A method for accessing heterogeneous nodes for monitoring terminals

By constructing a parsing syntax tree and dynamically adjusting the time slice allocation weights in conjunction with kernel parameters, the problem of uneven task distribution among heterogeneous data streams in multi-core kernels is solved, achieving efficient task allocation and system stability.

CN123114725APending Publication Date: 2026-11-17NANCHANG INST OF TECH
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611010213.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-08
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies pose risks of core overload and queue overflow when processing concurrent heterogeneous data streams, and cannot effectively distribute tasks evenly, leading to a decrease in system stability and efficiency.

Method used

By constructing a parsing syntax tree to extract the protocol type identifier of the data stream, encapsulating it into a standardized computation thread task, and dynamically adjusting the time slice allocation weight in combination with kernel clock frequency, voltage and transient temperature, the task is adaptively and evenly distributed among multi-core kernels, avoiding core overload and queue overflow.

Benefits of technology

It achieves non-linear matching of heterogeneous data streams between multi-core kernels, avoiding core overload and queue overflow, and improving system stability and processing efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114725A_ABST
    Figure CN123114725A_ABST
Patent Text Reader

Abstract

This invention relates to the field of processor task scheduling and discloses a method for monitoring heterogeneous node access in terminals. The method includes: a multi-protocol adaptation middleware constructs a parsing syntax tree and extracts communication protocol type identifiers; encapsulates multimodal data streams into standardized computation thread tasks; a task scheduler calculates the expected computational intensity based on the communication protocol type identifier and data packet length, and calculates time slice allocation weights by combining the ready queue length, nominal main frequency, operating voltage, and transient temperature of each physical computing core; accordingly, it rewrites the control block pointer and attaches the task to the tail of the ready list of the qualified kernels. This invention achieves feedforward calculation of computing power requirements, enabling non-linear matching between high load streams and kernel allocation weights, avoiding ready queue overflow and crashes, effectively suppressing the risk of high-heat frequency limiting, and evenly distributing the load.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a heterogeneous node access method for monitoring terminals, belonging to the field of processor task scheduling technology. Background Technology

[0002] Currently, when centrally processing data interaction from multiple monitoring terminals, the common practice is to use a task scheduler to distribute the access threads corresponding to multimodal monitoring data streams to different physical computing cores to maintain real-time data throughput. After the monitoring data packets arrive at the multi-protocol adaptation middleware, they are converted into task control blocks and attached to the processor's ready list. The physical computing core sequentially polls the ready list and allocates time slices to parse heterogeneous data payloads. As the complexity of the monitoring environment increases, the data streams generated by concurrently accessing heterogeneous terminals exhibit asymmetric processing overhead. The computation cycle required for parsing time-series text is relatively small, while the cycle required for parsing image streams increases significantly. When high computational load streams flood the processor in a short period of time, the traditional static distribution mechanism causes instantaneous task accumulation in the ready list of specific physical cores, resulting in thread scheduling cascading delays. This not only triggers thermal frequency limiting protection due to internal hardware heat accumulation but also leads to receiver buffer saturation, causing packet loss and system stability degradation.

[0003] To alleviate task backlog, simply increasing the number of physical cores or linearly improving the clock speed leads to increased processor power consumption. Under limited on-site heat dissipation conditions, this can easily accelerate the formation of localized hot spots, and even force frequency throttling, resulting in a reduction in the overall system computing power. Existing technologies face a fundamental contradiction when handling concurrent heterogeneous streams: the imbalance between the heterogeneous access data parsing load and the processor's electrothermal and physical limits. Linear improvements at the hardware level are subject to objective constraints in power consumption and heat dissipation, and the task scheduling logic at the software level also has shortcomings. For example, the paper with publication number CN111274024A... A Chinese invention patent application discloses a method and data structure for optimizing the average load of the ready queue based on the CFS scheduler. It is based on conventional general computing scenarios, but the underlying logic relies on the accumulation of decay time within the historical scheduling cycle. When processing multi-source concurrent monitoring data, it lacks feedforward calculation of the computational intensity of data stream protocol parsing before thread distribution, and cannot predict the asymmetric computing power overhead between heterogeneous protocol streams of text and images. Moreover, the scheduling strategy is fragmented and coupled in a closed loop with the physical state fluctuations of processor main frequency voltage and core transient temperature. Under the condition of electric thermal frequency limiting, scheduling mismatch occurs, and the risk of core overload and queue overflow under sudden high load flow cannot be avoided.

[0004] Therefore, the technical problem to be solved by this invention is how to calculate the computational density of protocol parsing based on the characteristics of concurrent heterogeneous data packets, and dynamically reconstruct the ready list topology by coupling the main frequency, voltage and transient temperature of multi-core kernels in a closed loop, so as to achieve adaptive balanced distribution of heterogeneous access streams among multiple physical kernels and thus avoid core overload and queue overflow. Summary of the Invention

[0005] To address the problems in the background art, the technical solution of the present invention is as follows: A method for heterogeneous node access for monitoring terminals, comprising the following steps: Step S1: The multi-protocol adaptation middleware loads the protocol library to build a parsing syntax tree, extracts the communication protocol type identifier of the multimodal data stream, and encapsulates the multimodal data stream into a normalized computation thread task containing the communication protocol type identifier. Step S2: The task scheduler reads the communication protocol type identifier and the current data packet byte length of the normalized computing thread task, matches the single-byte parsing instruction cycle constant through the computing power consumption matrix, and multiplies the constant by the byte length to obtain the expected computing intensity. In step S3, the task scheduler obtains the current ready queue length, nominal main frequency, current operating voltage, and transient temperature of each physical computing core through the ready queue register, nominal main frequency register, operating voltage register, and kernel transient temperature register. When the transient temperature is not greater than the safe temperature threshold, the nominal main frequency is multiplied by the current operating voltage to obtain the working capacity product, and this product is divided by the sum of the current ready queue length and the number 1 to obtain the time slice allocation weight. When the transient temperature is greater than the safe temperature threshold, the safe temperature threshold is divided by the transient temperature and then multiplied by the working capacity product to obtain the controlled capacity product, and this controlled capacity product is divided by the sum of the current ready queue length and the number 1 to obtain the reduced time slice allocation weight. Based on the expected computational intensity and the time slice allocation weight, the control block pointer of the normalized computing thread task is rewritten, and the normalized computing thread task is attached to the tail of the physical computing core ready list that meets the state judgment conditions.

[0006] Preferably, the process of data encapsulation and translation by the multi-protocol adaptation middleware in step S1 includes: step S11, loading the containerized protocol parsing plugin corresponding to the communication protocol type identifier, and constructing a complete parsing syntax tree for the multimodal data stream; step S12, using the complete parsing syntax tree to eliminate the instruction cycle differences between heterogeneous protocols, converting the payload in the current data packet into a normalized time-series text data stream with a unified semantic structure, and encapsulating the normalized time-series text data stream into an independent normalized computation thread task.

[0007] Preferably, the process of calculating the time slice allocation weight in step S3 includes: step S31, the task scheduler reads the number of currently queued task control blocks in the ready queue register inside each physical computing core in each control cycle, and uses it as the current ready queue length; step S32, the transient temperature is compared with the safe temperature threshold, and when the transient temperature is greater than the safe temperature threshold, the temperature rise reverse feedback suppression is enabled to forcibly reduce the scheduling priority of physical computing cores with high heat frequency limiting risk.

[0008] Preferably, the process of rewriting the control block pointer of the normalized computing thread task in step S3 includes: step S33, when the current ready queue length is greater than the activation load threshold and the transient temperature exceeds the safe temperature threshold, the task scheduler intercepts the thread assignment signal for the target physical computing kernel; step S34, rewriting the control block pointer of the normalized computing thread task, and attaching the newly accessed multimodal data stream to the tail of the physical computing kernel ready list that meets the state judgment conditions, bypassing the physical computing kernel with high thermal frequency limiting risk.

[0009] Preferably, after the multi-protocol adaptation middleware translates the multimodal data stream into a normalized computation thread task, the method further includes: step S4, adjusting the kernel ready list allocation priority for image data streams with higher computational overhead than time-series text data streams, and prioritizing the allocation of image data streams to the physical computation kernel with the highest nominal main frequency and the shortest current ready queue length.

[0010] Preferably, when calculating the time slice allocation weight in step S3, the method further includes: step S35, introducing a dynamic decay coefficient based on the kernel transient temperature change rate into the time slice allocation weight, wherein the dynamic decay coefficient changes monotonically with the increase of the kernel transient temperature change rate.

[0011] Preferably, after step S3, the method further includes a hardware health status assessment process: step S37, periodically record the historical transient temperature sequence and historical power consumption yield frequency of each physical computing core in a continuous scheduling cycle, and calculate the physical core health degradation index characterizing the core heat accumulation rate; step S38, when the physical core health degradation index continuously exceeds the safe aging threshold, output a core warning signal, and at the same time limit the maximum allowable ready queue length of the physical computing core.

[0012] Preferably, the method for determining whether the physical computing kernel meets the state determination conditions in step S3 includes: step S36, calculating the temperature difference between the transient temperature of the current physical computing kernel and the safe temperature threshold, and determining that the physical computing kernel meets the state determination conditions when the temperature difference is greater than the temperature drop margin threshold and the current ready queue length is less than the ready queue baseline length.

[0013] Preferably, the ready queue register, nominal main frequency register, operating voltage register, and core transient temperature register are all connected to the task scheduler through the chip's internal micro-rail hardware bus, and the hardware parameters of each physical computing core are read under the control of the scheduling clock signal with a clock frequency of 100MHz to 1000MHz.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. In the heterogeneous node access of the monitoring terminal, by capturing the characteristics of the data stream to be processed and the length of the data packets in real time, the scheduler converts the heterogeneous input stream with variable length and asymmetric protocol characteristics into computing thread tasks to be allocated at the logical layer. Based on the single-byte parsing instruction cycle constant, the expected computing intensity is determined, thereby breaking the disconnect between communication layer access and multi-core computing resource allocation. This allows the system to complete a fine-grained feedforward calculation of computing power requirements before the input stream is written to the general stack buffer. By changing the linked list mounting topology of the task control block, the allocation weight of high load streams and core kernels is non-linearly matched, avoiding overflow and crash of the ready queue of the physical kernel when concurrent data floods in.

[0015] 2. By integrating the current ready queue length, nominal clock frequency, and operating voltage of each physical core, and combining the real-time comparison of the kernel transient temperature with the safe temperature threshold, this method establishes a discrete gradient adjustment mechanism based on the hardware transient power consumption yield characteristics within the processor. At this time, the allocation weight generates reverse feedback suppression with the temperature rise curve, forcibly changes the pointer of the task control block of the thread to be allocated, guides the new access flow to automatically bypass the core node at high heat frequency limiting risk, and attaches it to the tail of the cold kernel ready list in place, thus eliminating the instantaneous collapse of computing node processing capacity and task stack deadlock loss caused by heat accumulation in the physical core from the underlying logic.

[0016] 3. To address the challenge of asymmetric protocol parsing overhead caused by multi-protocol concurrency, this method dynamically loads protocol libraries through a lightweight containerized architecture to extract parsing syntax trees, and encapsulates multimodal data streams into a standardized data format with standard semantic annotations. This directly eliminates the imbalance in computational overhead caused by differences in the number of instruction cycles of heterogeneous protocol parsing plugins. Combined with the weight allocation of the thread ready queue, this allows image streams with high computational overhead and time-series text streams with low computational density to achieve non-linear alignment in the kernel ready list. This eliminates reliance on retransmission or packet loss tolerance of upper-layer communication links, and achieves direct coupling between the parsing task stream and the physical kernel load at the node scheduling level. Attached Figure Description

[0017] Figure 1 This is a flowchart of a heterogeneous node access method for a monitoring terminal according to the present invention; Figure 2 This is an interactive diagram of a heterogeneous node access method for a monitoring terminal according to the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] A method for heterogeneous node access for monitoring terminals includes the following steps: Step S1: The multi-protocol adaptation middleware loads the protocol library to build a parsing syntax tree, extracts the communication protocol type identifier of the multimodal data stream, and encapsulates the multimodal data stream into a normalized computation thread task containing the communication protocol type identifier. Step S2: The task scheduler reads the communication protocol type identifier and the current data packet byte length of the normalized computing thread task, matches the single-byte parsing instruction cycle constant through the computing power consumption matrix, and multiplies the constant by the byte length to obtain the expected computing intensity. In step S3, the task scheduler obtains the current ready queue length, nominal main frequency, current operating voltage, and transient temperature of each physical computing core through the ready queue register, nominal main frequency register, operating voltage register, and kernel transient temperature register. When the transient temperature is not greater than the safe temperature threshold, the nominal main frequency is multiplied by the current operating voltage to obtain the working capacity product, and this product is divided by the sum of the current ready queue length and the number 1 to obtain the time slice allocation weight. When the transient temperature is greater than the safe temperature threshold, the safe temperature threshold is divided by the transient temperature and then multiplied by the working capacity product to obtain the controlled capacity product, and this controlled capacity product is divided by the sum of the current ready queue length and the number 1 to obtain the reduced time slice allocation weight. Based on the expected computational intensity and the time slice allocation weight, the control block pointer of the normalized computing thread task is rewritten, and the normalized computing thread task is attached to the tail of the physical computing core ready list that meets the state judgment conditions.

[0021] Preferably, the process of data encapsulation and translation by the multi-protocol adaptation middleware in step S1 includes: step S11, loading the containerized protocol parsing plugin corresponding to the communication protocol type identifier, and constructing a complete parsing syntax tree for the multimodal data stream; step S12, using the complete parsing syntax tree to eliminate the instruction cycle differences between heterogeneous protocols, converting the payload in the current data packet into a normalized time-series text data stream with a unified semantic structure, and encapsulating the normalized time-series text data stream into an independent normalized computation thread task.

[0022] Preferably, the process of calculating the time slice allocation weight in step S3 includes: step S31, the task scheduler reads the number of currently queued task control blocks in the ready queue register inside each physical computing core in each control cycle, and uses it as the current ready queue length; step S32, the transient temperature is compared with the safe temperature threshold, and when the transient temperature is greater than the safe temperature threshold, the temperature rise reverse feedback suppression is enabled to forcibly reduce the scheduling priority of physical computing cores with high heat frequency limiting risk.

[0023] Preferably, the process of rewriting the control block pointer of the normalized computing thread task in step S3 includes: step S33, when the current ready queue length is greater than the activation load threshold and the transient temperature exceeds the safe temperature threshold, the task scheduler intercepts the thread assignment signal for the target physical computing kernel; step S34, rewriting the control block pointer of the normalized computing thread task, and attaching the newly accessed multimodal data stream to the tail of the physical computing kernel ready list that meets the state judgment conditions, bypassing the physical computing kernel with high thermal frequency limiting risk.

[0024] Preferably, after the multi-protocol adaptation middleware translates the multimodal data stream into a normalized computation thread task, the method further includes: step S4, adjusting the kernel ready list allocation priority for image data streams with higher computational overhead than time-series text data streams, and prioritizing the allocation of image data streams to the physical computation kernel with the highest nominal main frequency and the shortest current ready queue length.

[0025] Preferably, when calculating the time slice allocation weight in step S3, the method further includes: step S35, introducing a dynamic decay coefficient based on the kernel transient temperature change rate into the time slice allocation weight, wherein the dynamic decay coefficient changes monotonically with the increase of the kernel transient temperature change rate.

[0026] Preferably, after step S3, the method further includes a hardware health status assessment process: step S37, periodically record the historical transient temperature sequence and historical power consumption yield frequency of each physical computing core in a continuous scheduling cycle, and calculate the physical core health degradation index characterizing the core heat accumulation rate; step S38, when the physical core health degradation index continuously exceeds the safe aging threshold, output a core warning signal, and at the same time limit the maximum allowable ready queue length of the physical computing core.

[0027] Preferably, the method for determining whether the physical computing kernel meets the state determination conditions in step S3 includes: step S36, calculating the temperature difference between the transient temperature of the current physical computing kernel and the safe temperature threshold, and determining that the physical computing kernel meets the state determination conditions when the temperature difference is greater than the temperature drop margin threshold and the current ready queue length is less than the ready queue baseline length.

[0028] Preferably, the ready queue register, nominal main frequency register, operating voltage register, and core transient temperature register are all connected to the task scheduler through the chip's internal micro-rail hardware bus, and the hardware parameters of each physical computing core are read under the control of the scheduling clock signal with a clock frequency of 100MHz to 1000MHz.

[0029] Example 1: The method claimed in this invention is applicable to industrial monitoring conditions that require high-frequency vibration and temperature cycling. When a multi-source monitoring terminal performs continuous observation of ecological and environmental indicators over a wide area and around the clock, the multi-protocol adaptation middleware loads the protocol library and constructs a parsing syntax tree, extracts the communication protocol type identifier of the multimodal data stream, and encapsulates the multimodal data stream into a normalized computation thread task containing the communication protocol type identifier. The task scheduler reads the communication protocol type identifier of the normalized computation thread task and the current data packet byte length, matches the single-byte parsing instruction cycle constant through the computing power consumption matrix, and multiplies the constant by the byte length to obtain the desired computational density. The computing power consumption matrix is ​​stored in the read-only memory area of ​​the scheduler. Its row dimension corresponds to different communication protocol type identifiers, and its column dimension corresponds to different processor microarchitecture feature values. The elements in the matrix are the single-byte parsing instruction cycle constants, which are the average number of instructions pre-measured by a benchmark test program based on typical industrial protocols under the target kernel architecture. For example, for a time-series text stream, the corresponding constant is set to 10-50 cycles. Periods / byte; for image streams involving encoding and decoding calculations, the corresponding constant is set to 500-2000 periods / byte to ensure that the load intensity can be feedforward quantized through simple multiplication operations before the task is mounted. All parameters are aligned with the working range of the method claimed in this invention. The task scheduler, through the chip's internal hardware bus, under the control of a scheduling clock signal with a clock frequency of 100MHz to 1000MHz, cyclically reads the ready queue register, nominal main frequency register, operating voltage register, and kernel transient temperature register inside each physical computing core to obtain the current ready queue length, nominal main frequency, current operating voltage, and transient temperature of each physical computing core. The nominal main frequency is selected as 2.0GHz, and the transient temperature is measured by a temperature sensor inside the physical computing core at a sampling frequency of not less than 10Hz. After system startup, hardware performance is calibrated within a preset period using a benchmark program stored in read-only memory, running each core under no-load conditions. The task scheduler measures the actual time taken for each core to execute a fixed number of instruction sequences. Compared with theoretical standards, the time required is Compare and calculate performance correction factors. ;in, A dimensionless correction factor between 0.9 and 1.1; and The unit is microseconds; during scheduling decisions, the task scheduler assigns weights... Physical kernel performance correction factor Multiply to obtain the corrected weights. This offsets the deviation in core computing power caused by differences in leakage current or process fluctuations.

[0030] When the transient temperature of the physical computing kernel does not exceed the safe temperature threshold, the task scheduler multiplies the nominal main frequency by the current operating voltage to obtain the working capacity product, and divides this working capacity product by the sum of the current ready queue length and the number 1 to calculate the time slice allocation weight. The calculation formula is as follows: ,in, For the first The weights are assigned to the current time slice of each physical computing kernel. For the first The nominal main frequency of a physical computing core For the first The current operating voltage of each physical computing core. For the first The current ready queue length of each physical computing kernel, where F i With V i The product of F and F is defined as the transient computational potential factor of the kernel, which in engineering practice serves as a surrogate parameter characterizing the highest frequency envelope that the kernel can support under the current voltage swing. i With V i Although their physical dimensions are different, their product can reflect the upper limit of the kernel's computing power supply within the power budget, F i This determines the toggling rate of the logic gate circuit, while V i This determines the energy margin during the flipping process; the two are combined by dividing by L. i+1 This system enables nonlinear prediction of the computing resources expected to be allocated to individual tasks, thereby constructing a mapping path from physical electrical parameters to logical allocation weights at the scheduling level. When the transient temperature of the physical computing kernel exceeds the safe temperature threshold, the task scheduler activates temperature rise inverse feedback suppression, downgrading the scheduling priority of physical computing kernels with high thermal throttling risk. The task scheduler reads the transient temperature... Perform time-domain smoothing filtering by weighted summing of the original temperature value of the current sampling period with the historical sequence of the previous N sampling periods to obtain the filtered transient temperature. The sampling frequency is set to 10Hz, and N is 5; the task scheduler utilizes... It participates in the time slice allocation weight reconstruction, intervenes in the response frequency constraint within the millisecond range, matches the chip package thermal time constant, and suppresses scheduling oscillations caused by temperature transients.

[0031] Divide the safe temperature threshold by the transient temperature and multiply by the working capacity product to obtain the controlled capacity product. Divide this controlled capacity product by the sum of the current ready queue length and the number 1 to calculate the reduced time slice allocation weight. The calculation formula is as follows: ,in, For the first The current reduced time slice allocation weights for each physical computing kernel. For the first The nominal main frequency of a physical computing core For the first The current operating voltage of each physical computing core. The preset safe temperature threshold, For the first Transient temperature of a physical computing kernel. For the first The current ready queue length of each physical computing kernel is determined by the task scheduler, which adds a dynamic decay coefficient based on the kernel transient temperature change rate to the time slice allocation weight. This coefficient decreases monotonically as the kernel transient temperature change rate increases, in order to offset the task list accumulation caused by high computational load.

[0032] When the current ready queue length exceeds the activation load threshold and the transient temperature exceeds the safe temperature threshold, the task scheduler intercepts the thread assignment signal for the target physical computing core, rewrites the control block pointer of the normalized computing thread task, and bypasses the high-temperature throttling risk of the physical computing core by allowing newly accessed multimodal data streams. Based on the expected computational density and time slice allocation weights, it calculates the temperature difference between the current physical computing core's transient temperature and the safe temperature threshold. If the temperature difference is greater than the temperature drop margin threshold and the current ready queue length is less than the ready queue baseline length, the physical computing core is determined to meet the state judgment conditions. The normalized computing thread task is then attached to the tail of the ready list of physical computing cores that meet the state judgment conditions. The task scheduler periodically records the historical transient temperature sequence and historical power yield frequency of each physical computing core to assess hardware health and address performance drift; it also monitors the temperature rise rate of each core under the set load. Changes can be used to determine the performance degradation of the thermally conductive medium or the packaging structure. If the mean exceeds the initial baseline value by 15% over multiple consecutive observation windows, the kernel safety temperature threshold will be lowered. This setting limits the maximum allowed length of the ready queue for the physical computing kernel.

[0033] Image data streams with higher computational overhead than time-series text data streams are prioritized for allocation using a ready list. They are preferentially assigned to the physical computing core with the highest nominal main frequency and the shortest current ready queue length. The task scheduler periodically records the historical transient temperature sequence and historical power yield frequency of each physical computing core in a continuous scheduling cycle, and calculates the physical core health degradation index. When the physical core health degradation index continuously exceeds the safe aging threshold, a kernel warning signal is output, and the maximum allowable ready queue length of the physical computing core is limited. The nonlinear alignment between the heterogeneous access data parsing load and the processor's electrothermal physical limit constraints is achieved through dynamic topology reconstruction of the underlying ready list.

[0034] Example 2: This invention was validated in an engineering environment with heterogeneous monitoring terminals that included load fluctuations and transient thermal effects. By comparing the performance changes of the sample group and the sample group of this invention, the effectiveness of the computing power demand feedforward calculation mechanism and the dynamic topology reconstruction of the ready linked list was verified. The test platform relied on an embedded computing module equipped with a 4-core heterogeneous processor, configured with an external noise generator, and injected random Gaussian white noise sequences into the operating environment to simulate electromagnetic interference in actual industrial sites, while setting a safe temperature threshold. The temperature was set at 85℃, with a temperature drop margin threshold of 15℃ and a ready queue baseline length of 10. This experiment included a comparative sample group A and the present invention sample group B. Comparative sample group A used a fixed polling strategy to allocate tasks, did not perform feedforward calculations of data stream computational intensity, and ignored transient temperature feedback from the processor cores. The present invention sample group B used a normalized computational thread task allocation method, performed task load hedging based on the expected computational intensity, and adjusted weights according to changes in kernel transient temperature. In the initial state of the experimental simulation environment, the ready queue lengths of the four physical computational cores were all 0. The transient temperature was consistently maintained at 45℃, and the operating voltage was uniformly kept constant at 1.2V. During the experiment, a heterogeneous data stream was input at a rate of 1000 task packets per second through an external data generator. The monitoring results of the comparison sample A showed that when the transient temperature climbed to 82℃ due to continuous high-load operation, the physical computing kernel frequently triggered the hardware overheat protection mechanism due to the lack of a power yield strategy, resulting in a forced halt of the execution pipeline. At this time, the total ready queue length of the system in the comparison sample A rapidly increased to 42, some computing thread tasks experienced packet loss, and the task stack recorded deadlock.

[0035] In sample group B of this invention, under the same data load, the task scheduler monitors the transient temperature change of the kernel in real time. When the transient temperature of a certain kernel reaches 75°C, the task scheduler calculates the reduced time slice allocation weight according to the formula. Based on kernel health degradation indicators, some long text data tasks were proactively offloaded to lighter kernels, while image data streams were mapped to the computing kernels with the optimal combination of current clock frequency and ready queue. Experimental data showed that, within the same test duration, the transient temperature peak of sample group B was limited to 78℃, the peak total ready queue length was 8, and there was no task loss throughout the process. To verify the rationality of the numerical range, an out-of-range control sample group C was established, in which the safe temperature threshold was set. Setting the temperature to 120℃ exceeds the material's thermal stability boundary. Experimental results show that when the transient temperature reaches 95℃, due to the lack of physical parameter warnings for the risk of core thermal degradation, some computational core logic units experience bit flip errors, causing the calculation result deviation rate to surge from 0.01% to 5.8%. This verifies the engineering necessity of setting the safe temperature threshold within the range of 85℃. In summary, this invention transforms heterogeneous input streams into computational thread tasks and achieves dynamic load balancing of the task ready queue based on a nonlinear matching weight constructed according to the desired computational density and core physical parameters, thus avoiding the risk of high-temperature frequency limiting. The experimental data shows that, compared with the comparative sample group A, the ready queue overflow risk of sample group B of this invention is reduced by 81% under complex high-load environments, and the task processing stability is improved.

[0036] Example 3: This example combines Figures 1 to 2 This describes a method for accessing heterogeneous nodes for monitoring terminals, such as... Figure 1 As shown, the operations of the multi-protocol adaptation middleware include: constructing a parsing syntax tree to extract communication protocol identifiers, and encapsulating multimodal data streams into normalized computation thread tasks. The results are then passed to the task scheduler, which sequentially executes operations such as reading the communication protocol type identifier and data packet byte length, calculating the desired computational intensity by matching the instruction cycle constant, and obtaining the current kernel ready queue length, voltage, and transient temperature. Next, it determines whether the transient temperature is greater than the safe temperature threshold. If the result is no, it calculates the product of working capacity to obtain the time slice allocation weight. If the result is yes, it calculates the product of controlled capacity to obtain the reduced time slice allocation weight. Afterward, the results of the above two branches are converged and the operation of rewriting the control block pointer of the normalized computation thread task is performed. Finally, the flow is directed to the physical computation kernel and the ready list, completing the operation of attaching it to the tail of the physical kernel ready list that meets the conditions. Figure 2As shown, the multi-source monitoring terminal sends a multimodal data stream to the multi-protocol adaptation middleware. After receiving the data stream, the multi-protocol adaptation middleware sequentially executes operations such as loading the protocol library, constructing the parsing syntax tree, and extracting the communication protocol type identifier. It then sends a normalized computation thread task to the task scheduler. After receiving the task, the task scheduler sequentially executes operations such as reading the type identifier and data packet length, matching the single-byte parsing instruction cycle constant, and calculating the expected computational density.

[0037] Example 4: When the monitoring terminal is processing high-concurrency, multi-source heterogeneous data streams, after the containerized plugin completes semantic parsing, the task scheduling system optimizes task distribution based on the computing power redundancy of specific computing nodes to avoid the asymmetric parsing latency caused by multi-protocol adaptation middleware impacting the general computing stack. When the application layer faces the challenge of sudden access to monitoring tasks leading to a backlog of ready queues in specific processing cores, the scheduler captures data stream packet characteristics in real time, obtains the communication protocol type identifier and byte length, substitutes them into the computing power consumption matrix, and calculates the expected computing intensity of the access task. The task scheduler updates the time slice allocation weight based on the current ready queue length, nominal clock frequency, operating voltage, and transient temperature of the physical computing core. When the transient temperature of the physical computing core exceeds the safe temperature threshold, the scheduler automatically triggers the logical redirection of the task mounting strategy. By calculating the ratio of the current transient power consumption of the physical computing core to the preset power threshold, the load allocation adjustment coefficient is obtained. The calculation logic for this coefficient is as follows: ,in, For load distribution adjustment factor, This represents the current kernel's transient power consumption. The preset power threshold is set to 1.2. This threshold is based on the transient power redundancy measurement of a typical embedded processor under thermal steady-state conditions. Experiments show that when the transient power consumption exceeds 20% of the preset power threshold, localized heat accumulation inside the chip cannot be dissipated through thermal grease and heat sinks in the next scheduling cycle. If this limit is continuously exceeded, hardware-level frequency limiting and oscillation shutdown will be triggered. Therefore, [the following text is incomplete and requires further context]. As an activation condition for triggering the discrete reconstruction of the ready linked list topology, it can maximize the use of the hardware's short-term pulse computing power without sacrificing peak performance.

[0038] During the logical judgment process, if the load distribution adjustment coefficient If the value exceeds 1.2, the scheduler determines that the current physical computing kernel has a high risk of overheating and frequency limiting. It immediately intercepts the pointers of newly accessed task control blocks and, by searching the ready queue lengths of the remaining physical computing kernels, redirects the new task to the kernel with the shortest ready queue length. Simultaneously, task control blocks already queued on the current kernel are migrated to a less loaded kernel. This discrete reconstruction of the task ready list topology mitigates the risk of overheating and frequency limiting. The task scheduler executes specific queue protection procedures for high-computation-density image stream tasks. Before mounting a task, the scheduler extracts the task weight factor of the current image data stream. When the weight factor exceeds the scheduling threshold When an image task is assigned to a physical computing kernel with the highest nominal clock frequency and a transient temperature below the safe temperature threshold, the assignment logic eliminates processing delays caused by mixed parsing of heterogeneous tasks, enabling dynamic coupling between high-load data tasks and the heat dissipation capabilities of the physical computing kernel. This invention constructs a task type discrimination arbitrator at the application layer. When the multi-protocol adaptation middleware identifies the communication protocol type as an image encoding protocol, the arbitrator issues a forced interrupt signal, suspending the general weight calculation thread. At this time, the system executes image stream-specific dispatch logic, directly retrieving the node with the highest clock frequency and shortest queue from the physical computing kernel register. Only when no kernel meets the temperature safety conditions will the arbitrator release control and fall back to the kernel based on... The general weight balancing mode, through this logical priority hard isolation, ensures the real-time response of high-value, high-load tasks. The above control logic, while ensuring the real-time processing of incoming tasks, avoids the blocking of the ready queue caused by the surge in transient power consumption inside the processor, and reduces the instantaneous power consumption fluctuation of the system.

[0039] Example 5: Before deployment, the current multi-source heterogeneous monitoring terminal access system needs to perform a standardized pre-calibration process for electromagnetic interference noise floor and computing load fluctuations in specific environments. During the offline configuration phase, the task scheduler uses a standard clock signal generator and a benchmark simulated load source to cyclically inject simulated task sequences of different cycles into the physical computing kernel, and measures the idle instruction execution efficiency of each kernel under interference-free conditions. The system is connected to a controllable electromagnetic interference source, and the power frequency interference intensity is set to 50Hz. The interference amplitude is gradually adjusted, and the fluctuation amplitude of the computing task completion cycle of each kernel under different disturbance intensities is recorded. Based on the above experimental data, the task scheduler generates an environmental adaptability calibration curve, which is used as the initial benchmark value for the correction of computing task density during system operation, ensuring that the computing power demand feedforward measurement mechanism has consistent initial judgment accuracy when deployed in different field environments.

[0040] During on-site deployment, the system automatically initiates a baseline deviation calibration process to eliminate individualized computing power deviations caused by differences in processor core manufacturing processes. The system utilizes a high-precision timer within the core to measure the actual time taken to execute the standard decompression algorithm and maps this time to the theoretically optimal time specified at the factory, thus constructing a core computing power correction coefficient. This coefficient is defined as the ratio of the theoretically optimal time to the actual test time. It is used to offset the time slice allocation weight deviation caused by individual processor performance differences during dynamic scheduling. In the specific calculation logic, the correction coefficient... It acts on the initial weights in the form of a product factor. In other words, the final decision on link list mounting is based on the corrected time slice allocation weight. Through this coupling path, the scheduler can incorporate the closed-loop hardware individual differences obtained from offline calibration into real-time load balancing decisions, ensuring that tasks with the same load receive higher allocation weights on physical cores with better performance. Specifically, before each physical computing core is formally connected to the multi-source heterogeneous data stream, it undergoes adaptive initial calibration by calculating the deviation rate to ensure that the allocation weights calculated subsequently are accurate. Maintaining logical determinism in heterogeneous task processing paths, such parameter calibration procedures are automatically re-executed after each system power outage and restart or hardware configuration change, thereby ensuring that the task scheduling system has long-term operational stability and engineering consistency in analyzing the computing power requirements of heterogeneous access streams.

[0041] Example 6: Before the formal deployment of the heterogeneous node access system of the current monitoring terminal, the system calibrates the instruction cycle benchmark table based on the node communication protocol parsing computing power overhead. This table is stored in the central processing unit core register array. When the monitoring terminal accesses, the task scheduler extracts the data flow communication protocol type identifier and data packet byte length, and matches the single-byte parsing instruction cycle constant with the computing power consumption matrix. Calculate the expected computational density Based on the current ready queue length of each physical computing kernel Nominal main frequency and operating voltage Calculate the weights for time slices The system is configured with a thermal steady-state feedback factor. This factor reflects the transient temperature of the physical calculation kernel. With safe temperature threshold The degree of deviation is calculated using the following logic: ,in, For thermal steady-state feedback factor, For safe temperature threshold, For the first Transient temperature of a physical computing kernel. This is an empirical coefficient with a value of 0.5. (Regarding the empirical coefficient...) The value of is determined by polynomial fitting of the chip thermal runaway curve at 85℃ in this invention. When the value is 0.5, the thermal steady-state feedback factor exist near It exhibits a moderate decay slope, ensuring timely triggering of scheduling intervention before the temperature crosses the safety threshold, while avoiding system task jitter caused by excessively steep decay. If it is too small, the feedback inhibition will be too slow and unable to block the formation of hot spots; if... If the value is too high, the kernel will be frequently downgraded within the normal fluctuation range, reducing the overall system throughput.

[0042] The task scheduler calculates the corrected time slice allocation weights based on the above factors. The formula is Under this processing logic, the transient power consumption of the physical computing kernel With the frequency of processing task load and task voltage reference value Fluctuations, when Exceeding the predetermined power threshold When this occurs, the ready queue is reconfigured, and the system compares the difference between the transient power consumption of each physical computing core and a preset threshold. Identify nodes at risk of overflow and shift the ready list pointer of that node to the hashrate fluctuation coefficient. The minimum idle computing cores, and to ensure the determinism of this scheduling strategy under complex operating conditions, the computing power fluctuation coefficient of the physical computing cores. The calculation is based on the following: ,in, For the first The computing power fluctuation coefficient of a physical computing kernel To determine the total number of samples for the task package, For the first The actual parsing time for each task package To analyze the average time consumption when When the power margin exceeds the migration threshold, the task scheduler will attach the computing thread task at the end of the queue to the physical computing core with the smallest computing power fluctuation coefficient, and complete the dynamic update of the ready list topology. This configuration path maintains the task parsing order while reducing the risk of local processor overheating through the spatial mapping of ready list nodes, and avoids logical deadlock caused by kernel computing core ready queue overflow when high concurrency access occurs.

[0043] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for heterogeneous node access in monitoring terminals, characterized in that, Includes the following steps: Step S1: The multi-protocol adaptation middleware loads the protocol library to build a parsing syntax tree, extracts the communication protocol type identifier of the multimodal data stream, and encapsulates the multimodal data stream into a normalized computation thread task containing the communication protocol type identifier. Step S2: The task scheduler reads the communication protocol type identifier and the current data packet byte length of the normalized computing thread task, matches the single-byte parsing instruction cycle constant through the computing power consumption matrix, and multiplies the constant by the byte length to obtain the expected computing intensity. Step S3: The task scheduler obtains the current ready queue length, nominal main frequency, current operating voltage, and transient temperature of each physical computing core through the ready queue register, nominal main frequency register, operating voltage register, and kernel transient temperature register. When the transient temperature is not greater than the safe temperature threshold, the nominal main frequency is multiplied by the current operating voltage to obtain the working capacity product, and the product is divided by the sum of the current ready queue length and the number 1 to obtain the time slice allocation weight. When the transient temperature is greater than the safe temperature threshold, the safe temperature threshold is divided by the transient temperature and then multiplied by the working capacity product to obtain the controlled capacity product, and the controlled capacity product is divided by the sum of the current ready queue length and the number 1 to obtain the reduced time slice allocation weight. Based on the expected computational intensity and time slice allocation weight, the control block pointer of the normalized computation thread task is rewritten, and the normalized computation thread task is attached to the tail of the physical computation kernel ready list that meets the state determination conditions.

2. The heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, The process of data encapsulation and translation by the multi-protocol adaptation middleware in step S1 includes: step S11, loading the containerized protocol parsing plugin corresponding to the communication protocol type identifier, and constructing a complete parsing syntax tree for the multimodal data stream; step S12, using the complete parsing syntax tree to eliminate the instruction cycle differences between heterogeneous protocols, converting the payload in the current data packet into a normalized time-series text data stream with a unified semantic structure, and encapsulating the normalized time-series text data stream into an independent normalized computation thread task.

3. The heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, The process of calculating the time slice allocation weight in step S3 includes: step S31, the task scheduler reads the number of currently queued task control blocks in the ready queue register inside each physical computing core in each control cycle, and uses it as the current ready queue length; step S32, the transient temperature is compared with the safe temperature threshold, and when the transient temperature is greater than the safe temperature threshold, the temperature rise reverse feedback suppression is enabled to forcibly reduce the scheduling priority of physical computing cores with high heat frequency limiting risk.

4. The heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, The process of rewriting the control block pointer of the normalized computing thread task in step S3 includes: step S33, when the current ready queue length is greater than the activation load threshold and the transient temperature exceeds the safe temperature threshold, the task scheduler intercepts the thread assignment signal for the target physical computing kernel; step S34, rewriting the control block pointer of the normalized computing thread task, and attaching the newly accessed multimodal data stream to the tail of the ready list of the physical computing kernel that meets the state judgment conditions, bypassing the physical computing kernel with high thermal frequency limiting risk.

5. A heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, After the multi-protocol adaptation middleware translates the multimodal data stream into a normalized computation thread task, the method further includes: step S4, adjusting the kernel ready list allocation priority for image data streams with higher computational overhead than time-series text data streams, and prioritizing the allocation of image data streams to the physical computation kernel with the highest nominal main frequency and the shortest current ready queue length.

6. The heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, When calculating the time slice allocation weight in step S3, the method further includes: step S35, introducing a dynamic decay coefficient based on the kernel transient temperature change rate into the time slice allocation weight, wherein the dynamic decay coefficient changes monotonically with the increase of the kernel transient temperature change rate.

7. A heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, After step S3, the method also includes a hardware health status assessment process: Step S37, periodically record the historical transient temperature sequence and historical power consumption yield frequency of each physical computing core in a continuous scheduling cycle, and calculate the physical core health degradation index characterizing the core heat accumulation rate; Step S38, when the physical core health degradation index continuously exceeds the safe aging threshold, output a core warning signal, and at the same time limit the maximum allowable ready queue length of the physical computing core.

8. A heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, The method for determining whether the physical computing kernel meets the state determination conditions in step S3 includes: step S36, calculating the temperature difference between the transient temperature of the current physical computing kernel and the safe temperature threshold, and determining that the physical computing kernel meets the state determination conditions when the temperature difference is greater than the temperature drop margin threshold and the current ready queue length is less than the ready queue baseline length.

9. A heterogeneous node access method for a monitoring terminal according to claim 1, characterized in that, The ready queue register, nominal main frequency register, operating voltage register, and core transient temperature register are all connected to the task scheduler through the chip's internal micro-rail hardware bus. Under the control of the scheduling clock signal with a clock frequency of 100MHz to 1000MHz, the hardware parameters of each physical computing core are read.

Citation Information

Patent Citations

  • CFS scheduler-based ready queue average load optimization method and data structure

    CN111274024A