served from the buffer; 3. By implementing cautious rules for preemption / non-preemption, selfish behavior of nodes when transmitting data packets is prevented; 4. By implementing cautious rules for preemption / non-preemption, low-priority data packets are protected from interruption by high-priority data packets during non-preemption periods based on their cumulative service time. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention; Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention are described clearly and completely below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention. To avoid obscuring the essence of this invention, well-known methods and processes are not described in detail.

[0033] The following is combined with Figure 1 The following describes specific embodiments of the present invention:

[0034] Step 1: First, divide the nodes in the network topology into two groups: non-preemptible nodes NPN and preemptible nodes PN. Non-preemptible nodes have applications with hard real-time or soft real-time requirements, while preemptible nodes have applications with soft real-time or no real-time requirements.

[0035] Step 2: Assign a priority class to each node, where non-preemptible nodes have the highest priority, and preemptible nodes have different priority classes. The lower the priority class, the easier it is for data packets to be interrupted.

[0036] Step 3: Establish a first-in-first-out priority queue for each priority category, and place data packets into the corresponding queue according to their priority category;

[0037] Step 4: When a data packet arrives at the server, check whether it belongs to a non-preemptible node or a preemptible node; if the data packet belongs to a non-preemptible node, put it into queue Q1, push it to the front of the queue, and remove it from the queue by the server; if the data packet belongs to a preemptible node PN... g If a packet is not found, it is placed in queue Q2 and checked whether it can leave the server without interference from packets from other priority categories of non-preemptible or preemptible nodes. If it can, it is pushed to the front of the queue and removed by the server. If not, the discretionary rule is applied to decide whether to interrupt the currently transmitted preemptible node packet based on the service time of the preemptible node.

[0038] Step 5: If an interruption occurs, the transmission of the currently transmitting preemptible node data packet is paused and pushed to the front of queue Q2. The newly arriving preemptible node data packet is pushed to the front of the queue and ejected by the server. If there is no interruption, the newly arriving preemptible node data packet is placed at the rear of queue Q2 and waits for the currently transmitting preemptible node data packet to be completely ejected by the server before being ejected by the server.

[0039] The delay of each PN packet can be divided into two parts: (1) initial delay, which is the time required to serve the initial packet; and (2) busy delay period, which is the time spent serving high-priority packets before considering PN packets. Therefore, the delay of PN packets is determined based on the high-priority packets. g Due to packet queuing time, nodes are further categorized into three types: Type A, Type B, and Type C. Type A nodes include all NPNs and PNs with a priority higher than g-1 (g-1). Type C nodes include PNs with lower priorities g+1 to M. Type B nodes are all PNs with priority g. Assuming packet arrival follows a Poisson distribution, the Type A delay period begins when an NPN or Type A packet arrives at the server. The Type B delay period begins when a Type B packet arrives at the server. Finally, the Type C delay period begins when a Type C packet arrives at the server. A typical delay period ends when the packet initiating the delay leaves the server and the server no longer has Type A or Type B packets. Generally, a typical busy delay period can be viewed as a series of mutually exclusive delay periods (i.e., Type A delay periods, Type B delay periods, or possibly multiple Type C delay periods). g Data packet occupancy time R g It can be considered as the sum of N interruptions, plus the preempted service time S.pg and a successful service time S sg However, during the successful service time S sg Unpreemptible period S Bg There may be packets from high-priority nodes waiting in the queue. Therefore, PN g Data packet completion time C g Including the time R g and by S bg The delay caused by high-priority data packets waiting in the queue during the busy period Y g . Pay attention to Y g The independent failure time D is generated by high-priority data packets. g The combination of . Furthermore, each failure time D g The length of PN is uniformly distributed for each interruption. g The data packet is empty at the moment it arrives at the server, which can be represented as B. g The busy cycle.

[0040] Therefore, B g The Laplace transform can be expressed as:

[0041]

[0042] λ g This represents the average rate at which a node (which may be an intermediate node or a leaf node) generates data packets, measured in data packets per second.

[0043] Because of D g Triggered by a type A packet (in this case, a PN with priority g-1), its length is equal to that of type B. g-1 However, if the failure time is caused by an NPN packet (i.e., a node with non-preemptive priority), then the initial delay period D is initiated. g-1 Furthermore, in the initial delay period D g-1 During this period, each type A data packet waiting in the server generates a B. g-1 The sub-busy cycle. The probability of each failure time occurring is λ. g-1 / Λ g-1 and Λ g-1 -λ g-1 / Λ g-1 Therefore, D g The Laplace transform can be expressed recursively as:

[0044]

[0045] Where g≥2, D1(s)=1, Subsequently, PN gThe number of interruptions a data packet encounters before it completely leaves the server is equal to the number of interruptions encountered in S. Ag The number of high-priority packets arriving during the (preemptible period) and the completion time can be considered as a delay period plus S. g (S Ag With S Bg The initial delay (sum of) during which high-priority packets waiting in the queue generate a D g The sub-busy cycle. Therefore, the Laplace transform of the completion time can be expressed as:

[0046]

[0047] With the server in state j (where j∈{0, A, B, C}, using the previously described delay period, and assuming that packet arrivals follow a Poisson distribution), PN g Steady-state probability π of data packet arrival g It can be represented as:

[0048] π0=1-ρ,π A =ρ A (1-ρ) / (1-ρ B -ρ A ), π B =ρ B (1-ρ) / 1-ρ B -ρ A ), π C =ρ C / (1-ρ B -ρ A )

[0049] Here, the steady-state condition is utilized, where ρ represents the utilization rate of PNs, defined as ρ = λ. g E[S eg ], where S eg It is PN g The effective service time of data packets. Furthermore, assume the server is in state 0 when it is empty. Therefore, the Laplace transform of the queuing time can be expressed as:

[0050]

[0051] in It is PN g The conditional waiting time for a data packet while the server is in state j (i.e., a type A delay period, a type B delay period, or multiple type C delay periods). Therefore, if ψ g / j Indicates the initial delay, C g Let j represent the service time for a delay period of type j, where j∈{A,B,C}. Then the conditional waiting time can be obtained using the following formula:

[0052]

[0053] in, The initial delay for type A and type B delay periods only needs to be estimated. You can then get the queue time. In order to get... Consider type C delay periods. In this case... This can be viewed as a type C busy period triggered by type h packets (where h ∈ {C}). This type C busy period ends when a type h packet leaves the server and the server is no longer in a type A or type B delay period. Therefore, This can be considered as the failure time caused by packets from type C nodes, with priorities ranging from g+1 to h-1. When packets arrive at the link server, they are immediately classified according to priority, with high-priority packets moving directly to the front of the queue. Furthermore, if the judgment rules are met, high-priority packets are pushed off the link server, while low-priority packets wait in the queue. Clearly, this results in a lower average latency for high-priority packets.

[0054] When a PN has experienced n interruptions g The total time required for a data packet to be pushed to the server includes queuing delays caused by these n interruptions and PN. g The service time of data packets. Therefore, assuming the arrival follows a Poisson distribution, the response time T g It consists of two independent random variables, namely queuing time W. g and stay time R g Therefore, the Laplace transform of the response time can be expressed as:

[0055]

[0056] Compared to existing technologies, this invention proposes a deterministic network congestion control method based on priority queues. This algorithm utilizes a priority-based preemptible / non-preemptible priority queue model to control congestion in node buffers. This model first defines the network topology as a tree structure of sink nodes, intermediate nodes, and leaf nodes. Nodes in the network are organized according to preemptible and non-preemptible service categories, and data packets are pushed out through link servers, thereby preventing node buffer overflow. This invention also provides a queuing model based on discretionary rules, used to determine whether a data packet can leave the server without interference from data packets of other priority categories based on the data packet's service time. This invention further provides a method for implementing different first-in-first-out priority queues for data packets of different priority groups, used to distinguish traffic types and prevent head-end blocking.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for deterministic network congestion control based on priority queues, characterized in that The method comprises the following steps: Step 1: divide the nodes in the network topology into two groups: non-preemptible nodes NPN and preemptible nodes PN, wherein the non-preemptible nodes have hard real-time or soft real-time demand application programs, and the preemptible nodes have soft real-time or non-real-time demand application programs; Step 2: assign a priority class to each node, wherein the non-preemptible nodes have the highest priority, the preemptible nodes have a low priority class, and the lower the priority class, the easier the data packet is interrupted; Step 3: establish a first-in-first-out priority queue for each priority class, and put the data packet into the corresponding queue according to the priority class to which the data packet belongs; Step 4: When a data packet arrives at the server, check whether it belongs to non-preemptible node or preemptible node; if the data packet belongs to non-preemptible node, put it into queue Q1 and push it to the front of the queue and push it out by the server; if the data packet belongs to preemptible node , g refers to priority, put it into queue Q2 and check whether it can leave the server without the interference of non-preemptible node or preemptible node data packet of other priority categories; if it can, push it to the front of the queue and push it out by the server; if it cannot, apply discretionary rules to decide whether to interrupt the currently transmitting preemptible node data packet according to the served time of the preemptible node; Step 5: if an interruption occurs, the preemptible node data packet currently being transmitted is paused and pushed to the head of the queue Q2, and the newly arrived preemptible node data packet is pushed to the front of the queue and pushed out by the server; if no interruption occurs, the newly arrived preemptible node data packet is put into the tail of the queue Q2 and waits to be pushed out by the server after the currently transmitted preemptible node data packet is completely pushed out of the server.

2. The priority queue based deterministic network congestion control method of claim 1, wherein Comprise: The delay of each PN data packet in step 4 can be divided into two parts: (1) initial delay, that is, the time required to serve the initial data packet; (2) delay busy period, that is, the time spent for serving high-priority data packets before considering PN data packets.

3. The priority queue based deterministic network congestion control method of claim 2, wherein In step 4, the influence of packet queuing time on the high-priority packet The nodes are further classified into three types: Type A, Type B and Type C. Type A nodes include all NPNs and PNs with higher priority than g-1. Type C nodes include PNs with low priority from g+1 to M. Type B nodes are all PNs with priority g. If it is assumed that the data packet arrival conforms to the Poisson distribution, then the type A delay period starts when the NPN data packet or the type A data packet arrives at the server; the type B delay period starts when the type B data packet arrives at the server; finally, the type C delay period starts when the type C data packet arrives at the server.

4. The priority queue based deterministic network congestion control method of claim 3, wherein A typical delay period in Step 4 ends when the data packet that initiated the delay leaves the server and the server has no more Type A and Type B data packets; a typical delay busy period can be viewed as a series of mutually exclusive delay periods; the occupancy time of the data packet the sum of the N interruptions, plus the pre-empted service time , and a successful service time , during the successful service time , during the non-pre-emptable period , there can be data packets of high priority nodes waiting in the queue; the completion time of the data packet includes the occupancy time and the delay busy periods caused by high priority data packets waiting in the queue during , is the combination of independent failure times generated by high priority data packets; the length of each failure time is identically distributed for each interruption; from the instant the data packet arrives at the server to the server being empty, denoted as the busy period; the Laplace transform of can be expressed as: ； denotes the average rate at which a node generates data packets, in packets per second; since is triggered by a type A data packet, in this case, a pre-emptable node PN having a priority of g - 1, whose failure time length is equal to .

5. The priority queue based deterministic network congestion control method of claim 4, wherein If the fault time is initiated by an NPN packet in Step 4, start the initial delay period ; During the initial delay period , wait for each type A data packet in the server to generate a busy sub-period of 1; the probability of occurrence of each fault time is and ; The Laplace transform of the busy period is expressed in a recursive form as follows: ； wherein g > 2, (s) = 1, .

6. The priority queue based deterministic network congestion control method of claim 5, wherein In step 4 The number of interruptions a data packet encounters before it completely leaves the server is equal to the number of interruptions encountered in... The number of high-priority packets arriving during the preemptible period, and the completion time can be considered as a delay period plus... The initial delay, for and The sum, during which time, a high-priority data packet waiting in the queue generated a The sub-busy cycle; the Laplace transform of the completion time is expressed as: 。 7. The priority queue based deterministic network congestion control method of claim 1, wherein Comprise: The step 4 assumes that the data packet arrivals follow a Poisson distribution, the server is in state j, where j ∈ {0, A, B, C}, uses the delay period described earlier, and assumes that the data packet arrivals follow a Poisson distribution, Steady state probabilities of data packet arrivals is represented as: ； where the steady state condition is exploited, p denotes the utilization, defined as where is the effective service time of a data packet.

8. The priority queue based deterministic network congestion control method of claim 1, wherein Comprise: In step 4, it is assumed that when the server is empty, the server is in state 0; the Laplace transform of the queuing time is represented as: ； wherein is Condition waiting time of the data packet at the server in state j, i.e. type A delay period or type B delay period or multiple type C delay periods.

9. The priority queue based deterministic network congestion control method of claim 1, wherein Comprise: in step 4 denotes the initial delay, denotes the service time of type j delay period, where j e {A, B, C}, and the conditional waiting time is given by: ； wherein, denotes the initial delay of the type A and type B delay periods, which only needs to be estimated to obtain the queuing time; to obtain , the type C delay period is considered; may be considered as a type C busy cycle triggered by a type h packet, where h e {C}; this type C busy cycle ends when the type h packet leaves the server and the server is not in a type A or type B delay cycle; are considered as fault times triggered by packets of type C nodes, with priorities from g+1 to h-1; when the packets arrive at the link server, they are immediately classified according to the priority, and the high-priority packets are directly moved to the front of the queue; if the judgment rule is met, the high-priority packets are pushed out of the link server, and the low-priority packets wait in the queue.

Images