Traffic theory time delay upper bound calculation method in network equipment scheduling system

Abstract

The invention discloses a flow theoretical time delay upper bound calculation method in a network equipment scheduling system, which is characterized in that theoretical time delay analysis is closely combined with a scheduling structure and algorithm of the network equipment scheduling system, and a mode of combining deterministic parameter analysis and non-deterministic parameter analysis is adopted; the quantitative calculation mode of the upper bound of the theoretical time delay of the flow in the network equipment scheduling system is determined under the condition of not depending on statistical data, and the problem that the time delay is difficult to calculate due to uncertainty of a part of algorithms of the network equipment is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of network communications, and in particular relates to a method for calculating an upper bound of theoretical flow delay in a network equipment scheduling system. Background Art

[0002] A network device scheduling system typically refers to a software system used to manage, monitor, and optimize the resource allocation and operational status of network devices, including switches, routers, firewalls, and servers. Its core goal is to improve network resource utilization, ensure business continuity, and reduce operational costs through automated and intelligent scheduling strategies. Traffic latency in a network refers to the total time it takes for data to travel from the source node to the destination node. It primarily includes processing latency, queuing latency, transmission latency, and propagation latency. Key influencing factors include network topology, traffic characteristics, device scheduling strategies, and quality of service requirements. Device scheduling strategies include queue management and congestion control.

[0003] Existing theoretical upper bounds for traffic latency are primarily calculated using methods based on queuing theory. Queuing theory is a fundamental theory for analyzing network latency, deriving the statistical characteristics of latency by establishing an "arrival process-service process-queue model." Queuing theory typically relies on statistical averages, making it difficult to directly provide a deterministic upper bound on latency. Approximate estimates based on probability distributions (such as the 99th percentile) are required. Furthermore, the simplicity of queuing theory methods lacks analysis of the specific scheduling structures and algorithms within switches, impacting the accuracy of the calculated results. Summary of the Invention

[0004] In view of this, the present invention provides a method for calculating the theoretical upper bound of flow delay in a network device scheduling system. It adopts a combination of deterministic parameter analysis and non-deterministic parameter analysis to determine a strict formula for the theoretical upper bound of flow delay, thereby improving the accuracy of the calculation results.

[0005] The present invention provides a method for calculating the upper bound of theoretical traffic delay in a network device scheduling system, comprising the following steps:

[0006] At the beginning of network device scheduling, all non-congested flows enter the congested queue. During the scheduling process, if the input rate of the congested flow packets completely occupies the total output bandwidth of the switch, the smaller value of the time when the token of the congested flow is exhausted or the time when the accumulated number of packets of the congested flow exceeds its accumulated token is obtained. The upper bound of the theoretical flow delay of the network device is the product of this smaller value and a first ratio. The first ratio is the ratio of the accumulated rate of the non-congested flow packets to the committed bandwidth of the non-congested flow. Otherwise, the time when the token of the congested flow is exhausted is recorded as the base time. The upper bound of the theoretical flow delay of the network device is the product of this base time and the first ratio.

[0007] Furthermore, the method for judging whether the input rate of the congested flow data packets completely occupies the total output bandwidth of the switch is as follows: if K c *R c *CIR c ≥M, the congested flow packet input rate fully occupies the total output bandwidth of the switch during the scheduling process; otherwise, the congested flow packet input rate does not fully occupy the total output bandwidth of the switch, where K c is the total number of congested flows, R c The actual sending rate of data packets for congested flows, CIR c is the committed bandwidth for the congested flow, and M is the total sending bandwidth of the switch.

[0008] Furthermore, when the congested flow packet input rate completely occupies the total output bandwidth of the switch, the time t at which the congested flow token is exhausted is c1 for Where P is the number of pre-allocated tokens, v ccrd1 is the net token consumption rate and K c is the total number of congested flows, M is the total sending bandwidth of the switch, and CIR c Commit bandwidth for congested flows.

[0009] Furthermore, when the congested flow packet input rate completely occupies the total output bandwidth of the switch, the time t at which the congested flow packet accumulation exceeds its token accumulation is p1 for Where P is the number of pre-allocated tokens, v ccrd1 is the net token consumption rate and K c is the total number of congested flows, M is the total sending bandwidth of the switch, and CIR c Committed bandwidth for congested flows, v ncpak1 is the cumulative rate of packets in the non-congested flow and v ncpak1 =R nc *CIR nc , R nc The actual sending rate of data packets for non-congested flows, CIR nc Commit bandwidth for non-congested flows.

[0010] Furthermore, when the input rate of the congested flow packets does not completely occupy the total output bandwidth of the switch, the moment when the tokens of the congested flow are exhausted is the same as the moment when the accumulated amount of the congested flow packets exceeds its accumulated amount of tokens.

[0011] Furthermore, when the congested flow packet input rate does not fully occupy the total output bandwidth of the switch, the non-congested flow packet accumulation rate v ncpak2 for Among them, R ncThe actual sending rate of data packets for non-congested flows, CIR nc is the committed bandwidth of the non-congested flow, M is the total sending bandwidth of the switch, K c is the total number of congested flows, R c The actual sending rate of data packets for congested flows, CIR c Committed bandwidth for congested flows, K nc is the total number of non-congested flows.

[0012] Beneficial effects:

[0013] The present invention closely combines theoretical delay analysis with the scheduling structure and algorithm of the network equipment scheduling system, and adopts a combination of deterministic parameter analysis and non-deterministic parameter analysis to determine the quantitative calculation method of the upper bound of the theoretical delay of traffic in the network equipment scheduling system without relying on statistical data, thus solving the problem that some algorithms of network equipment have uncertainty and are difficult to calculate delay. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of the structure of a network device scheduling system targeted by a method for calculating the upper bound of theoretical traffic delay in a network device scheduling system provided by the present invention. DETAILED DESCRIPTION

[0015] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0016] The present invention provides a method for calculating the upper bound of theoretical flow delay in a network device scheduling system. The core idea is to analyze the switch scheduling process, combine the uncertainty of the actual scheduling process, obtain the various stages of theoretical flow scheduling delay, and find the stage with the largest delay; by considering the worst-case scheduling delay, the uncertainty factor is converted into a deterministic factor, and the upper bound of the theoretical flow delay of this stage is calculated.

[0017] The network equipment scheduling system structure is as follows Figure 1 As shown, the system contains two queues of different priorities, namely, the non-congested flow queue and the congested flow queue. The non-congested flow queue has a higher priority, while the congested flow queue has a lower priority. Specifically, after a data packet enters the network device scheduling system, the congestion identification mechanism first sets a label, including congested and non-congested labels, and then stores it in the corresponding flow according to the label. The congestion identification mechanism checks the flow corresponding to the data packet. If the accumulated number of data packets in the flow is greater than the accumulated number of tokens, the label of the data packet is set to congested; otherwise, the label of the data packet is set to non-congested. The flow enters the corresponding queue based on the label of the first data packet, and the network device scheduling system selects the flow to send the data packet through priority scheduling.

[0018] To perform latency analysis, this paper assumes that a certain number of congested and non-congested flows enter the network device simultaneously, and that the current total packet input volume exceeds the total output bandwidth of the network device. Because the increasing latency of congested flows has no upper bound, this paper only analyzes the upper bound of the latency of non-congested flows. Furthermore, for ease of analysis, packet reception and transmission, as well as token allocation consumption, are considered at the bit granularity level.

[0019] The present invention provides a method for calculating the upper bound of theoretical traffic delay in a network device scheduling system, which specifically includes the following steps:

[0020] Step 1: Analyze the scheduling process of the network device scheduling system and determine that the time when the theoretical upper bound of its traffic delay is generated is the time when the congestion flow token is exhausted or the time when the accumulated amount of congestion flow data packets exceeds the threshold and enters the congestion queue. The theoretical upper bound of the traffic delay is the smaller value of the above two times.

[0021] The specific analysis process is as follows. By analyzing the switch scheduling process and taking into account the uncertainty of the actual scheduling process, we derive the theoretical delays for each stage of flow scheduling and identify the stage with the highest delay. Ideally, at the start of scheduling, all packets are labeled as non-congested, resulting in flows being placed in the non-congested queue, and scheduling is approximately round-robin. However, in actual scheduling, packets may arrive in bursts, and token allocation may be delayed. As a result, some packets may be labeled as congested, causing the corresponding flows to enter the congested queue. Because the congested queue has a lower priority, these flows must wait until all non-congested flows have completed their transmission before they can be sent. This can also increase scheduling delays for non-congested flows.

[0022] The scheduling process can be divided into two cases: First, when the tokens of a congested flow are exhausted first, the congested flow can only send packets at the committed bandwidth rate. At this time, non-congested flows in the congested queue can also send packets normally, and the packet accumulation does not increase. Second, when the accumulated packets of a congested flow exceed its token accumulation, the congestion identification mechanism is triggered. At this time, the congested flow's packets are labeled as congested. After sending packets with non-congested labels, the congested flow enters the congested queue. At this time, the accumulated packets of non-congested flows do not increase. Therefore, the theoretical upper bound of the flow delay occurs when the tokens of a congested flow are exhausted or when the accumulated packets of a congested flow exceed the threshold and enter the congested queue.

[0023] Step 2: When all non-congested flows enter the congested queue due to misidentification of congestion at the beginning of scheduling, according to K c *R c *CIR c The relationship between the size of M is divided into two situations: the input rate of congested flow packets completely occupies the total output bandwidth of the switch; and the input rate of congested flow packets cannot completely occupy the total output bandwidth of the switch.

[0024] When the input rate of congested flow packets completely occupies the total output bandwidth of the switch, the theoretical upper bound of the delay of non-congested flows is:

[0025]

[0026] Among them, t c1 is the moment when the token of the congested flow is exhausted, and P is the number of pre-allocated tokens, v ccrd1 is the net token consumption rate and K c is the total number of congested flows, M is the total sending bandwidth of the switch, and CIR c Commit bandwidth for congested flows; v ncpak1 is the cumulative rate of packets in the non-congested flow, and v ncpak1 =R nc *CIR nc , R nc The actual sending rate of data packets for non-congested flows, CIR nc Commit bandwidth for non-congested flows; t p1 is the moment when the accumulated amount of packets of the congested flow exceeds its accumulated amount of tokens, and

[0027] When the input rate of congested flow packets cannot fully occupy the total output bandwidth of the switch, the theoretical upper bound of the delay of non-congested flows is:

[0028]

[0029] Among them, t c2 is the moment when the token of the congested flow is exhausted, and v cpak2 is the congestion flow packet accumulation rate, and its value is zero; v ncpak2 is the cumulative rate of packets for non-congested flows, and

[0030] Example 1:

[0031] In this embodiment, a network device uses a method for calculating the theoretical upper bound of traffic delay in a network device scheduling system provided by the present invention to calculate the theoretical upper bound of the delay of non-congested flows in the scheduling mechanism within the switch, specifically including:

[0032] By considering the worst-case scheduling delay, we transform the uncertainty into a deterministic factor and calculate the theoretical upper bound of the flow delay for this phase. The relevant variables used in the calculation are shown in Table 1.

[0033] Table 1 Variables related to the upper bound of flow theory delay

[0034]

[0035] In the fluid model, discussing the upper bound of the theoretical delay for non-congested flows is meaningful only when the total amount of input packets to the network device is greater than the total output bandwidth of the network device, that is:

[0036] K c *R c *CIR c +K nc *R nc *CIR nc >M

[0037] At the beginning of scheduling, some flows may enter the congestion queue. Therefore, the present invention considers the worst case in analysis and assumes that all non-congested flows enter the congestion queue due to misidentification of congestion at the beginning of scheduling. c *R c *CIR c Different from the size relationship of M, it can be divided into the following two cases:

[0038] 1) The congested flow packet input rate can fully occupy the total output bandwidth of the switch, that is, K c *R c *CIR c ≥M, at this time, the non-congested flow has no data packets to send, so the congested flow data packet accumulation rate v cpak1 and the net token consumption rate v ccrd1 for:

[0039]

[0040] Non-congested flow packet accumulation rate v ncpak1 for:

[0041] v ncpak1 =R nc *CIR nc

[0042] Then there are two situations: a) the token of the congested flow is exhausted first; b) the accumulated amount of data packets of the congested flow exceeds its accumulated amount of tokens first.

[0043] For case a), it happens at:

[0044]

[0045] For case b), it happens at:

[0046]

[0047] Therefore, the theoretical upper bound of the delay of non-congested flows is:

[0048]

[0049] 2) The input rate of congested flow packets cannot fully occupy the total output bandwidth of the switch, i.e., K c *R c *CIR c <M, at this time, there are some packets sent by non-congested flow. Therefore, the accumulation rate v of congested flow packets cpak2 and the net token consumption rate v ccrd2 are as follows:

[0050] v cpak2 = 0

[0051] v ccrd2 = (R c - 1)*CIR c

[0052] The accumulation rate v of non-congested flow packets ncpak2 is as follows:

[0053]

[0054] Subsequently, there are the following two situations: a) The tokens of the congested flow are exhausted first; b) The accumulated amount of congested flow packets exceeds its token accumulation amount first.

[0055] For situation a), the moment when it occurs is:

[0056]

[0057] For situation b). The moment when it occurs is:

[0058]

[0059] Therefore, the upper bound of the theoretical delay of non-congested flow is:

[0060] In summary, the above is only the preferred embodiment of the present invention, and is not used to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for calculating the upper bound of theoretical traffic delay in a network device scheduling system, characterized in that: The following steps are involved: At the start of network device scheduling, all non-congested flows enter the congested queue. During the scheduling process, if the input rate of the congested flow packets completely occupies the total output bandwidth of the switch, the smaller value of the time when the token of the congested flow is exhausted or the time when the accumulated number of packets of the congested flow exceeds its accumulated token is obtained. The upper bound of the theoretical flow delay of the network device is the product of the smaller value and a first ratio. The first ratio is the ratio of the accumulated rate of the non-congested flow packets to the committed bandwidth of the non-congested flow. Otherwise, the time when the token for obtaining the congested flow is exhausted is recorded as the reference time, and the upper bound of the theoretical delay of the flow of the network device is the product of the reference time and the first ratio.

2. The method for calculating the upper bound of theoretical traffic delay according to claim 1 is characterized in that: The method to judge whether the input rate of congested flow packets fully occupies the total output bandwidth of the switch is: if K c *R c *CIR c ≥M, the congested flow packet input rate fully occupies the total output bandwidth of the switch during the scheduling process; otherwise, the congested flow packet input rate does not fully occupy the total output bandwidth of the switch, where K c is the total number of congested flows, R c The actual sending rate of data packets for congested flows, CIR c is the committed bandwidth for the congested flow, and M is the total sending bandwidth of the switch.

3. The method for calculating the upper bound of theoretical traffic delay according to claim 1 is characterized in that: When the input rate of the congested flow packets completely occupies the total output bandwidth of the switch, the time t at which the token of the congested flow is exhausted is c1 for Where P is the number of pre-allocated tokens, v ccrd1 is the net token consumption rate and K c is the total number of congested flows, M is the total sending bandwidth of the switch, and CIR c Commit bandwidth for congested flows.

4. The method for calculating the upper bound of theoretical traffic delay according to claim 1 is characterized in that: When the input rate of the congested flow packets completely occupies the total output bandwidth of the switch, the time t at which the accumulated amount of congested flow packets exceeds its token accumulation amount is reached. p1 for Where P is the number of pre-allocated tokens, v ccrd1 is the net token consumption rate and K c is the total number of congested flows, M is the total sending bandwidth of the switch, and CIR c Committed bandwidth for congested flows, v ncpak1 is the cumulative rate of packets in the non-congested flow and v ncpak1 =R nc *CIR nc , R nc The actual sending rate of data packets for non-congested flows, CIR nc Commit bandwidth for non-congested flows.

5. The method for calculating the upper bound of theoretical traffic delay according to claim 1 is characterized in that: When the input rate of the congested flow packets does not fully occupy the total output bandwidth of the switch, the moment when the tokens of the congested flow are exhausted is the same as the moment when the accumulated amount of the congested flow packets exceeds its accumulated amount of tokens.

6. The method for calculating the upper bound of theoretical traffic delay according to claim 1, characterized in that: When the input rate of congested flow packets does not fully occupy the total output bandwidth of the switch, the accumulation rate of non-congested flow packets v ncpak2 for Among them, R nc The actual sending rate of data packets for non-congested flows, CIR nc is the committed bandwidth of the non-congested flow, M is the total sending bandwidth of the switch, K c is the total number of congested flows, R c The actual sending rate of data packets for congested flows, CIR c Committed bandwidth for congested flows, K nc is the total number of non-congested flows.

Images