Core network service chain rapid reconstruction method based on network function virtualization

Abstract

The invention discloses a core network service chain rapid reconstruction method based on network function virtualization, and relates to the technical field of communication network data transmission and network service orchestration, and the method comprises the steps: obtaining a cabinet loop differential pressure sequence and an error correction counting sequence where each candidate takeover host is located, and determining a reconstruction observation window and a reference window based on an orchestration event timestamp; respectively calculating a differential pressure ripple index and an error correction peak cluster index according to the change condition of the corresponding sequence; calculating the physical interference risk quantity based on the indexes and the synchronization intensity of the two sequences in the observation window; and finally, carrying out risk scoring on the candidate service chains according to the risk quantity, selecting the service chain with the lowest score to execute flow cutting, and outputting a reconstruction posterior consistency quantity. According to the method, cross-layer interference between short-time instability of a physical layer and communication errors can be quantified, hardware transient fluctuation is prevented from being misjudged as software overload, and wrong drop point selection and reconstruction oscillation are prevented.

Description

Technical Field

[0001] This invention relates to the field of communication network data transmission and network service orchestration technology, and in particular to a method for rapid reconfiguration of core network service chains based on network function virtualization. Background Technology

[0002] The core network's control plane and user plane functions are gradually being deployed virtualized on cloud infrastructure. Multiple virtual network functions are organized sequentially through a service chain to carry various communication services. To meet high reliability requirements, when a lower-level node fails or triggers a reconfiguration, the core network service chain needs to be rapidly reconfigured to ensure service continuity. In high-density deployment scenarios with liquid-cooled cabinets, service chain reconfiguration often occurs concurrently with the migration of network function instances. At this time, the cabinet's physical cooling loop is highly susceptible to short-term hydraulic transients due to pump control adjustments or flow redistribution, manifesting as short-term differential pressure ripple fluctuations. Simultaneously, the uplink ports of candidate takeover hosts often experience a sudden surge in clustered spikes in forward error correction counts or cyclic redundancy check counts. These transient physical layer environmental disturbances and link transmission errors have a significant negative impact on the reliability of flow switching within the reconfiguration observation window.

[0003] Existing service chain reconfiguration strategies often rely on network link reachability and general server performance metrics for candidate chain evaluation and selection. This makes it difficult to identify and quantify the potential cross-layer interference between short-term fluctuations in rack cooling circuit differential pressure and sudden increases in host port error correction counts. Due to the lack of quantitative and comprehensive methods for assessing short-term physical layer instability, existing technologies are highly prone to misjudging communication anomalies caused by hardware transient fluctuations as software-level processing overload or system orchestration failures within the reconfiguration window. This lack of analytical perspective and misjudgment directly leads to incorrect service chain endpoint selection decisions, resulting in repeated flow switching and reconfiguration oscillations between multiple endpoints in the core network service chain. This not only significantly prolongs network service recovery time but also severely degrades the overall user experience. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as the difficulty in quantifying the cross-layer interference between short-term physical layer instability and link communication errors, the tendency to misjudge hardware transient fluctuations as software overload, and the resulting errors in point selection and repeated reconstruction oscillations. This invention proposes a rapid reconstruction method for core network service chains based on network function virtualization.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A method for rapid reconfiguration of core network service chains based on network function virtualization includes: S1. Obtain the cabinet loop differential pressure sequence, error correction count sequence, orchestration event timestamp, and candidate service chain set of each candidate takeover host cabinet, and determine the reconstruction observation window and reference window based on the orchestration event timestamp. S2. Calculate the differential pressure ripple index based on the differential pressure sequence of the cabinet circuit within the reconstructed observation window and the reference window; S3. Calculate the error correction peak cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window; S4. Based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, calculate the physical interference risk. S5. Based on the physical interference risk, score each candidate service chain in the candidate service chain set, select the candidate service chain with the lowest score for flow cutting, and output the post-reconstruction consistency value.

[0006] Preferably, the reconstruction observation window and the reference window are determined based on the timestamps of the arranged events, including: Obtain the orchestration trigger time and the flow cut completion time, and determine the time period from the orchestration trigger time to the flow cut completion time as the reconstruction observation window; Obtain the timestamp of the most recent service chain change or instance lifecycle change event immediately preceding the orchestration trigger time as the preceding timestamp, and determine the time period from the preceding timestamp to the orchestration trigger time as the baseline window.

[0007] Preferably, the differential pressure ripple index is calculated based on the differential change of the cabinet loop differential pressure sequence within the reconstructed observation window and the reference window, including: The difference between the differential pressure values ​​of adjacent timestamps in the differential pressure sequence of the computer cabinet circuit is used to obtain the differential pressure differential sequence; The median absolute deviation of the differential pressure differential sequences within the reconstructed observation window and the reference window is calculated respectively to obtain the robust scales of the reconstructed observation window and the reference window. The minimum absolute value of the non-zero differential pressure differential sequence within the reference window is obtained as the first adaptive stabilizing term; The differential pressure ripple index is obtained by dividing the robust scale of the reconstructed observation window by the sum of the robust scale of the reference window and the first adaptive stability term.

[0008] Preferably, the error correction peak cluster index is calculated based on the incremental change of the error correction count sequence within the reconstructed observation window and the reference window, including: The error correction counting sequence includes the forward error correction counting sequence and the cyclic redundancy check counting sequence; Calculate the count increments of adjacent timestamps in the forward error correction count sequence and the cyclic redundancy check count sequence respectively to obtain the forward error correction count increment sequence and the cyclic redundancy check count increment sequence; The median of the forward error correction count increment sequence and the cyclic redundancy check count increment sequence within the reference window are used as the first peak discrimination baseline and the second peak discrimination baseline, respectively. The incremental samples within the reconstructed observation window that are larger than the first peak discrimination baseline are constructed into a forward error correction peak sample set; The incremental samples within the reconstructed observation window that are larger than the second peak discrimination baseline are constructed into a cyclic redundancy check peak sample set. The minimum sum of the forward error correction count increment and the cyclic redundancy check count increment within the reconstructed observation window that is greater than zero is obtained as the second adaptive stabilizing term; The summation of the increments of the forward error correction peak sample set is divided by the sum of the total increments of the forward error correction within the reconstructed observation window and the second adaptive stabilization term to obtain the first energy proportion. The summation of the increments of the cyclic redundancy check peak sample set is divided by the sum of the total increments of the cyclic redundancy check within the reconstructed observation window and the second adaptive stabilization term to obtain the second energy proportion. The maximum value between the first energy percentage and the second energy percentage is used as the error correction spike cluster index.

[0009] Preferably, the physical interference risk is calculated based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, including: The absolute value of the difference between adjacent timestamps in the cabinet loop differential pressure sequence is used as the first calculation sequence, and the sum of the forward error correction count increment and the cyclic redundancy check count increment of the corresponding timestamp is used as the second calculation sequence. The median time span of adjacent samples is obtained as the sampling interval. The duration of the reconstructed observation window is divided by the sampling interval and rounded up to determine the lag range of the correlation calculation. Within the lag range, the normalized cross-correlation sequence of the first and second calculated sequences is calculated, and the maximum non-negative correlation value in the normalized cross-correlation sequence is taken as the synchronization strength. Multiplying the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity yields the physical interference risk.

[0010] Preferably, based on the physical interference risk level, a risk score is assigned to each candidate service chain in the candidate service chain set, and the candidate service chain with the lowest score is selected for traffic cutting, including: For each candidate service chain in the candidate service chain set, obtain the corresponding set of landing hosts; Calculate the physical interference risk of each candidate takeover host in the set of landing hosts; Obtain the byte count increment of the uplink port of each candidate takeover host within the baseline window, and calculate the traffic weight of each candidate takeover host by the proportion of the byte count increment of a single candidate takeover host to the total byte count increment of the landing host set. The risk score of the corresponding candidate service chain is obtained by multiplying the traffic weight of each candidate host with the corresponding physical interference risk. The candidate service chain with the lowest risk score is selected as the target service chain, and traffic is switched to the target service chain.

[0011] Preferably, the output of the reconstructed posterior consistency quantity includes: Obtain the actual duration of the system health check cycle record, and extend the cut-off completion time backward by the actual duration of the system health check cycle record to form a posterior window; The risk score of the target service chain is recalculated within the posterior window to obtain the posterior risk score. The difference between the posterior risk score of the target service chain and the risk score of the target service chain before the flow cut is output as the posterior consistency quantity of the reconstruction.

[0012] To address the above problems, the present invention also provides a rapid reconfiguration system for core network service chains based on network function virtualization, the system comprising: The data acquisition module is used to acquire the differential pressure sequence of the cabinet loop of each candidate takeover host, the error correction count sequence of the candidate takeover host, the orchestration event timestamps and the candidate service chain set, and to determine the reconstruction observation window and the reference window based on the orchestration event timestamps. The differential pressure ripple calculation module is used to calculate the differential pressure ripple index based on the differential change of the cabinet circuit differential pressure sequence within the reconstructed observation window and the reference window. The error correction spike calculation module is used to calculate the error correction spike cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window. The risk calculation module is used to calculate the physical interference risk based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window. The flow cut and reconstruction decision module is used to score the risk of each candidate service chain in the candidate service chain set based on the physical interference risk, select the candidate service chain with the lowest score for flow cut, and output the post-reconstruction consistency value.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention acquires the differential pressure sequence of the cabinet loop and the error correction count sequence of the candidate host to be taken over, and calculates the differential pressure ripple index and the error correction spike cluster index based on time windows, thereby extracting the synchronization intensity of the two within the reconstruction observation window, and finally fusing them to construct the physical interference risk quantity; it accurately quantifies the cross-layer synchronization characteristics of the short-term hydraulic transient of the liquid-cooled cabinet cooling loop and the communication error of the host port physical layer, effectively identifying the deep correlation between hardware transient fluctuations and network link transmission anomalies; based on this quantitative index, the system can accurately distinguish between short-term physical layer instability and software layer processing overload, fundamentally avoiding misjudgment of orchestration failure caused by the lack of analysis perspective, and providing a reliable physical layer interference quantification basis for the stable reconstruction of the service chain.

[0014] 2. This invention further performs a comprehensive risk score on each candidate service chain in the candidate service chain set based on the physical interference risk quantity, combined with traffic weights, and strictly selects the candidate service chain with the lowest score to perform the flow switching operation; it directly transforms the cross-layer physical interference risk into the selection criterion for the landing host, ensuring that core network service traffic is diverted to the safest node least affected by physical layer transient disturbances, significantly reducing the probability of repeated flow switching and reconstruction oscillations caused by incorrect landing point selection; at the same time, this invention achieves closed-loop verification of the risk assessment caliber and risk change trend before and after flow switching by outputting the post-reconstruction consistency quantity, greatly improving the convergence speed of network self-healing, and effectively ensuring the continuity of core network services and the high-quality service experience of users. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating a method for rapid reconfiguration of core network service chains based on network function virtualization provided by this invention; Figure 2 A functional module diagram of a core network service chain rapid reconfiguration system based on network function virtualization provided by the present invention; Figure 3 This is a schematic diagram of the liquid-cooled cabinet structure provided by the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example: This example provides a method for rapid reconfiguration of core network service chains based on network function virtualization. See [link to example]. Figure 1Specifically, including: S1. Obtain the cabinet loop differential pressure sequence, error correction count sequence, orchestration event timestamp, and candidate service chain set of each candidate takeover host cabinet, and determine the reconstruction observation window and reference window based on the orchestration event timestamp. S2. Calculate the differential pressure ripple index based on the differential pressure sequence of the cabinet circuit within the reconstructed observation window and the reference window; S3. Calculate the error correction peak cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window; S4. Based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, calculate the physical interference risk. S5. Based on the physical interference risk, score each candidate service chain in the candidate service chain set, select the candidate service chain with the lowest score for flow cutting, and output the post-reconstruction consistency value.

[0018] In embodiments of the present invention, the differential pressure sequence of the rack loop of each candidate takeover host, the error correction count sequence of the candidate takeover host, the orchestration event timestamp, and the candidate service chain set are obtained, and the reconstruction observation window and the reference window are determined based on the orchestration event timestamp, including: In the high-density deployment of network function virtualization core network in liquid-cooled cabinets, the network function orchestrator of the cloud-based core network acts as the execution entity to perform data acquisition and time window determination operations. The cabinet loop differential pressure sequence is collected in real time by the loop differential pressure sensors configured in the liquid-cooled cabinets where each candidate takeover host is located. The sensors acquire the differential pressure value of the cabinet technical cooling system loop at a sampling frequency of 50 Hz and upload it to the cabinet cooling control system in real time. The cooling control system matches the corresponding timestamp for each differential pressure value according to the network unified time synchronization UTC time format and stores them in an orderly manner to form the cabinet loop differential pressure sequence corresponding to each cabinet.

[0019] The error correction count sequence of the candidate takeover host is continuously collected by the port performance counter of the uplink optical port of each candidate takeover host. The port performance counter obtains the forward error correction count and cyclic redundancy check count of the uplink port at a sampling interval of 100 milliseconds. Similarly, the timestamp of each collected count value is matched according to the UTC time format and continuously recorded to form the error correction count sequence corresponding to each candidate takeover host. The orchestration event timestamp and the candidate service chain set are generated and stored autonomously by the network function orchestrator. When the network function orchestrator detects that a hardware failure of a virtual network function instance in the core network triggers the service chain reconstruction requirement, the orchestrator generates and records the orchestration trigger time of the event at the same time as triggering the reconstruction orchestration event. It will also automatically generate a candidate service chain set containing multiple ordered virtual network function sequences and corresponding candidate takeover host landing point mappings based on the current network topology, server resource availability and deployment constraints of the virtual network function in the core network. At the same time, the orchestrator estimates and determines the completion time of the traffic switching of this reconstruction based on the historical execution time of the core network service chain reconstruction and the current network load status.

[0020] The orchestrator directly determines the continuous time period from the orchestration trigger time to the flow cut completion time as the reconstruction observation window. Then, the orchestrator retrieves its own full event log database and searches the database for the most recent service chain change event or instance lifecycle change event that occurred immediately before the orchestration trigger time. The timestamp of the retrieved event is determined as the preceding timestamp. The orchestrator further determines the continuous time period from the preceding timestamp to the orchestration trigger time as the baseline window. Throughout the data acquisition process, the rack cooling control system, host port performance counter, and network function orchestrator use the same network time source to achieve time synchronization, ensuring that the timestamps of the collected rack loop differential pressure sequence and error correction count sequence are completely aligned with the orchestration event timestamps, thus ensuring the consistency of the time dimension in subsequent data calculations.

[0021] It should be noted that the candidate takeover host refers to the server entity selected to carry the original business flow or continue the original session processing during rapid service chain reconstruction. The rack in which the host is located is the physical rack where the server is installed and runs. The rack loop differential pressure sequence refers to the numerical sequence of the pressure difference between the supply and return liquid lines in the liquid cooling loop of the physical rack, changing over time at continuous sampling moments. This pressure difference reflects the flow resistance and flow redistribution state of the coolant between the rack manifold and its branches. Short-term fluctuations in differential pressure correspond to hydraulic transients in the loop and flow fluctuations caused by pump control regulation. The Network Function Orchestrator is the core hardware and software management unit of the orchestration layer under the Network Function Virtualization architecture. Deployed on top of the core network cloud infrastructure, it is the core scheduling entity connecting the underlying computing / storage / network physical resources, virtual network function instances, and upper-layer core network service requirements. It is also the core execution device for realizing the full lifecycle management and dynamic reconstruction of the core network service chain in the network function virtualization environment. It is a conventional and core network management component in the fields of communication network data transmission and network service orchestration technology.

[0022] It should be noted that the error correction count sequence of the candidate takeover host refers to the numerical sequence of transmission error correction related counts accumulated over time at continuous sampling moments of the host's uplink port or network card port. The count changes reflect the strength of bit errors and error correction processing activities occurring during electrical or optical transmission. Count increment spikes typically indicate a decrease in the instantaneous signal margin or an increase in noise disturbance. The orchestration event timestamp refers to the standard time stamp recorded by the network function virtualization orchestration system when key events such as triggering reconfiguration and completing flow switching occur, used to characterize the start and end boundaries of the reconfiguration process and the sequence of events. The candidate service chain set is... This refers to a set of multiple service chain schemes available under the same reconstruction task. Each service chain scheme includes network function instances and their landing host mappings arranged in the order of business processing, representing the processing paths that business packets pass through in sequence in the network. The reconstruction observation window refers to the time interval during reconstruction determined by the orchestration event timestamps, used to centrally analyze the synchronous relationship between differential pressure fluctuations and error correction changes during the reconstruction process. The baseline window refers to the stable operating time interval immediately before the reconstruction is triggered, used to provide a reference level for differential pressure and error correction in the non-reconstruction state, so that subsequent calculations can characterize the degree of abnormal fluctuations introduced by reconstruction in terms of relative changes.

[0023] In an embodiment of the present invention, the differential pressure ripple index is calculated based on the differential change of the cabinet circuit differential pressure sequence within the reconstructed observation window and the reference window, including: Based on the differential pressure sequence of the cabinet loops after time synchronization and alignment in the previous step, the differential pressure ripple index is calculated. The differential pressure sequence of the cabinet loops is arranged in ascending order of timestamps, and the difference in differential pressure values ​​corresponding to adjacent timestamps is calculated sequentially to obtain the differential pressure differential sequence. The calculation formula is as follows: ,in This represents the i-th differential pressure value in the differential pressure sequence of the cabinet loop. This represents the differential pressure value corresponding to the i-th timestamp in the differential pressure sequence of the cabinet loop. The differential pressure value corresponding to the (i-1)th timestamp in the differential pressure sequence of the cabinet circuit is calculated using this formula. The reason for using this formula is to accurately capture the instantaneous change in differential pressure of the liquid-cooled cabinet circuit by directly subtracting the differential pressure values ​​of adjacent timestamps, intuitively reflecting the real-time fluctuation characteristics of differential pressure ripple, and providing basic data for subsequent ripple scale quantification.

[0024] Extract all differential pressure values ​​within the reconstructed observation window and the reference window respectively, and calculate the median absolute deviation of the differential pressure differential sequences within the two windows to obtain the robust scales of the reconstructed observation window and the reference window. The formula for calculating the median absolute deviation is as follows: The formula for calculating the robust scale of the reconstructed observation window is as follows: The formula for calculating the robustness metric of the reference window is: ,in This corresponds to the differential pressure differential sequence within the window. To perform the median calculation, To reconstruct a robust scale for the observation window, As a benchmark window robustness measure, To reconstruct the observation window, As the reference window, The reason for using the median absolute deviation to calculate the ripple scale is that this statistic is more robust than variance and standard deviation, and can effectively avoid the interference of extreme differential pressure values ​​caused by short-term hydraulic transients in the liquid cooling circuit on the ripple scale calculation results, thus adapting to the non-normal data characteristics of differential pressure sequences.

[0025] All non-zero differential pressure differential values ​​are selected from the differential pressure differential sequence within the reference window. The absolute values ​​of these non-zero differential pressure differential values ​​are calculated, and the minimum value is taken as the first adaptive stability term. This method is used to determine the first adaptive stability term because the reference window represents the normal operating period before service chain reconstruction. Determining the stability term based on the non-zero differential pressure differential data within this window better reflects actual operating conditions. It also fundamentally avoids calculation anomalies such as a zero denominator in subsequent division operations and eliminates the need for manually preset fixed values, ensuring the adaptiveness of the calculation. Finally, the robust scale of the reconstruction observation window is used as the numerator, and the sum of the robust scale of the reference window and the first adaptive stability term is used as the denominator. Division is then performed to obtain the differential pressure ripple index, calculated using the following formula: ,in The differential pressure ripple index is calculated using this formula because by reconstructing the ratio of the ripple scale of the observation window to the reference window, it is possible to accurately quantify the enhancement of the differential pressure ripple of the liquid-cooled cabinet circuit within the observation window relative to the normal operating state of the equipment. The larger the ratio, the more severe the differential pressure ripple within the reconstruction window, and the higher the risk of physical layer interference from the liquid-cooled circuit. At the same time, the addition of the first adaptive stabilizing term to the denominator eliminates the calculation risk of the denominator being zero, and does not significantly affect the normal scale ratio calculation, thus ensuring the accuracy and effectiveness of the differential pressure ripple index.

[0026] It should be noted that the robust scale of the reconstruction observation window refers to the fluctuation scale obtained by statistically measuring the adjacent sampled difference values ​​of the differential pressure sequence of the cabinet loop within the reconstruction observation window. It is based on the deviation of the difference value from the median level and is characterized by the absolute deviation of the median, thus reflecting the typical ripple intensity of the differential pressure caused by the hydraulic transients and flow redistribution of the loop during reconstruction and is insensitive to occasional extreme spikes. The robust scale of the reference window refers to the fluctuation scale obtained by calculating the adjacent sampled difference values ​​of the differential pressure sequence of the cabinet loop in the same way within the reference window. It is used to characterize the natural fluctuation background of the cabinet cooling loop under stable operating conditions, so that subsequent judgments can use this background as a reference to measure whether the ripple is abnormally enhanced during reconstruction. The first adaptive stability term is to select the minimum value greater than zero from the set of absolute values ​​of differential pressure difference values ​​within the reference window as the numerical stability quantity. It reflects the minimum non-zero differential pressure change amplitude that can be observed under the reference operating conditions. It is used to avoid numerical divergence caused by the denominator being zero or too small when performing ratio calculations, while maintaining that this stability quantity comes from the actual sampling resolution and the minimum discernible change of the loop.

[0027] It should be noted that the differential pressure ripple index is a normalized ratio index composed of the robust scale of the reconstructed observation window, the robust scale of the reference window, and the first adaptive stability term. It measures the relative enhancement of differential pressure ripple in the cabinet circuit during the reconstruction process by amplifying the differential pressure fluctuation scale relative to the reference background fluctuation scale. The larger the index value, the more significant the differential pressure fluctuation within the reconstruction window and the more likely it is to correspond to the disturbance of the network carrying environment by the hydraulic transient of the circuit. Among them, the differential pressure ripple of the cabinet loop refers to the periodic or quasi-periodic fluctuation of the pressure difference between the supply and return lines of the liquid-cooled cabinet during continuous sampling. Essentially, it is a slight oscillation of the pressure difference caused by the speed and pressure control of the pump, the opening of the cabinet manifold and branch valves, the redistribution of branch flow among different cold plates, and the change of local resistance in the loop with temperature and flow rate during the flow of coolant. The ripple is usually superimposed on the stable pressure difference level and is reflected by the alternation of positive and negative values ​​or amplitude fluctuations of adjacent sampling differences. The changes in the amplitude and frequency of the ripple reflect the strength of the hydraulic transient in the loop and the stability of the flow state. When the ripple is enhanced, it means that the pressure difference in the loop is more likely to fluctuate significantly in a short period of time, which may cause short-term fluctuations in the heat dissipation conditions and the operating margin of the devices in the cabinet.

[0028] In an embodiment of the present invention, the error correction peak cluster index is calculated based on the incremental change of the error correction count sequence within the reconstructed observation window and the reference window, including: Subsequent calculations are performed based on the error correction count sequence of the candidate takeover host after time synchronization and alignment. This error correction count sequence specifically includes a forward error correction count sequence and a cyclic redundancy check (CRC) count sequence, both of which share the same timestamp dimension as the cabinet loop differential pressure sequence and are arranged in ascending order of timestamps. The difference between the count values ​​corresponding to adjacent timestamps is calculated sequentially for the forward error correction count sequence to obtain the forward error correction count increment sequence. The calculation formula is as follows: Then, perform the same adjacent subtraction operation on the cyclic redundancy check (CRC) count sequence to obtain the CRC count increment sequence, calculated using the following formula: ,in This represents the i-th increment value in the forward error correction count increment sequence. Let be the forward error correction count value corresponding to the i-th timestamp in the forward error correction count sequence. For the forward error correction counting sequence, the th The forward error correction count value corresponding to each timestamp This represents the i-th increment value in the cyclic redundancy check (CRC) count increment sequence. This represents the cyclic redundancy check (CRC) count value corresponding to the i-th timestamp in the CRC count sequence. For the cyclic redundancy check counting sequence, the th The reason for using the adjacent difference formula to calculate the cyclic redundancy check count value corresponding to each timestamp is to directly capture the real-time change of the error correction count in each sampling interval of the uplink port of the candidate takeover host, accurately reflect the real-time occurrence status of communication errors at the port physical layer, and provide basic data that reflects the actual error changes of the port for subsequent peak discrimination. At the same time, this calculation method is consistent with the calculation logic of differential pressure differential sequence, ensuring the time dimension consistency of subsequent synchronization strength analysis.

[0029] From the generated forward error correction count increment sequence and cyclic redundancy check count increment sequence, extract all increment values ​​falling within the reference window to form the forward error correction count increment subsequence and the cyclic redundancy check count increment subsequence within the reference window. Then calculate the median of each subsequence. Use the median of the forward error correction count increment subsequence within the reference window as the first peak discrimination baseline. The calculation formula is as follows: The median of the incremental subsequence of cyclic redundancy check counts within the baseline window is used as the second peak discrimination baseline, calculated using the following formula: ,in The first peak is the discrimination baseline. The second peak is the discrimination baseline. To perform the median calculation, As the reference window, The reason for using the median of the incremental sequence within the reference window as the peak discrimination baseline is that the reference window is the normal operating period of the device before the service chain reconstruction. The median of its incremental sequence can accurately characterize the error correction count increment baseline level under normal port communication conditions. Moreover, as a robust statistic, the median can effectively avoid the interference of a small number of abnormal incremental values ​​that occasionally occur during normal operation periods on the baseline judgment. It is also completely adaptively generated based on the actual operating data of the device, adapting to the port communication status under different racks and different loads.

[0030] It should be noted that the forward error correction count sequence refers to the numerical sequence formed by the accumulation of forward error correction related counts recorded at the uplink port or network card port of the candidate takeover host over time during continuous sampling. This count reflects the intensity of the error correction mechanism's correction when bit flips or symbol decision errors occur during transmission. The larger the increment of the count over time, the more frequently the correctable errors occur per unit time, which usually corresponds to signal attenuation, crosstalk, noise increase, or reduced link margin caused by device temperature drift. The cyclic redundancy check count sequence refers to the numerical sequence formed by the accumulation of cyclic redundancy check error related counts recorded at the same port over time during continuous sampling. This count reflects the change in the number of times the receiver finds inconsistencies when checking the arriving data frames. The larger the increment of the check error, the more non-negligible frame-level errors there are per unit time, which usually corresponds to an increase in physical layer bit errors, link transient instability, or abnormal working state of interface devices.

[0031] It should be noted that the first peak discrimination baseline and the second peak discrimination baseline refer to the median level calculated from the forward error correction count increment sequence and the cyclic redundancy check count increment sequence within the reference window, respectively. They are used as reference thresholds to determine whether the incremental samples within the reconstruction observation window belong to anomalous enhancement. The first peak discrimination baseline is used to measure the enhancement degree of the forward error correction increment relative to the stable operating background, and the second peak discrimination baseline is used to measure the enhancement degree of the cyclic redundancy check increment relative to the stable operating background. Both are adaptively formed from the actual background data.

[0032] All incremental values ​​falling within the reconstruction observation window are extracted from the forward error correction count increment sequence and the cyclic redundancy check count increment sequence, respectively. Then, for each forward error correction count increment value within the reconstruction observation window, it is determined whether it is greater than the first peak discrimination baseline. All increment samples that meet the determination criteria are integrated to construct a forward error correction peak sample set. The same determination operation is performed for each cyclic redundancy check count increment value within the reconstruction observation window. All increment samples greater than the second peak discrimination baseline are integrated to construct a cyclic redundancy check peak sample set. Then, a point-by-point summation operation is performed on the forward error correction count increment and cyclic redundancy check count increment corresponding to all timestamps within the reconstruction observation window to obtain the error correction count increment sum corresponding to each timestamp. From all error correction count increment sums, results with values ​​greater than zero are selected, and the minimum value among them is taken as the second adaptive stabilizing term. The stabilizing term is adaptively determined based on the actual error correction increment data within the reconstruction observation window. The stabilizing term can avoid the calculation anomaly of zero denominator in the subsequent energy proportion calculation, and only the minimum value of the increment sum greater than zero is selected, which will not significantly interfere with the normal energy proportion calculation results.

[0033] Summing all incremental values ​​within the forward error correction spike sample set, using this sum as the numerator, summing all incremental values ​​of the forward error correction count increment sequence within the reconstructed observation window to obtain the total forward error correction increment, and using the sum of the total forward error correction increment and the second adaptive stabilization term as the denominator, performing a division operation to obtain the first energy proportion, the calculation formula is as follows: Then, sum all the increment values ​​within the Cyclic Redundancy Check (CRBC) spike sample set, using this sum as the numerator. Summing all the increment values ​​of the CRBC count increment sequence within the reconstructed observation window yields the total CRBC increment. The sum of the total CRBC increment and the second adaptive stabilization term is used as the denominator, and division is performed to obtain the second energy percentage. The calculation formula is as follows: ,in As the first energy percentage, This is the set of forward error correction spike samples. This is the summation of all incremental values ​​within the forward error correction spike sample set. To reconstruct the total increment of forward error correction within the observation window, As the second energy percentage, For the cyclic redundancy check peak sample set, To perform a cyclic redundancy check on the summation of all incremental values ​​within the spike sample set, The total increment of cyclic redundancy check within the reconstructed observation window is calculated using this formula. The reason for using this formula to calculate the energy ratio is to accurately quantify the energy ratio of the peak cluster within the reconstructed observation window by using the ratio of the total increment of the peak sample to the total increment within the reconstructed observation window. The larger the ratio, the higher the proportion of the peak cluster in the overall error correction increment, and the more significant the error peaks of the port physical layer. Adding a second adaptive stabilizing term to the denominator further avoids the computational risk of the denominator being zero, ensuring the effectiveness of the calculation.

[0034] The calculated first energy percentage and the second energy percentage are compared numerically, and the maximum value is selected as the error correction spike cluster index. The reason is that forward error correction and cyclic redundancy check are two different error checking methods in the port physical layer. The occurrence of a significant spike cluster in either of them will cause physical layer interference to the service chain reconstruction. Taking the maximum value of the two energy percentages can accurately capture the most significant error correction spike cluster characteristics of the port physical layer within the reconstruction observation window, objectively reflect the error risk level of the port caused by physical layer disturbance, and provide an accurate port error quantification index for the subsequent calculation of synchronization risk.

[0035] It should be noted that the forward error correction spike sample set refers to the set of incremental sample indices selected from the forward error correction count increment sequence within the reconstruction observation window. Each incremental sample satisfies the condition that its value is greater than the first spike discrimination baseline, thus representing a time segment in which forward error correction activity is significantly enhanced relative to the stable background during reconstruction. The sum of the increments corresponding to the set reflects the contribution intensity of the forward error correction enhancement segment to the total error correction increment within the reconstruction window. The cyclic redundancy check spike sample set refers to the set of incremental sample indices selected from the cyclic redundancy check count increment sequence within the reconstruction observation window. Each incremental sample satisfies the condition that its value is greater than the second spike discrimination baseline, thus representing a time segment in which frame-level check failure is significantly enhanced relative to the stable background during reconstruction. The sum of the increments corresponding to the set reflects the contribution intensity of the check error enhancement segment to the total check error increment within the reconstruction window.

[0036] It should be noted that the first energy ratio refers to the ratio of the sum of increments corresponding to the peak sample set of forward error correction to the total increment of forward error correction within the reconstruction observation window. A second adaptive stabilizing term, formed by the minimum non-zero increment within the reconstruction observation window, is added to the denominator to ensure the numerical stability of the ratio calculation. This ratio is used to characterize the concentration of the forward error correction enhancement segment in the error correction activity throughout the reconstruction window. A larger ratio indicates that the error correction activity is more concentrated in a few peak time segments. The second energy ratio refers to the ratio of the sum of increments corresponding to the peak sample set of cyclic redundancy check to the total increment of cyclic redundancy check within the reconstruction observation window. Similarly, a second adaptive stabilizing term is added to the denominator to avoid ratio distortion caused by an excessively small total increment. This ratio is used to characterize the concentration of the check error enhancement segment in the check error activity throughout the reconstruction window. A larger ratio indicates that the check error is more concentrated in a few peak time segments.

[0037] It should be noted that the error correction spike cluster index is a comprehensive index obtained by taking the maximum value of the first energy proportion and the second energy proportion within the same reconstruction observation window. This index is used to characterize the degree of cluster concentration of error correction-related abnormal spikes on a uniform scale. When the error correction spike cluster index is large, it indicates that at least one type of count in forward error correction or cyclic redundancy check shows obvious spike clustering phenomenon within the reconstruction window, which is more likely to correspond to a sudden increase in error correction activity caused by short-term instability of the link physical layer. Among them, the phenomenon of spike clustering refers to the phenomenon that a certain count or measurement quantity does not change uniformly and slowly on the continuous sampling time axis, but instead appears in a few adjacent or nearby sampling moments within the reconstruction observation window, forming one or more sudden enhancement segments. This is manifested by the incremental sequence continuously showing values ​​significantly higher than the background level within these segments and maintaining a lower level outside the segments, so that the total increment is mainly contributed by these short-term segments. The clustering usually corresponds to the system entering an unstable state due to transient disturbances in a short period of time. For example, the link signal margin decreases in a short time, leading to a sudden increase in error correction activities or a sudden increase in verification failures, and then recovers to near the background level after the disturbance subsides. This causes the abnormal energy to be clustered rather than dispersed in time.

[0038] In embodiments of the present invention, the physical interference risk is calculated based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, including: For each timestamp within the reconstructed observation window, the absolute value of the cabinet loop differential pressure differential is calculated. The result is used as the corresponding value for the first calculation sequence to construct the first calculation sequence. The calculation formula is as follows: Then, for the same timestamp within the reconstructed observation window, the forward error correction count increment and the cyclic redundancy check count increment are summed. The summation result is used as the corresponding value of the second calculation sequence to construct the second calculation sequence. The calculation formula is as follows: ,in This represents the value corresponding to the i-th timestamp in the first calculation sequence. To reconstruct the differential pressure differential value of the cabinet loop corresponding to the i-th timestamp within the observation window, This represents the value corresponding to the i-th timestamp in the second calculation sequence. To reconstruct the forward error correction count increment value corresponding to the i-th timestamp within the observation window, To reconstruct the cyclic redundancy check (CRC) count increment value corresponding to the i-th timestamp within the observation window, the absolute value of the differential pressure differential is taken when constructing the first calculation sequence. This is because the interference of the cabinet loop differential pressure ripple on the physical layer is determined by the differential pressure fluctuation amplitude. The positive or negative sign of the differential pressure differential only represents the fluctuation direction and does not affect the synchronization analysis with the error correction increment. Taking the absolute value can accurately characterize the actual fluctuation intensity of the differential pressure ripple. The reason for summing the two types of error correction increments when constructing the second calculation sequence is that both forward error correction and CRC are error correction methods for the port physical layer. The sum of their increments can comprehensively characterize the total error increment level of the port physical layer under a single timestamp. Furthermore, both sequences are constructed based on the same timestamp dimension, ensuring that there is a corresponding error correction increment for each timestamp. and There are no data misalignments or missing data, which lays the foundation for a unified time dimension for subsequent correlation calculations.

[0039] Extract all timestamps within the reconstructed observation window, arranged in ascending order. Calculate the time span between any two adjacent timestamps to obtain a set of time spans for all adjacent samples. Perform a median operation on this set and use the result as the sampling interval. The median is used to calculate the sampling interval because, in actual deployments, there may be slight time deviations in the sampling operations of sensors and counters, leading to minor fluctuations in adjacent sampling time spans. The median, as a robust statistic, effectively avoids interference from individual abnormal time spans on the calculated sampling interval, making the sampling interval more closely reflect actual sampling conditions. Calculate the total duration of the reconstructed observation window, i.e., the time difference between the flow cut completion time and the orchestration trigger time. Divide the total duration by the sampling interval The calculated result is rounded up to obtain the lag step size for the correlation calculation, and the lag range for the correlation calculation is determined as starting from... arrive integer range ,in For the lagging step size, The moment when the flow cut is completed. To schedule the trigger times, The reason for using an up-rounding operation to determine the lag step size for the lag range in the correlation calculation is to ensure that the total time length corresponding to the lag range can completely cover the total duration of the reconstructed observation window, avoiding the problem of insufficient time length due to rounding, and preventing the omission of some time dimension synchronization relationships in subsequent correlation calculations. The reason for determining the lag range as an interval containing positive and negative integers is that the cross-layer synchronization relationship between the cabinet circuit differential pressure ripple and the port error correction spike cluster may have three situations: differential pressure ripple occurs ahead, error correction spike cluster occurs ahead, or the two are completely synchronized. The positive and negative lag range can fully consider all the above possible synchronization timing relationships, ensuring the comprehensiveness and accuracy of subsequent synchronization strength calculations.

[0040] Within the lag range, calculate the normalized cross-correlation value corresponding to each integer lag value to obtain the normalized cross-correlation sequence. The calculation formula is as follows:

[0041] in lagged value The corresponding normalized cross-correlation value, For the i-th value in the first calculation sequence, The mean of the first calculated sequence is... For the second calculated sequence at lag value The i-th corresponding value, The mean of the second calculated sequence is used; normalized cross-correlation can eliminate the dimensional differences between differential pressure ripple and error correction increment, so that the correlation value is within the standard numerical range, which is convenient for determining the degree of synchronization.

[0042] All non-negative normalized cross-correlation values ​​were selected from the calculated normalized cross-correlation sequences, and the maximum value was chosen as the synchronization strength. The reason for determining the synchronization strength is that a negative correlation value indicates that the differential pressure ripple and the error correction spike cluster are changing in opposite directions. This situation will not produce physical layer cooperative interference and has no negative impact on service chain reconstruction. Only non-negative correlation values ​​can characterize the synchronous change characteristics of the two. Taking the maximum value can accurately capture the strongest synchronous correlation between the two within the reconstruction observation window. The larger the value of the synchronization strength, the stronger the cross-layer synchronization of the differential pressure ripple and the error correction spike cluster, and the higher the risk of interference caused by short-term instability of the physical layer.

[0043] The differential pressure ripple index and error correction spike cluster index obtained in the previous calculation are multiplied together with the synchronization intensity obtained in this calculation to obtain the physical interference risk. The calculation formula is as follows: ,in This represents the risk of physical interference. The differential pressure ripple index, To correct the peak cluster index, The reason for using this formula to calculate the physical interference risk is that the interference risk of the physical layer to the service chain reconstruction is determined by three core factors: the severity of differential pressure ripple, the significance of error correction spike clusters, and the synchronization strength between the two. These three factors are coupled, and the physical layer interference risk will only reach a high level when the differential pressure ripple is severe, the error correction spike clusters are significant, and the synchronization between the two is strong. By multiplying the results, the quantitative indicators of the three dimensions can be integrated to accurately and objectively characterize the reconstruction interference risk of a single candidate takeover host caused by short-term instability of the physical layer, providing a core quantitative basis for the subsequent risk scoring of candidate service chains.

[0044] It should be noted that the first calculation sequence refers to the time series formed by taking the absolute value of the differential pressure difference between adjacent sampling times in the cabinet loop differential pressure sequence. It is used to characterize the transient change amplitude of differential pressure in each sampling interval within the reconstruction observation window. The larger the value, the more drastic the loop differential pressure change in that interval and the more likely it corresponds to the enhancement of hydraulic transients. The second calculation sequence refers to the time series formed by summing the forward error correction count increment and the cyclic redundancy check count increment at the sampling time corresponding to the first calculation sequence. It is used to characterize the combined enhancement degree of link error correction activities and frame-level check failure activities in the same sampling interval. The larger the value, the more concentrated the error events caused by link physical layer instability in that interval.

[0045] It should be noted that the lag range refers to the allowed time alignment offset interval when calculating the correlation between the first and second calculated sequences. It is determined by the duration of the reconstruction observation window and the sampling interval, and is used to cover the possible propagation and response delays of differential pressure changes on port error performance, enabling the correlation calculation to search for the optimal alignment relationship between the two sequences within a reasonable time offset. The synchronization strength refers to the maximum non-negative correlation value obtained after calculating the normalized cross-correlation between the first and second calculated sequences within the lag range. It is used to characterize the temporal synchronization consistency between the amplitude of the differential pressure transient change and the degree of enhancement of the error correction activity. A higher synchronization strength indicates that the two sequences exhibit more coordinated changes with rising and falling at a certain time offset, and are more consistent with the specific characteristics of differential pressure ripple and error correction spike synchronization. The physical interference risk is a comprehensive quantity obtained by multiplying the differential pressure ripple index, the error correction spike cluster index and the synchronization strength. It is used to reflect the possibility and strength of the adverse effects of the synchronization enhancement of hydraulic transients in the liquid cooling loop and link error activities on the rapid reconfiguration process of the service chain in a unified scale. A higher risk indicates that the differential pressure fluctuation is more significant, the error correction spike is more concentrated and the synchronization between the two is stronger within the reconfiguration window, which may lead to instability and misjudgment in the flow cut and health determination stages.

[0046] In an embodiment of the present invention, based on the physical interference risk level, a risk score is performed on each candidate service chain in the candidate service chain set, the candidate service chain with the lowest score is selected for traffic cutting, and a post-reconstruction consistency level is output, including: The orchestrator retrieves a pre-generated landing point mapping table for each candidate service chain. This mapping table contains the association between each virtual network function in the candidate service chain and the corresponding deployed candidate takeover host. The orchestrator extracts all associated candidate takeover hosts from the mapping table, and after deduplication, forms a unique set of landing point hosts corresponding to the candidate service chain. Subsequently, for each candidate takeover host in the set of landing point hosts, the orchestrator performs the following complete steps: rack loop differential pressure sequence processing, differential pressure ripple index calculation, error correction count increment sequence processing, error correction spike cluster index calculation, synchronization strength calculation, and physical interference risk quantity synthesis. Based on the differential pressure data of the rack where the host is located and the error correction count data of its own uplink port, the orchestrator independently completes the calculation to obtain the unique physical interference risk quantity corresponding to each candidate takeover host.

[0047] The performance counters of each candidate takeover host's uplink port are retrieved within the baseline window. The byte count values ​​corresponding to the start and end times of the baseline window are extracted and subtracted to obtain the byte count increment of a single candidate takeover host within the baseline window. This operation is performed on all candidate takeover hosts within the landing host set. Then, the traffic weight of each candidate takeover host is obtained by calculating the ratio of the byte count increment of a single candidate takeover host to the total byte count increment of the landing host set. The calculation formula is as follows: ,in The traffic weight of a single candidate takeover host in the set of landing hosts. This represents the increment of the uplink port byte count for the candidate takeover host within the baseline window. This refers to the set of landing hosts corresponding to the current candidate service chain. For any candidate takeover host in the set of landing hosts, The formula for calculating traffic weights is based on the total byte count increment of the landing host set. The baseline window is the normal operation period before service chain reconstruction. The byte count increment during this period can truly reflect the actual business traffic carrying capacity of each candidate takeover host in the service chain. The higher the traffic carrying capacity of a host, the greater its impact on the overall operation of the service chain. The weights constructed through this ratio can make the subsequent service chain risk scoring fit the actual business distribution. At the same time, the reason for deduplicating the landing host set is to avoid the same candidate takeover host being counted repeatedly because it is associated with multiple virtual network functions, ensuring that the calculation objects of traffic weights and physical interference risk are unique and correspond.

[0048] It should be noted that the landing host set refers to a set of candidate takeover host entities corresponding to the deployment landing points of each network function instance in the candidate service chain. Each host in this set carries at least one network function instance or its critical processing path in the candidate service chain, thus collectively forming the physical server-level carrying foundation of the candidate service chain. This set is used to characterize the resource distribution and network carrying distribution of the candidate service chain among different hosts. The byte count increment refers to the increment value obtained by reading the cumulative byte count from the byte counter of the uplink port of the candidate takeover host within the baseline window and calculating the difference. It is reflected in the baseline window. Within a sampling interval, the data transmission volume actually carried by the uplink port of the host is denoted as byte count increment. The larger the byte count increment, the more service traffic the host carries during that time period. Traffic weight refers to the ratio obtained by normalizing the byte count increment of a single candidate takeover host in the landing host set to the total byte count increment of the landing host set. It is used to characterize the relative contribution of the host to the overall traffic carried by the candidate service chain. The larger the traffic weight, the higher the traffic proportion of the host in the candidate service chain. Therefore, when performing risk scoring on the candidate service chain, the physical interference risk of the host should have a relatively greater impact on the chain-level risk score.

[0049] For a single candidate service chain in the candidate service chain set, the traffic weight of each candidate takeover host in its landing host set is multiplied point-by-point with the corresponding physical interference risk. Then, all product results are summed to obtain the risk score for that candidate service chain. The calculation formula is as follows: ,in For candidate service chains Risk score, For candidate service chains A single candidate takeover host corresponds to the set of landing hosts. For candidate service chains The corresponding set of landing hosts, Assign traffic weights to candidate takeover host $h$ The physical interference risk of the candidate takeover host $h$ is calculated using this formula because the overall physical interference risk of the candidate service chain is determined by the risks of each landing host. Different hosts carry different amounts of business traffic in the service chain, and traffic weight can accurately characterize the impact of a single host on the overall operation of the service chain. By weighted summing of traffic weight and single-machine risk, the service chain risk score can be made to fit the actual business carrying capacity distribution characteristics and objectively reflect the actual interference risk level of the entire service chain. The higher the traffic carrying capacity of a host, the more significant its risk impact on the overall service chain, thus ensuring the accuracy and practicality of the risk score.

[0050] A global numerical comparison of the risk scores of all candidate service chains in the candidate service chain set is performed, and the candidate service chain with the smallest risk score is selected as the target service chain. The reason for selecting the target service chain is that the magnitude of the risk score directly corresponds to the overall risk level of the service chain due to short-term instability interference from the physical layer. The smaller the score, the lower the possibility that the service chain will be affected by differential pressure ripple and error correction spike cluster synchronous interference after reconstruction. This can minimize the problems of flow sliding failure and repeated reconstruction oscillation caused by physical layer interference, which meets the core requirements of reliability and convergence for rapid reconstruction of service chains.

[0051] Once the target service chain is determined, the Network Function Orchestrator immediately issues deployment and traffic switching instructions to the Core Network Virtual Infrastructure Manager. The Virtual Infrastructure Manager first instantiates and deploys each virtual network function in the target service chain, establishes virtual links, and configures network policies according to the instructions, ensuring that the new service chain meets the functional requirements of core network service transmission. Then, using a seamless traffic switching method, the core network service traffic carried by the original service chain is gradually forwarded from the original chain to the target service chain according to a preset diversion rate. During the traffic switching process, the orchestrator collects key indicators such as packet loss rate and latency jitter in real time through the core network data plane monitoring module to ensure the continuity and transmission quality of service traffic. Once all service traffic from the original service chain has been diverted to the target service chain and all key indicators are within normal threshold ranges, the traffic switching is considered complete. The Virtual Infrastructure Manager reports the completion status of the traffic switching to the Network Function Orchestrator, which synchronously updates the core network service chain operation status log, recording information such as the target service chain identifier, the time of traffic switching completion, and the operating parameters of each landing host, providing data support for subsequent post-verification consistency checks.

[0052] It should be noted that the risk score is a comprehensive evaluation metric calculated for each candidate service chain in the candidate service chain set. It characterizes the degree to which the chain is affected by liquid cooling loop differential pressure ripple and port error correction spike synchronization interference within the service chain fast reconfiguration window. This evaluation metric is obtained by weighting and summing the physical interference risk of each candidate takeover host in the set of landing hosts corresponding to the candidate service chain according to their traffic weight. This results in the host with a higher proportion of traffic carrying a greater contribution to the risk of the candidate service chain. The smaller the risk score, the less likely the candidate service chain is to suffer from flow switching instability and health judgment error caused by synchronization interference within the reconfiguration window. The target service chain is the service chain scheme determined by selecting the candidate service chain with the smallest risk score after calculating the risk scores for each candidate service chain in the candidate service chain set. This service chain scheme corresponds to a set of network function instances and their landing host mappings arranged in the order of business processing, and serves as the final reconfiguration result for actually executing flow switching and continuing business carrying.

[0053] Retrieve the system health check cycle configuration item stored in the core network operation and maintenance configuration database. This configuration item is the actual duration of the health check task execution interval continuously recorded during the actual operation of the core network, denoted as... This duration is dynamically configured and updated in real time by the operation and maintenance system based on the core network service type and operational requirements. This actual duration is directly extracted as the time span of the posterior window. Subsequently, the orchestrator obtains the posterior window completion time returned by the virtual infrastructure manager after the preceding posterior window operation is completed. The time when the flow cut is completed is taken as the starting time, and the actual duration of the system health check cycle is extended forward. The continuous time period from the completion of the flow cut to the end of the subsequent delay is defined as the posterior window, and the time range of the posterior window is... Next, the orchestrator, within the posterior window, recalculates the target service chain risk score using the same calculation logic and steps as before the traffic cut. This involves first collecting the rack loop differential pressure sequence of each host in the target service chain's corresponding landing host set, and the error correction count sequence of the host's uplink port. Then, it sequentially completes the time window determination, differential pressure ripple index calculation, error correction spike cluster index calculation, synchronization strength calculation, and single-host physical interference risk calculation. Simultaneously, it retrieves the byte count increment of each host's uplink port within the posterior window to recalculate the traffic weight. Finally, it multiplies and sums the traffic weight of each host with the corresponding posterior physical interference risk to obtain the target service chain's posterior risk score. The calculation dimensions, operation steps, and data collection methods of this posterior risk score are highly consistent with the target service chain risk score before the traffic cut, ensuring direct numerical comparability between the two.

[0054] The calculated posterior risk score is subtracted from the target service chain risk score calculated before the traffic cut, and this difference is used as the reconstruction posterior consistency quantity. After the calculation of the reconstruction posterior consistency quantity is completed, multi-dimensional output and storage operations are immediately performed. On the one hand, the reconstruction posterior consistency quantity is integrated with relevant data such as the posterior window time range, the target service chain identifier, the posterior physical interference risk quantity of each host in the landing host set, and the posterior traffic weight, and written into the core network service chain reconstruction full log library for persistent storage. On the other hand, the reconstruction posterior consistency quantity is pushed to the core network operation and maintenance monitoring platform in real time for visualization on the monitoring interface. At the same time, it is associated with the target service chain's operating status indicators to provide operation and maintenance personnel with a quantitative basis for judging the reconstruction effect. The numerical change of the reconstruction posterior consistency quantity can directly reflect the risk trend of physical layer interference to the target service chain after the traffic cut, providing accurate quantitative data support for subsequent service chain operating status review and reconstruction strategy optimization.

[0055] It should be noted that the post-hoc risk score refers to the process of re-collecting the differential pressure sequence and error correction count sequence of the cabinet loop corresponding to the target service chain within the post-hoc window after the target service chain completes the flow cut. The physical interference risk of each landing host is obtained using the same calculation method as before the flow cut, and then the chain-level comprehensive evaluation is obtained by weighted summing of each physical interference risk based on the traffic weight of the landing host set. This evaluation is used to characterize the degree of differential pressure ripple and error correction spike synchronization interference experienced by the target service chain during the actual operation phase after the flow cut, thereby reflecting whether the environment is stable after the flow cut and whether the risk has fallen back to an acceptable level. The post-reconstruction consistency value refers to the difference between the post-reconstruction risk score obtained by the target service chain within the post-reconstruction window and the risk score obtained before the cutoff within the reconstruction observation window. It is used to characterize the consistency and trend of risk assessment before and after reconstruction. When the post-reconstruction consistency value is positive and has a large magnitude, it indicates that the chain-level risk after the cutoff is relatively higher than before the cutoff, suggesting that the reconstruction has not eliminated synchronization interference or that new physical disturbances have increased. When the post-reconstruction consistency value is negative and has a large magnitude, it indicates that the chain-level risk after the cutoff is relatively lower than before the cutoff, suggesting that the synchronization between differential pressure fluctuation and error correction peaks is weakened and the carrying environment tends to be stable after the cutoff.

[0056] like Figure 2 The diagram shown is a functional block diagram of a core network service chain rapid reconfiguration system based on network function virtualization provided in an embodiment of the present invention.

[0057] In this embodiment, the functions of each module / unit are as follows: The data acquisition module is used to acquire the differential pressure sequence of the cabinet loop of each candidate takeover host, the error correction count sequence of the candidate takeover host, the orchestration event timestamps and the candidate service chain set, and to determine the reconstruction observation window and the reference window based on the orchestration event timestamps. The differential pressure ripple calculation module is used to calculate the differential pressure ripple index based on the differential change of the cabinet circuit differential pressure sequence within the reconstructed observation window and the reference window. The error correction spike calculation module is used to calculate the error correction spike cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window. The risk calculation module is used to calculate the physical interference risk based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window. The flow cut and reconstruction decision module is used to score the risk of each candidate service chain in the candidate service chain set based on the physical interference risk, select the candidate service chain with the lowest score for flow cut, and output the post-reconstruction consistency value.

[0058] like Figure 3As shown, a technical cooling system loop is installed inside the liquid-cooled cabinet. This loop is arranged along the cabinet frame, forming a supply and return liquid circulation path. A cabinet manifold is installed within the loop to distribute the coolant from the main path to multiple branch pipes. Each branch pipe is connected to multiple server nodes, allowing the coolant to flow sequentially through the server cold plates within each server node to remove the heat generated during server operation. The coolant then returns to the cabinet manifold via the loop and merges back into the main path of the loop, completing the circulation. A pump control and regulation unit is installed on the technical cooling system loop to adjust the flow rate and differential pressure within the loop. The system is designed to maintain stable operation of the coolant loop when server node load changes or branch flow is redistributed. Differential pressure sensors are installed between the supply and return fluid paths to collect pressure difference changes between the supply and return fluids and form a differential pressure sequence for the cabinet loop, thus characterizing the pressure difference fluctuation state of the cabinet loop during operation. The arrows in the figure indicate the flow direction of coolant in the main loop path, cabinet manifold, and branch pipelines. Through the above structural configuration, coolant distribution and circulation recovery for multiple server nodes in the liquid-cooled cabinet can be realized, and a data acquisition basis is provided for the operation status analysis based on cabinet loop pressure difference changes.

[0059] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for rapid reconfiguration of core network service chains based on network function virtualization, characterized in that, include: S1. Obtain the cabinet loop differential pressure sequence, error correction count sequence, orchestration event timestamp, and candidate service chain set of each candidate takeover host cabinet, and determine the reconstruction observation window and reference window based on the orchestration event timestamp. S2. Calculate the differential pressure ripple index based on the differential pressure sequence of the cabinet circuit within the reconstructed observation window and the reference window; S3. Calculate the error correction peak cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window; S4. Based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, calculate the physical interference risk. S5. Based on the physical interference risk, score each candidate service chain in the candidate service chain set, select the candidate service chain with the lowest score for flow cutting, and output the post-reconstruction consistency value.

2. The method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 1, characterized in that, The reconstructed observation window and baseline window are determined based on the timestamps of the orchestrated events, including: Obtain the orchestration trigger time and the flow cut completion time, and determine the time period from the orchestration trigger time to the flow cut completion time as the reconstruction observation window; Obtain the timestamp of the most recent service chain change or instance lifecycle change event immediately preceding the orchestration trigger time as the preceding timestamp, and determine the time period from the preceding timestamp to the orchestration trigger time as the baseline window.

3. The method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 2, characterized in that, Based on the differential change of the cabinet loop differential pressure sequence within the reconstructed observation window and the reference window, the differential pressure ripple index is calculated, including: The difference between the differential pressure values ​​of adjacent timestamps in the differential pressure sequence of the computer cabinet circuit is used to obtain the differential pressure differential sequence; The median absolute deviation of the differential pressure differential sequences within the reconstructed observation window and the reference window is calculated respectively to obtain the robust scales of the reconstructed observation window and the reference window. The minimum absolute value of the non-zero differential pressure differential sequence within the reference window is obtained as the first adaptive stabilizing term; The differential pressure ripple index is obtained by dividing the robust scale of the reconstructed observation window by the sum of the robust scale of the reference window and the first adaptive stability term.

4. The method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 3, characterized in that, Based on the incremental changes in the error correction count sequence within the reconstructed observation window and the baseline window, the error correction peak cluster index is calculated, including: The error correction counting sequence includes the forward error correction counting sequence and the cyclic redundancy check counting sequence; Calculate the count increments of adjacent timestamps in the forward error correction count sequence and the cyclic redundancy check count sequence respectively to obtain the forward error correction count increment sequence and the cyclic redundancy check count increment sequence; The median of the forward error correction count increment sequence and the cyclic redundancy check count increment sequence within the reference window are used as the first peak discrimination baseline and the second peak discrimination baseline, respectively. The incremental samples within the reconstructed observation window that are larger than the first peak discrimination baseline are constructed into a forward error correction peak sample set; The incremental samples within the reconstructed observation window that are larger than the second peak discrimination baseline are constructed into a cyclic redundancy check peak sample set. The minimum sum of the forward error correction count increment and the cyclic redundancy check count increment within the reconstructed observation window that is greater than zero is obtained as the second adaptive stabilizing term; The summation of the increments of the forward error correction peak sample set is divided by the sum of the total increments of the forward error correction within the reconstructed observation window and the second adaptive stabilization term to obtain the first energy proportion. The summation of the increments of the cyclic redundancy check peak sample set is divided by the sum of the total increments of the cyclic redundancy check within the reconstructed observation window and the second adaptive stabilization term to obtain the second energy proportion. The maximum value between the first energy percentage and the second energy percentage is used as the error correction spike cluster index.

5. A method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 4, characterized in that, Based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window, the physical interference risk is calculated, including: The absolute value of the difference between adjacent timestamps in the cabinet loop differential pressure sequence is used as the first calculation sequence, and the sum of the forward error correction count increment and the cyclic redundancy check count increment of the corresponding timestamp is used as the second calculation sequence. The median time span of adjacent samples is obtained as the sampling interval. The duration of the reconstructed observation window is divided by the sampling interval and rounded up to determine the lag range of the correlation calculation. Within the lag range, the normalized cross-correlation sequence of the first and second calculated sequences is calculated, and the maximum non-negative correlation value in the normalized cross-correlation sequence is taken as the synchronization strength. Multiplying the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity yields the physical interference risk.

6. The method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 5, characterized in that, Based on the physical interference risk level, each candidate service chain in the candidate service chain set is scored, and the candidate service chain with the lowest score is selected for traffic cutting, including: For each candidate service chain in the candidate service chain set, obtain the corresponding set of landing hosts; Calculate the physical interference risk of each candidate takeover host in the set of landing hosts; Obtain the byte count increment of the uplink port of each candidate takeover host within the baseline window, and calculate the traffic weight of each candidate takeover host by the proportion of the byte count increment of a single candidate takeover host to the total byte count increment of the landing host set. The risk score of the corresponding candidate service chain is obtained by multiplying the traffic weight of each candidate host with the corresponding physical interference risk. The candidate service chain with the lowest risk score is selected as the target service chain, and traffic is switched to the target service chain.

7. A method for rapid reconfiguration of core network service chains based on network function virtualization according to claim 6, characterized in that, Output the reconstructed posterior consistency measure, including: Obtain the actual duration of the system health check cycle record, and extend the cut-off completion time backward by the actual duration of the system health check cycle record to form a posterior window; The risk score of the target service chain is recalculated within the posterior window to obtain the posterior risk score. The difference between the posterior risk score of the target service chain and the risk score of the target service chain before the flow cut is output as the posterior consistency quantity of the reconstruction.

8. A rapid reconfiguration system for core network service chains based on network function virtualization, applied in the rapid reconfiguration method for core network service chains based on network function virtualization as described in any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to acquire the differential pressure sequence of the cabinet loop of each candidate takeover host, the error correction count sequence of the candidate takeover host, the orchestration event timestamps and the candidate service chain set, and to determine the reconstruction observation window and the reference window based on the orchestration event timestamps. The differential pressure ripple calculation module is used to calculate the differential pressure ripple index based on the differential change of the cabinet circuit differential pressure sequence within the reconstructed observation window and the reference window. The error correction spike calculation module is used to calculate the error correction spike cluster index based on the incremental changes in the error correction count sequence within the reconstructed observation window and the reference window. The risk calculation module is used to calculate the physical interference risk based on the differential pressure ripple index, the error correction spike cluster index, and the synchronization intensity of the cabinet loop differential pressure sequence and the error correction count sequence within the reconstruction observation window. The flow cut and reconstruction decision module is used to score the risk of each candidate service chain in the candidate service chain set based on the physical interference risk, select the candidate service chain with the lowest score for flow cut, and output the post-reconstruction consistency value.

Images