A resource scheduling method and related equipment for a hybrid computing cluster

Abstract

This invention relates to a resource scheduling method and related equipment for a hybrid computing cluster. The method includes: acquiring computing tasks and extracting task characteristic parameters; estimating the quantum computing speedup ratio based on the task characteristic parameters; if the quantum computing speedup ratio is greater than a threshold, allocating the computing task to a quantum computing unit for execution; acquiring the physical state parameters of each quantum computing unit to calculate a quantum resource health score; dividing the computing task based on coherence time constraints according to the quantum resource health score and a decoherence time window, generating multiple quantum subtasks; before the execution of the quantum subtasks, if it is determined that the estimated execution time of any quantum subtask exceeds the corresponding decoherence time window, or the quantum resource health score is lower than the score threshold, triggering an automatic rollback mechanism to schedule the corresponding quantum subtask to a classical computing unit for computation. This solves the problems of poor resource scheduling and low task processing efficiency in related technologies.

Description

Technical Field

[0001] This invention relates to the field of quantum computing technology, and in particular to a resource scheduling method and related equipment for a hybrid computing cluster. Background Technology

[0002] Quantum computing has developed rapidly in recent years, and quantum computing units (also known as quantum processing units, or QPUs) have begun to work in conjunction with classical computing units (CPUs, GPUs) to form a quantum computing system. Classical hybrid computing clusters. Hybrid computing clusters have shown potential to surpass pure classical computing in fields such as quantum chemical simulations, combinatorial optimization, and machine learning. However, how to efficiently schedule heterogeneous resources within hybrid computing clusters, especially how to rationally utilize expensive and fragile quantum computing power, has become a key bottleneck restricting the performance of hybrid computing clusters.

[0003] However, in practical applications, the behavior of quantum computing units differs fundamentally from that of classical computing units. If the scheduler ignores these quantum-specific properties, tasks may be assigned to qubits that are about to expire, resulting in a waste of computing resources; or a large quantum circuit may be submitted to a QPU with insufficient coherence time, causing the quantum state to decoherent before the computation is completed, and the final result to be meaningless.

[0004] There is currently no effective solution to the problems of poor resource scheduling and low task processing efficiency in related technologies. Summary of the Invention

[0005] The present invention provides a resource scheduling method and related equipment for a hybrid computing cluster, which at least partially solves the problems of poor resource scheduling effect and low task processing efficiency in related technologies.

[0006] To address the aforementioned problems, one aspect of this invention provides a resource scheduling method for a hybrid computing cluster, the hybrid computing cluster including quantum computing units and classical computing units, the method comprising: The system acquires a computing task to be processed and extracts the task feature parameters. Based on the task feature parameters, it estimates the quantum computing speedup ratio. If the quantum computing speedup ratio is greater than a preset speedup ratio threshold, the system assigns the computing task to the quantum computing unit for execution. The physical state parameters of each quantum computing unit in the hybrid computing cluster are acquired in real time, and a quantum resource health score is calculated based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the temporal degradation characteristics of quantum states; and the quantum resource health score is used to dynamically characterize the availability state of quantum resources. Based on the quantum resource health score and the decoherence time window determined by the remaining decoherence time, the computing tasks assigned to the quantum computing unit are divided into multiple quantum subtasks based on coherence time constraints. Before the execution of the quantum task, if it is determined that the expected execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than the preset health score threshold, an automatic rollback mechanism is triggered to schedule the corresponding quantum task to the classical computing unit for computation.

[0007] In some embodiments, the task segmentation step based on coherent time constraints includes: The maximum allowed circuit depth within the decoherence time window is calculated based on the decoherence time window, the gate operation time of the quantum computing unit performing a single quantum gate operation, and the fidelity information contained in the quantum resource health score. The computational task assigned to the quantum computing unit for execution is divided into multiple quantum subtasks, such that the circuit depth corresponding to each quantum subtask is less than or equal to the maximum circuit depth.

[0008] In some embodiments, prior to the task segmentation step based on coherent time constraints, the method further includes: Based on the partitioning strategy corresponding to the computing task, the estimated number of quantum tasks generated after task partitioning is estimated, and based on the estimated number, the communication overhead required to merge and reconstruct the computing results of each quantum task in the classical computing unit is calculated. If the communication overhead exceeds a preset overhead threshold, then the computation task is prohibited from being split, and the computation task is allocated as a whole to the classic computing unit for execution.

[0009] In some embodiments, the step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time and calculating the quantum resource health score based on the physical state parameters includes: Multiple physical state parameters characterizing the operating state of the quantum computing unit are acquired in real time; wherein, the physical state parameters include the remaining decoherence time, gate fidelity, computational error rate, and number of available qubits; Based on the physical state parameters, determine the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the computation error rate to the preset error rate threshold, and the proportion of available qubits for the quantum computing unit. The quantum resource health score is calculated based on a weighted regression model; wherein the input variables of the weighted regression model include: the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the calculation error rate to the preset error rate threshold, and the proportion of available qubits.

[0010] In some embodiments, the step of estimating the quantum computing speedup based on the task characteristic parameters includes: The problem size corresponding to the computing task, the connectivity characterizing the degree of interaction between qubits in the task, and the proportion of the computing task that can be quantized are obtained from the task characteristic parameters, as the quantizable proportion. The quantum computing speedup is obtained by calculating the problem size, the connectivity, and the quantizability ratio using a speedup prediction function.

[0011] In some embodiments, the step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time further includes: Based on time series analysis of historical operating data, interpolation is used to predict the remaining decoherence time of each quantum computing unit at the current moment.

[0012] In some of these embodiments, it also includes: The communication adaptation unit adaptively switches between blocking and non-blocking call modes based on the task type. Specifically, tasks assigned to the quantum computing unit are executed in a blocking mode, while tasks scheduled to the classical computing unit after triggering the automatic rollback mechanism are executed in a non-blocking mode.

[0013] In some embodiments, the preset acceleration ratio threshold ranges from [1.5, 5], and the preset health score threshold ranges from [0.1, 0.5].

[0014] To address the aforementioned problems, one aspect of this invention provides a resource scheduling device for a hybrid computing cluster, the hybrid computing cluster including quantum computing units and classical computing units, the device comprising: The first scheduling module is used to acquire computing tasks to be processed, extract task feature parameters of the computing tasks, estimate the quantum computing speedup based on the task feature parameters, and if the quantum computing speedup is greater than a preset speedup threshold, then the computing task is assigned to the quantum computing unit for execution. The computing module is used to acquire the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time, and calculate the quantum resource health score based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of quantum states; the quantum resource health score is used to dynamically characterize the availability state of quantum resources; The task segmentation module is used to segment the computing tasks assigned to the quantum computing unit based on coherence time constraints according to the quantum resource health score and the decoherence time window determined by the remaining decoherence time, thereby generating multiple quantum subtasks. The second scheduling module is used to, before the execution of the quantum task, if it is determined that the estimated execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than a preset health score threshold, trigger an automatic rollback mechanism to schedule the corresponding quantum task to the classical computing unit for computation.

[0015] To address the aforementioned problems, one aspect of this invention provides a non-transitory machine-readable medium storing computer instructions for causing a computer to execute any of the above-described hybrid computing cluster resource scheduling methods.

[0016] The beneficial effects of this invention are as follows: By acquiring the computational task to be processed, the task characteristic parameters of the computational task are extracted; the quantum computing speedup ratio is estimated based on the task characteristic parameters; if the quantum computing speedup ratio is greater than a preset speedup ratio threshold, the computational task is allocated to a quantum computing unit for execution; the physical state parameters of each quantum computing unit in the hybrid computing cluster are acquired in real time, and a quantum resource health score is calculated based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of quantum states; the quantum resource health score is used to dynamically characterize the available state of quantum resources; based on the quantum resource health score and the decoherence time window determined by the remaining decoherence time, the computational task allocated to the quantum computing unit for execution is divided into multiple quantum subtasks based on coherence time constraints; before the execution of any quantum subtask, if it is determined that any quantum subtask... If the expected execution time of a task exceeds the corresponding decoherence time window, or if the quantum resource health score falls below a preset health score threshold, an automatic rollback mechanism is triggered, scheduling the corresponding quantum task to a classical computing unit for computation. This technique overcomes the problems of poor resource scheduling and low task processing efficiency in related technologies. By first determining whether the computational task to be processed is suitable for quantum computing, then assessing the health of the quantum computing unit, then dividing the computational task based on coherence time constraints, and finally verifying the division result before task execution, and using classical computing units for simulation computation as a fallback path, this technique mitigates the risk of continued deterioration of resource status after task division. It achieves a significant improvement in computing power utilization and an effective reduction in scheduling latency, thus improving resource scheduling efficiency, reducing computational resource waste, and increasing task processing efficiency.

[0017] Details of one or more embodiments of the present invention are set forth in the following drawings and description, so that other features, objects and advantages of the invention will be more readily understood. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the main flow of a resource scheduling method for a hybrid computing cluster according to one embodiment of the present invention; Figure 2 This is a schematic diagram of the main flow of a resource scheduling method for a hybrid computing cluster, which is another embodiment of the present invention. Figure 3This is a schematic diagram of the main modules of a resource scheduling device for a hybrid computing cluster according to one embodiment of the present invention; Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0020] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the invention. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the invention.

[0021] In the realm of classic high-performance computing, scheduling systems (such as Kubernetes and SLURM) typically employ a resource-based "binary state" model—that is, determining whether a computing unit is idle, online, or under high or low load, and allocating tasks accordingly. This type of scheduler is unaware of the real-time degradation of the physical attributes of computing units, which is reasonable for CPUs or GPUs because their computing power remains relatively stable over time.

[0022] However, the behavior of quantum computing units differs fundamentally from that of classical computing units: the coherence time (decoherence time) of a qubit is continuously consumed during idle or operation, gate fidelity decays with the number of operations and crosstalk, and the error rate may gradually increase until it exceeds the fault tolerance threshold. These physical parameters change dynamically over time, and the changes are irreversible. If the scheduler ignores these quantum-specific properties, tasks may be assigned to qubits that are about to expire, resulting in wasted computing resources; or large quantum circuits (also known as quantum tasks) may be submitted to a QPU with insufficient coherence time, causing the quantum state to decoher before the computation is completed, resulting in a meaningless final result.

[0023] Furthermore, scheduling schemes in related technologies often lack the ability to predict whether a task is suitable for quantum acceleration. Not all computational tasks can benefit from quantum computing. For tasks with low quantum speedup ratios, forcibly allocating them to the QPU not only fails to improve performance but also reduces overall efficiency due to task encoding / decoding, queuing, and communication overhead. On the other hand, even if a task is determined to be suitable for quantum execution, how to decompose the task into multiple subtasks that can be completed within a finite decoherence time window, and how to automatically fall back to classical simulation when subtasks may time out or quantum resources deteriorate, are unresolved technical blind spots in the scheduling systems provided by related technologies.

[0024] To address the aforementioned problems, embodiments of the present invention provide a resource scheduling method for hybrid computing clusters, such as... Figure 1 As shown, the resource scheduling method of this hybrid computing cluster mainly includes: Step S101: Obtain the computing task to be processed and extract the task feature parameters of the computing task; estimate the quantum computing speedup based on the task feature parameters; if the quantum computing speedup is greater than the preset speedup threshold, then allocate the computing task to the quantum computing unit for execution. Step S102: Real-time acquisition of physical state parameters of each quantum computing unit in the hybrid computing cluster, and calculation of quantum resource health score based on physical state parameters; wherein, physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of quantum states; quantum resource health score is used to dynamically characterize the available state of quantum resources; Step S103: Based on the quantum resource health score and the decoherence time window determined by the remaining decoherence time, the computing tasks assigned to the quantum computing unit are divided into multiple quantum subtasks based on coherence time constraints. Step S104: Before the execution of any quantum task, if it is determined that the expected execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than the preset health score threshold, an automatic rollback mechanism is triggered to schedule the corresponding quantum task to the classical computing unit for computation.

[0025] Based on the above setup, by first determining whether the computational task to be processed is suitable for quantum computing, then assessing the health of the quantum computing unit, then dividing the computational task based on coherent time constraints, and finally verifying the division result before task execution, and using classical computing units for simulation as a fallback path, the risk of resource status continuing to deteriorate after task division is mitigated. This achieves a significant improvement in computing power utilization and an effective reduction in scheduling latency, thus improving resource scheduling efficiency, reducing computing resource waste, and increasing task processing efficiency.

[0026] According to an embodiment of the present invention, based on the above step S101, two key elements are introduced: task characteristic parameters and estimated quantum computing speedup. The task characteristic parameters (such as problem size, connectivity, and quantizability ratio) reflect the intrinsic structure of the computational task, and these parameters are highly correlated with the acceleration potential of the quantum algorithm. By estimating the quantum computing speedup, the expected performance gain of the computational task on a quantum computing unit compared to a classical computing unit can be quantified without actually executing the task. Comparing the estimated quantum computing speedup with a preset speedup threshold allows the scheduling device / system to make reasonable scheduling decisions—only computational tasks whose expected benefits exceed their costs (the preset speedup threshold can take into account communication and queuing overhead) are sent to the quantum computing unit. Based on this mechanism, the traditional binary mode of scheduling devices / systems, which either process entirely through quantum computing units or entirely through classical computing units, is changed, ensuring a fine-grained match between computational tasks and computing units. This enables the pre-determination of whether a current computing task is suitable to be executed by a quantum computing unit during the scheduling decision-making stage, avoiding blindly scheduling tasks with low quantum speedup ratios to quantum processing units, thereby reducing unnecessary quantum resource consumption and task encoding / decoding and communication overhead.

[0027] In some embodiments, the availability of quantum computing units differs fundamentally from that of classical computing units—the coherence time of qubits is continuously consumed during idle and operational periods, gate fidelity decreases with the number of operations, and the computational error rate may gradually approach the fault tolerance threshold. Furthermore, these parameters cannot be characterized by simple "online status" or "load level." Based on step S102, by real-time acquisition of physical state parameters corresponding to the quantum computing unit, including the remaining decoherence time, the remaining decoherence time serves as the core input variable, directly reflecting the time budget by which the quantum state can maintain effective computation. Simultaneously, by quantizing the physical properties of the quantum computing unit that dynamically degrade over time (especially the decoherence time) into a continuous health score, the scheduling device / system can grasp the true availability of quantum resources in real time, rather than relying solely on static idle / busy states. Specifically, the lower the quantum resource health score, the closer the quantum resource is to failure. Based on this, scheduling can proactively avoid risks by preventing the allocation of critical tasks to quantum computing units that are about to degrade.

[0028] In some examples, the correctness of quantum computing heavily depends on the coherence of quantum states during computation. If the total execution time of a quantum circuit (also known as a quantum computing task) exceeds the decoherence time of a qubit, the measurement result will become meaningless due to decoherence. Based on step S103 above, a time window (i.e., the decoherence time window) is determined using the remaining decoherence time obtained in the previous steps. Combined with the fidelity information contained in the health score (lower fidelity allows for smaller circuit depth), the original computation task is divided into multiple quantum subtasks based on coherence time constraints. This task partitioning strategy based on coherence time constraints essentially encodes the quantum physical law (decoherence) directly into the task partitioning rules of the scheduling device, thereby eliminating the possibility of invalid computation due to timeouts at the source. Understandably, based on this step, large computational tasks that might otherwise exceed the quantum state lifetime can be forcibly divided into multiple small quantum subtasks that can be completed within a single decoherence time window, ensuring that each subtask completes from the start of execution to the result readout before the quantum state decoherence, thus guaranteeing the physical validity of the computation result.

[0029] In some cases, the task partitioning in the aforementioned steps is based on the current health score and decoherence time window, but the state of quantum resources is constantly changing. There may be a small time difference between the completion of task partitioning and the actual start of subtask execution. During this period, the remaining decoherence time of the quantum computing unit may be further shortened, or the health score may drop below the threshold due to crosstalk, measurement interference, etc. Based on the above step S104, before execution, it is re-determined (rather than detected at runtime) whether the expected execution time of each subtask still meets the current decoherence time window and whether the health score is still within the available range. This "predictive rollback" mechanism does not wait for execution failure to occur, but actively avoids risks before execution. Once rollback is triggered, the subtask is scheduled to the classical computing unit for simulation computation—although the accuracy or speed of the simulation computation may be lower than that of successful quantum computation, it at least ensures the completion of the task and avoids the interruption of the entire computing task due to the instantaneous failure of quantum resources. Therefore, by performing a dual check on the quantum subtask and the state of quantum resources before execution, if the time constraint cannot be met or the resource health is too low, the subtask is immediately transferred to the classical computing unit for simulation calculation, ensuring that the task can obtain the final result under any circumstances, avoiding task suspension or failure due to quantum resource failure, and significantly improving the resource scheduling effect.

[0030] In some of these embodiments, the task segmentation step based on coherence time constraints includes: calculating the maximum allowed circuit depth within the decoherence time window based on the decoherence time window, the gate operation time of a single quantum gate operation performed by the quantum computing unit, and the fidelity information contained in the quantum resource health score; and segmenting the computational task assigned to the quantum computing unit for execution into multiple quantum subtasks, such that the circuit depth corresponding to each quantum subtask is less than or equal to the maximum circuit depth.

[0031] Based on the above setup, decoherence constraints and fidelity constraints in quantum physics are transformed into task partitioning rules executable by the scheduling device. This allows any task assigned to a quantum computing unit to be automatically decomposed into physically complete quantum subtasks, thus ensuring the success rate of quantum computing and avoiding the waste of computing resources caused by the mismatch between circuit depth and coherence time. Specifically, the two sub-steps constitute a complete "constraint quantization-task decomposition" chain. The first sub-step is responsible for extracting a static, deterministic upper limit of circuit depth (that is, transforming physical laws into engineering constraints) from dynamic physical parameters (decoherence time window, gate operation time, fidelity). The second sub-step uses this upper limit of circuit depth to partition the original task, concretizing the abstract circuit depth constraint into a hard requirement for each quantum subtask. Based on these two sub-steps, the use of quantum computing units changes from "attempting to run, possibly failing" to "pre-ensuring execution," significantly improving the determinism and efficiency of quantum resource scheduling.

[0032] According to an embodiment of the present invention, the total execution time of a quantum circuit (also called a quantum task) is approximately equal to the circuit depth multiplied by the average time of a single quantum gate operation. The decoherence time window represents the maximum duration for which a quantum state can remain coherent. To ensure the physical validity of the quantum computing results, the execution time of the quantum circuit must be ≤ the decoherence time window. The decoherence time window is determined by the remaining decoherence time, the gate operation time is an inherent parameter of the quantum computing unit, and the fidelity information comes from the quantum resource health score. Therefore, the calculated maximum circuit depth reflects not only the time constraint but also the precision limitation of quantum operations. That is, through the above settings, the dynamic physical constraints of the quantum computing unit (decoherence time window, gate operation speed, fidelity) are quantified into a specific numerical index (i.e., maximum circuit depth), enabling the scheduling device to clearly know how many layers of quantum gate operations a quantum task can contain without decoherence occurring before completion, given the current quantum resource state.

[0033] According to another embodiment of the present invention, many quantum subtasks (such as quantum approximation optimization algorithms, variational quantum eigenvalue solvers, etc.) often have large circuit depths. If directly submitted to the quantum computing unit for execution, they can easily exceed the decoherence time window. Based on the above settings, the original task is decomposed into several subtasks using the calculated maximum circuit depth as the hard partitioning granularity. The circuit depth of each subtask is forcibly limited to within the maximum depth. This partitioning strategy does not change the logical correctness of the original task (subtasks can be reorganized or parameter passed between them through classical computation), but it changes the execution time of each subtask on the quantum computing unit. Since the depth of each subtask does not exceed the maximum depth, its execution time will inevitably not exceed the decoherence time window, avoiding mid-computation decoherence failure due to excessive circuit depth, thereby fundamentally eliminating the risk of invalid computation due to timeout.

[0034] In some embodiments, before the task segmentation step based on coherent time constraints, the method further includes: estimating the estimated number of quantum tasks generated after task segmentation based on the segmentation strategy corresponding to the computing task; and calculating the communication overhead required to merge and reconstruct the computing results of each quantum task in the classical computing unit based on the estimated number; if the communication overhead is greater than a preset overhead threshold, then the task segmentation is prohibited, and the entire computing task is allocated to the classical computing unit for execution.

[0035] Based on the above settings, a cost verification step is added before task splitting. This ensures that tasks are only split and assigned to quantum computing units for execution if the estimated classical recombination communication overhead after task splitting is within an acceptable range. Otherwise, the entire computation task is directly processed by the classical computing unit. This fundamentally prevents the problem of quantum acceleration gains being swallowed up by communication overhead due to overly fine task splitting, thus improving the overall computational efficiency of the hybrid computing cluster. Specifically, the number of subtasks is first estimated to provide input for subsequent overhead calculations; then, the abstract task splitting cost is transformed into a specific value by calculating the communication overhead; finally, the execution results are compared to make the final scheduling decision. If the communication overhead exceeds a preset overhead threshold, task splitting is prohibited. In classical hybrid computing, quantum computing units execute sub-tasks, while classical computing units reassemble the results. The efficiency of their collaboration directly impacts the end-to-end latency of the entire task. Based on these steps, the traditional quantum scheduling scheme addresses the blind spot of focusing solely on whether quantum computation can be completed, while neglecting the cost of data integration afterward. Understandably, without these steps, the scheduling device might divide a computational task into hundreds of miniature quantum circuits. While each sub-circuit could complete within decoherence time, the classical reassembly stage would require hundreds of result transfers and complex merging, potentially resulting in a total time consumption far exceeding that of direct classical computation.

[0036] In some embodiments, the partitioning strategy for different computational tasks (such as partitioning by circuit depth, partitioning by the number of qubits, partitioning by subgraphs, etc.) determines the number of subtasks after partitioning. Based on the partitioning strategy corresponding to the computational task, the step of estimating the estimated number of quantum subtasks generated after task partitioning can pre-calculate the theoretical or approximate value of the number of subtasks by analyzing the scale of the original task (e.g., total circuit depth, number of qubits, total number of gate operations) and the preset partitioning granularity (e.g., maximum circuit depth). This estimation does not depend on the completion of the actual partitioning operation, and therefore can be performed in the early stages of scheduling decisions, avoiding the ineffective computation and resource waste caused by "performing task partitioning first and then discovering excessive overhead."

[0037] In some examples, a large computational task is divided into multiple quantum subtasks. Each subtask, after being executed on a quantum computing unit, produces partial results. To obtain the final complete result, the classical computing unit must collect the results of all subtasks and merge and reconstruct them (e.g., shrinking tensor networks, splicing sub-circuit results, and rearranging parameters). This process involves data transfer (communication overhead) of the subtask computation results from the quantum computing unit to the classical computing unit, as well as post-processing computation (computation overhead) on the classical computing unit. By calculating the steps required to merge and reconstruct the computation results of each quantum subtask in the classical computing unit based on an estimated number, and using the estimated number of subtasks, combined with the data volume of each subtask result, the communication bandwidth and processing power of the classical computing unit, the total classical reconstructing communication overhead can be estimated. This quantifies the side effects of task division (the data transfer and computation costs incurred by merging and reconstructing the quantum subtask computation results in the classical computing unit) into a comparable communication overhead value, providing a quantitative basis for subsequent scheduling decisions.

[0038] In some cases, while dividing the task into subtasks ensures that each subtask completes within the decoherence time window, excessively fine divisions can generate a large number of subtasks, leading to a surge in communication overhead during the classical reassembly phase. If this communication overhead exceeds the theoretical speedup gain of quantum computing relative to classical computing, then using quantum computing units may actually be slower or less efficient than using classical computing units directly. By setting a preset overhead threshold (which can be set based on historical system data or experience, for example, related to a certain proportion of the estimated quantum speedup), when the estimated communication overhead exceeds this threshold, the scheduler can proactively prohibit task division and roll back the entire task to classical computing units for execution. This effectively avoids ineffective quantum computing acceleration and ensures the overall computational efficiency of the hybrid computing cluster.

[0039] According to a specific embodiment of the present invention, before the above-described task partitioning step based on coherent time constraints, the method further includes: if the communication overhead is greater than the benefit brought by the speedup of quantum computing, then partitioning the computing task is prohibited, and the computing task is allocated as a whole to classical computing units for execution.

[0040] Based on the above settings, only tasks whose communication overhead after partitioning does not exceed the quantum acceleration gain are allowed to enter the quantum partitioning and execution process. This avoids the additional communication cost caused by partitioning from completely offsetting or even exceeding the performance gain brought by quantum computing itself, thus ensuring that the end-to-end acceleration effect of the hybrid computing cluster is positive.

[0041] The speedup benefits of quantum computing essentially lie in the time savings or increased efficiency achieved by having quantum computing units execute computational tasks compared to classical computing units. However, to accommodate the decoherence time constraints of quantum computing units, computational tasks typically need to be divided into multiple quantum subtasks. This division itself doesn't generate additional quantum speedup benefits; instead, it introduces new costs on the classical computing unit side: the communication overhead required to merge and reconstruct the computation results of each subtask. When the division granularity is too fine or the number of subtasks is too large, this communication overhead can become very significant, even exceeding the original benefits of quantum speedup. Based on the above steps, a simple and effective decision boundary is constructed by directly comparing the communication overhead with the quantum speedup benefits: if the communication overhead exceeds the benefits, it means that the overall time consumption using quantum computing units (after division) is actually higher than using classical computing units directly. In this case, the scheduler actively prohibits task division and allocates the entire task to classical computing units for execution, effectively choosing a "suboptimal but definitely effective" solution, avoiding the trap of "quantum speedup being dragged down by cost."

[0042] In some embodiments, the step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time and calculating the quantum resource health score based on the physical state parameters includes: acquiring multiple physical state parameters characterizing the operating state of the quantum computing unit in real time; wherein the physical state parameters include remaining decoherence time, gate fidelity, computational error rate, and number of available qubits; determining the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the computational error rate to a preset error rate threshold, and the proportion of available qubits based on the physical state parameters; and calculating the quantum resource health score based on a weighted regression model; wherein the input variables of the weighted regression model include: the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the computational error rate to a preset error rate threshold, and the proportion of available qubits.

[0043] Based on the above setup, a specific scheme for calculating the health score of quantum resources is provided. Through a complete chain of raw physical parameter acquisition, normalization processing, and weighted fusion scoring, the complex, time-varying, and multi-dimensional physical states of quantum computing units are compressed into a simple, stable, and comparable health score. This health score retains the physical meaning of the original parameters (through normalization) and achieves adaptation to different scenarios through learnable weights, thus providing a clear and reliable quantitative basis for upper-level scheduling decisions. This transforms the quantum computing unit from a black box into a quantifiable resource. The scheduling device can directly determine the availability of the current quantum resource based on the health score, greatly reducing the implementation complexity of resource scheduling in hybrid computing clusters.

[0044] Specifically, according to embodiments of the present invention, the physical state parameters of the quantum computing unit are highly dynamic. The remaining decoherence time is continuously consumed as the qubits are idle or operated, meaning the remaining decoherence time determines how long the quantum state can be maintained. Gate fidelity decreases with the number of operations and the accumulation of crosstalk, meaning the gate fidelity determines the accuracy of a single operation. The computational error rate may increase due to environmental noise or measurements of neighboring qubits, meaning the computational error rate determines the reliability of the computational result. The number of available qubits also changes in real time as some qubits are calibrated, reset, or marked as unavailable, meaning the number of available qubits determines the parallel processing capability. By performing a "real-time acquisition" operation (e.g., polling the state register at millisecond intervals or triggering through hardware interrupts), the instantaneous values ​​of these dynamically changing physical state parameters are captured. Based on these real-time acquired physical state parameters, the availability of the quantum computing unit can be comprehensively characterized from four dimensions: time, accuracy, reliability, and capacity, avoiding the one-sided judgment that may result from a single indicator.

[0045] In some embodiments, the maximum decoherence time of different quantum computing units may range from tens to hundreds of microseconds, and the total number of qubits may also range from tens to hundreds. If the absolute value of the remaining decoherence time (e.g., 50 microseconds) is used directly as the scoring input, then for a QPU with a maximum decoherence time of 100 microseconds, 50 microseconds means 50% of the lifetime remains; but for a QPU with a maximum decoherence time of 500 microseconds, 50 microseconds means only 10% of the lifetime remains—the two represent completely different health states. By calculating the ratio of "remaining decoherence time / maximum decoherence time," the absolute time is converted into a relative remaining proportion, allowing for fair comparison between different QPUs. Similarly, calculating the ratio of "number of available qubits / total number of qubits" to obtain the proportion of available qubits eliminates the influence of differences in the total number of qubits. For the error rate, directly using the raw error rate is insufficient to determine its acceptability. Therefore, the ratio of "calculated error rate / preset error rate threshold" is calculated, reflecting how close the current error rate is to the failure boundary (a ratio <1 indicates it is still acceptable, a ratio ≥1 indicates it has exceeded the fault tolerance capability). Gate fidelity is already a dimensionless number between 0 and 1, requiring no conversion. Through the above normalization process, all input variables are mapped to a similar numerical range (usually 0 to 1), eliminating the incomparability caused by hardware differences between different quantum computing units (such as different absolute values ​​of maximum decoherence time and different total number of bits). This ensures that the health score has cross-device consistency and comparability, laying the foundation for the stable calculation of the subsequent weighted regression model.

[0046] In some cases, different application scenarios may have different sensitivities to different metrics. For example, for deep circuit tasks requiring long sequences of gate operations, the decoherence time ratio should have a higher weight; while for precision measurement tasks requiring high accuracy, gate fidelity should have a higher weight. This embodiment of the invention employs a weighted regression model (i.e., a linear weighted sum: H = ... *(Decoherence time ratio) + *Door fidelity+ * (1 - error rate ratio) + * The ratio of qubits (where the error rate ratio can be converted into a health factor) is used to learn the optimal weight combination through methods such as Bayesian regression or gradient descent based on historical operational data (such as the success rate of previous tasks, the deviation between actual execution time and expectations, etc.). This makes the health score no longer a fixed formula, but can adaptively fit the actual performance pattern of the current quantum computing unit. Simultaneously, the calculated health score H is a value between 0 and 1 (usually H < 0.3 indicates that the quantum resources of the quantum computing unit are unavailable). The scheduling device can directly compare it with a preset health threshold to quickly make a scheduling decision on whether to use the quantum computing unit.

[0047] According to a specific embodiment of the present invention, a quantum resource health scoring model is provided, which is the expression corresponding to the above-mentioned weighted regression model:

[0048] Where H represents the quantum resource health score; The remaining decoherence time of the quantum computing unit; This represents the maximum decoherence time of a quantum computing unit; Indicates the current door fidelity; This indicates the error rate in the calculation; This indicates the preset error rate threshold; Indicates the number of available qubits. Indicates the total number of sub-bits. This indicates the proportion of available qubits. , , , The weight coefficients are obtained through Bayesian regression learning of historical operating data.

[0049] In some embodiments, the step of estimating the quantum computing speedup based on task characteristic parameters includes: obtaining from the task characteristic parameters the problem size corresponding to the computing task, the connectivity characterizing the degree of interaction between qubits in the task, and the proportion of the portion of the computing task that can be quantized as the quantizable proportion; and using the speedup estimation function to calculate the problem size, connectivity, and quantizable proportion to obtain the quantum computing speedup.

[0050] Based on the above settings, a rapid, quantitative, and reproducible prediction of the "quantum acceleration potential" of any computing task is achieved, enabling the scheduling device / system to determine "whether the task is worth using quantum computing" before the task is actually executed. This avoids the waste of resources and performance degradation caused by blindly scheduling computing tasks with low speedup ratios to quantum computing units.

[0051] According to embodiments of the present invention, the speedup potential of quantum computing over classical computing is not valid for all tasks; it is highly dependent on the inherent properties of the task. The above steps screen out three of the most decisive factors from the task's characteristic parameters. Among them, problem size (e.g., the number of nodes in graph theory problems, the matrix dimension in linear algebra problems, and the number of variables in combinatorial optimization problems) directly determines the complexity growth curve of classical algorithms and is also the basis for the scale effect of quantum algorithms (such as Grover's search algorithm and quantum approximation optimization algorithms). Connectivity reflects the interaction density between different variables or qubits in the task—highly connected tasks are generally more suitable for quantum computing because quantum entanglement excels at handling correlations; while low-connectivity tasks may be more suitable for classical decomposition. The quantizability ratio refers to the proportion of code or computational steps in the task that can be effectively accelerated by quantum algorithms. Since any practical task may contain classical preprocessing and post-processing parts, only the higher the quantizability ratio, the more significant the overall speedup effect. By extracting these three parameters, the abstract computational task is transformed into three quantifiable key indicators, which respectively characterize the task size, structural complexity, and quantum fitness, providing comprehensive and semantically clear input features for subsequent quantitative prediction of speedup ratio.

[0052] According to another embodiment of the present invention, the speedup prediction function is a pre-constructed mathematical model that, by analyzing a large amount of actual execution data (quantum execution time, classical execution time) of historical tasks and combining it with theoretical speedup models (such as the square root speedup of Grover's search algorithm, the exponential speedup of Shor's algorithm, etc.), fits the mapping relationship between problem size, connectivity, quantizability ratio, and speedup. This speedup prediction function can be a linear regression model, a nonlinear regression model, or a model based on an empirical lookup table. After inputting the problem size, the speedup prediction function can estimate the time complexity of classical algorithms (e.g., O(N²)) and quantum algorithms (e.g., O(N)), thus obtaining the theoretical speedup. Connectivity is used to correct this theoretical speedup value—high connectivity tasks will have a higher quantum speedup, while low connectivity tasks will have a lower one. The quantization ratio is used to calculate the end-to-end speedup (because only a portion of the code is accelerated, the overall speedup is limited by Amdahl's Law (a theoretical upper limit for performance improvement in parallel computing, stating that the overall speedup is limited by the proportion of the non-parallelizable part of the system)). Through the combined calculation of these three factors, the speedup prediction function can output a relatively accurate predicted speedup value. The scheduler / system can compare this value with a preset speedup threshold (e.g., 2): if the predicted speedup is greater than the preset speedup threshold, it indicates that quantum computing is expected to bring significant benefits, and the computation task can be allocated to a quantum computing unit; otherwise, even if the task is quantizable, it is not worthwhile to occupy scarce quantum resources.

[0053] According to a specific embodiment of the present invention, an expression for estimating the speedup ratio of quantum computing is provided:

[0054] in, This is the estimated speedup ratio for quantum computing; , , These are, respectively, the problem size corresponding to the computational task obtained from the task characteristic parameters, the connectivity characterizing the degree of interaction between qubits in the task, and the proportion of the computational task that can be quantized (as the quantizable proportion). f() represents the speedup prediction function. In one specific implementation, when the predicted quantum computing speedup is obtained... When the value is greater than 2, the computing task can be assigned to the quantum computing unit (it should be noted that the above values ​​are only examples and are not intended to limit the embodiments of the present invention).

[0055] In some embodiments, the step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time further includes: predicting the remaining decoherence time of each quantum computing unit at the current moment by interpolation based on time series analysis of historical operating data.

[0056] Based on the above settings, another specific implementation method for obtaining physical state parameters is provided, which enables the scheduling device to obtain an estimate of the remaining decoherence time with extremely low acquisition cost and real-time requirements. This satisfies the need for quantum resource state perception while protecting the quantum computing unit from decoherence aggravation and computing power loss caused by frequent measurements, thus achieving a balance between state monitoring and resource protection.

[0057] According to embodiments of the present invention, the remaining decoherence time in a quantum computing unit is a physical quantity that varies continuously with time, and its changing trend has certain regularity (e.g., it slowly recovers as idle time increases and decreases stepwise as operations are performed). In order to obtain the precise value at the current moment, it is usually necessary to run specialized benchmark tests (such as the Ramsey interference experiment, which is often used in the field of quantum computing to measure the decoherence time of qubits and evaluate the phase coherence characteristics of qubits by analyzing the decay of interference fringes). Such tests themselves consume quantum resources and may disrupt ongoing computing tasks. The above steps employ a non-invasive alternative: First, the remaining decoherence time data recorded by the quantum computing unit at various time points during its historical operation are collected (e.g., feedback values ​​after the previous task execution, archived data from periodic low-overhead sampling, etc.) to form a time series. Then, trend analysis is performed on the time series (e.g., linear regression, exponential smoothing, or more complex time series models) to identify the decay or recovery patterns of the decoherence time. Finally, based on the current position on the time axis, interpolation methods (e.g., linear interpolation, spline interpolation) are used to estimate the remaining decoherence time at that moment. Since interpolation prediction does not require actively sending measurement pulses to the quantum computing unit, it does not interfere with the coherence of the quantum state, nor does it occupy time slices that could be used for computational tasks. Therefore, without actively initiating benchmark tests or measurement operations, the approximate value of the remaining decoherence time at the current moment can be estimated using the decoherence time variation patterns recorded over a period of time, thus avoiding the interference and additional time overhead caused by frequent active measurements on the quantum state.

[0058] In some embodiments, the method further includes: adaptively switching between blocking call mode and non-blocking call mode according to the task type by a communication adaptation unit; wherein, blocking mode is used for tasks assigned to the quantum computing unit for execution, and non-blocking mode is used for tasks scheduled to the classical computing unit after triggering the automatic rollback mechanism.

[0059] Based on the above settings, the communication mechanism of the hybrid computing cluster has task type awareness capabilities. It can dynamically match the optimal calling mode according to the essential differences between quantum tasks and classical rollback tasks in terms of time constraints, resource consumption, and result dependencies. This maximizes the utilization efficiency of classical computing units while ensuring the real-time performance of quantum tasks, thereby optimizing the overall performance of the cluster.

[0060] According to embodiments of the present invention, quantum tasks and classical rollback tasks have distinctly different execution characteristics. Quantum tasks are constrained by the decoherence time window, have short execution times (typically microseconds to milliseconds), and their results are crucial for subsequent computations. Furthermore, quantum computing units can only process a limited number of tasks at a time. Therefore, adopting a blocking mode simplifies synchronization logic, reduces scheduling overhead, and ensures timely consumption of results. In contrast, classical rollback tasks are essentially classical simulation computations, with longer execution times (milliseconds to seconds), high determinism in results, and lower real-time requirements. Classical computing units typically support multi-task parallelism. If a blocking mode is also adopted for these tasks, the scheduler will be stuck in a waiting state for a long time, preventing other classical tasks from being scheduled in a timely manner and significantly reducing overall throughput. Based on the above settings, the mode switching logic is encapsulated through a communication adaptation unit, allowing the upper-layer scheduler to ignore the underlying communication details and simply mark the task type (quantum task or classical rollback task). The adaptation unit automatically selects the blocking or non-blocking mode. This design ensures low latency on the quantum critical path (blocking mode) while relieving the blocking pressure on the scheduler from classical rollback tasks (non-blocking mode). This allows classical computing units to perform simulations asynchronously in the background, while the scheduler continues to process newly arriving tasks. Consequently, when faced with numerous rollback scenarios, the hybrid computing cluster will not block the entire scheduling queue due to a few long-running classical simulation tasks, thus maintaining high task throughput and resource utilization.

[0061] In the blocking call mode, the scheduler (or calling thread) pauses subsequent operations after initiating a task, waiting for the task to complete and return a result before continuing. This mode is suitable for quantum tasks that have a strong dependency on the result and whose execution time is predictable (usually within the decoherence time window). Because quantum computing units are resource-scarce and have short execution times (microseconds to milliseconds), using the blocking mode avoids complex asynchronous state management while ensuring that quantum results can be immediately used for subsequent classical processing, preventing the decoherence window from being wasted due to delays.

[0062] In non-blocking call mode, the scheduler returns immediately after initiating a task without waiting for the result. The completion status can be retrieved later through a query or callback mechanism. This mode is suitable for tasks scheduled to the classical computing unit after triggering the automatic rollback mechanism (typically classical simulation computations, which may have long execution times and high uncertainty). If a blocking mode is used for classical rollback tasks, the scheduler will idle for a long time, wasting CPU resources and delaying the queuing of subsequent tasks. By switching to non-blocking mode, the scheduler can continue processing other tasks while classical rollback tasks are executing, thereby improving system concurrency and throughput. Furthermore, the communication adaptation unit, as an abstraction layer of the underlying communication interface, can automatically select the calling mode based on the task type label passed from the upper layer, without requiring explicit differentiation from the upper-layer logic, achieving adaptive switching.

[0063] In some of these embodiments, the preset acceleration ratio threshold ranges from [1.5, 5], and the preset health score threshold ranges from [0.1, 0.5].

[0064] Based on the above settings, optimal value ranges that have been tested in practice are provided for the two key decision parameters of the hybrid computing cluster (speedup threshold and health threshold). This enables the scheduling device to quickly configure a scheduling strategy that is neither too aggressive nor too conservative without relying on specific hardware models or task types, thereby ensuring the versatility and performance stability of the system in different scenarios.

[0065] According to embodiments of the present invention, a preset speedup threshold and a preset health threshold jointly determine the behavioral boundaries of the scheduling device. The former controls "which tasks are worth executing using quantum computing units," while the latter controls "when quantum resources are still available." Without constraints on the range of values, different users or different deployment environments may set extreme values, leading to the scheduling device almost never using quantum resources or overusing them. By limiting the optimal range of the preset thresholds, researchers, system administrators, and automatic configuration tools can all select parameters within a reasonable range, avoiding performance crashes due to improper parameter settings.

[0066] On the one hand, the speedup threshold for quantum computing is a key parameter determining whether a computational task is worth allocating to a quantum computing unit. If the speedup threshold is less than 1.5 (e.g., 1.1 or 1.2), then even if the task only provides a very slight speedup (less than 50%), it will be allowed to use quantum resources. However, considering the overhead of task encoding / decoding, queuing, communication, and possible splitting and reassembly, the actual end-to-end speedup is often lower than the theoretical estimate. Too low a speedup threshold will cause a large number of tasks with "minor speedups" to occupy scarce quantum computing units, thus reducing overall throughput. Conversely, if the speedup threshold is greater than 5, only a very small number of tasks that can provide speedups of more than 5 times (such as large number factorization or quantum simulations of a certain scale) can use quantum computing units, while the vast majority of tasks with moderate speedup potential (e.g., 2-4 times) will be rejected, resulting in idle and wasted quantum resources. The preset speedup threshold is set to a range of [1.5, 5], covering the speedup interval from "significant but not significant" to "significant." This excludes weak speedup scenarios where the benefits are insufficient to cover the costs, while retaining a sufficient number of schedulable tasks. This allows the scheduling device to select a specific threshold (e.g., the default 2) based on the load during actual operation, thereby optimizing resource utilization and scheduling efficiency. In other words, the preset speedup threshold range provides the scheduling device with a reasonable, non-extreme speedup judgment boundary. This avoids a large influx of low-yield tasks into the quantum computing unit due to an excessively low threshold, and also avoids the erroneous rejection of high-potential tasks due to an excessively high threshold, thus achieving a balance between quantum resource utilization and task scheduling security.

[0067] On the other hand, the quantum resource health score is designed to be a value between 0 and 1, reflecting the overall availability of the quantum computing unit. If the preset health score threshold is set too high (e.g., greater than 0.5), a slight decrease in the health score (e.g., from 0.8 to 0.6) will trigger a rollback, even though the quantum computing unit may still have sufficient computing power. This will cause many tasks that could have been successfully completed to be prematurely rolled back to classical computing units, reducing the actual utilization rate of quantum resources. If the preset health score threshold is set too low (e.g., less than 0.1), it means that rollback will only be triggered in extremely severe cases where the health score is below 0.1. However, when the health score is between 0.1 and 0.2, the quantum computing unit may already be close to failure (e.g., extremely short remaining decoherence time, error rate close to the threshold). Forcing the execution of tasks at this time will most likely fail, resulting in wasted time. The preset health score threshold is set to a range of [0.1, 0.5], covering the interval from "low availability" to "medium availability," allowing the scheduling device to select the threshold (e.g., the default 0.3) based on specific hardware characteristics and task requirements. This range avoids excessive rollback due to normal fluctuations while ensuring timely switching to classical simulation when resources become truly unavailable, thus guaranteeing task success rate and system stability. In other words, the aforementioned preset health score threshold range ensures that automatic rollback is only triggered when the health score of the quantum computing unit is at a relatively low level, avoiding frequent rollbacks caused by normal fluctuations in the health score (e.g., slight decreases). It also prevents the forced use of quantum resources when the health score is extremely low (close to failure), thereby achieving a balance between system robustness and resource utilization.

[0068] According to an embodiment of the present invention, a specific implementation of a resource scheduling scheme for a hybrid computing cluster is provided, such as... Figure 2 As shown, it mainly includes: (1) Computation task acquisition and task feature parameter extraction Hybrid computing task queue: First, retrieve the computing tasks to be processed from the hybrid computing task queue. These tasks include the parts that need to be processed by quantum computing.

[0069] Extracting task feature parameters: Extracting key task feature parameters from the computational task for subsequent analysis and decision-making.

[0070] (2) Core assessment and prediction (parallel processing) At this stage, the evaluation is conducted along two main paths: Path 1: Quantum Speedup Prediction. Based on the extracted task feature parameters, predict the speedup of the computational task in a quantum computing environment (denoted as A).

[0071] Path 2: Real-time monitoring of quantum resources. (a) Real-time monitoring of quantum resources: Monitor the current state of quantum resources in real time with a period of 100ms (T2) or based on the gate operation frequency (F_gate). (b) Joint scoring of health H: Combine the monitoring data to construct a health scoring mechanism and establish a degradation state evaluation function (i.e., the aforementioned weighted regression model) to quantitatively evaluate the current health status and performance degradation degree of the quantum processor (QPU).

[0072] (3) Decision branch logic Two levels of judgment will be made based on the above assessment results.

[0073] First-level judgment: Speedup ratio threshold. Judgment condition: Speedup ratio A > 2? If not (i.e., if the speedup is less than 2x), the computational task is rolled back to classical computing units (CPU / GPU) for processing, thus avoiding the use of expensive quantum resources. If yes (if the speedup is greater than 2x), then proceed to the next level of judgment.

[0074] Second-level judgment: QPU health threshold. Judgment condition: Optimal QPU health H > 0.3? If not (i.e., if the current optimal QPU's health is below 0.3), the computation task is rolled back to classical computing units (CPU / GPU) for processing. If yes (i.e., if the health meets the requirements), it is confirmed as a suitable task for quantum computing and enters the scheduling and execution phase.

[0075] (4) Scheduling Execution and Optimization For tasks that pass the double-check process, perform fine-grained scheduling operations: (a) Coherence Time Aware Segmentation Module: This module ensures that the depth of the quantum circuit is limited before the decoherence time (T2) window. This means that the task will be segmented or adjusted to fit the coherence time limit of the qubit and prevent computational errors caused by decoherence.

[0076] (b) Precise allocation: The processed quantum circuits are precisely allocated to the quantum processors (QPUs) with the highest health scores (H scores) for execution to ensure the success rate and efficiency of the computation.

[0077] Final result: Through the above process, the architecture can avoid ineffective decomposition and significantly improve the utilization efficiency of QPU resources.

[0078] The resource scheduling method for the hybrid computing cluster provided in this embodiment of the invention involves: acquiring the computing tasks to be processed and extracting the task characteristic parameters of the computing tasks; estimating the quantum computing speedup ratio based on the task characteristic parameters; if the quantum computing speedup ratio is greater than a preset speedup ratio threshold, then allocating the computing tasks to quantum computing units for execution; acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time and calculating the quantum resource health score based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of quantum states; the quantum resource health score is used to dynamically characterize the available state of quantum resources; according to the quantum resource health score and the decoherence time window determined by the remaining decoherence time, the computing tasks allocated to the quantum computing units for execution are divided into multiple quantum subtasks based on coherence time constraints; before the execution of the quantum subtasks, if it is determined that the expected execution time of any quantum subtask exceeds the corresponding decoherence time window, or the quantum resource health score is lower than a preset health score threshold, then an automatic rollback mechanism is triggered, and the corresponding quantum subtask is scheduled to a classical computing unit for computation. By first determining whether the computational task to be processed is suitable for quantum computing, then assessing the health of the quantum computing unit, then dividing the computational task according to the coherence time window, and finally verifying the division result before execution, and using classical computing units for simulation as a fallback path, the risk of resource status continuing to deteriorate after division is mitigated. This achieves a significant improvement in computing power utilization and an effective reduction in scheduling latency, thus improving resource scheduling efficiency, reducing computing resource waste, and increasing task processing efficiency.

[0079] Based on the resource scheduling method for hybrid computing clusters provided in the embodiments of the present invention, the embodiments of the present invention also provide a resource scheduling device for hybrid computing clusters, the hybrid computing clusters including quantum computing units and classical computing units; Figure 3 As shown, the resource scheduling device 300 of the hybrid computing cluster includes: The first scheduling module 301 is used to acquire computing tasks to be processed, extract task feature parameters of computing tasks, estimate quantum computing speedup based on task feature parameters, and if the quantum computing speedup is greater than a preset speedup threshold, allocate computing tasks to quantum computing units for execution. The computing module 302 is used to acquire the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time, and calculate the quantum resource health score based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of the quantum state; the quantum resource health score is used to dynamically characterize the available state of the quantum resource; The task segmentation module 302 is used to segment the computing tasks assigned to the quantum computing unit based on coherence time constraints according to the quantum resource health score and the decoherence time window determined by the remaining decoherence time, and generate multiple quantum subtasks. The second scheduling module 304 is used to, before the execution of a quantum task, if it is determined that the expected execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than the preset health score threshold, trigger an automatic rollback mechanism to schedule the corresponding quantum task to the classical computing unit for computation.

[0080] Based on the above setup, by first determining whether the computational task to be processed is suitable for quantum computing, then assessing the health of the quantum computing unit, then dividing the computational task based on coherent time constraints, and finally verifying the division result before task execution, and using classical computing units for simulation as a fallback path, the risk of resource status continuing to deteriorate after task division is mitigated. This achieves a significant improvement in computing power utilization and an effective reduction in scheduling latency, thus improving resource scheduling efficiency, reducing computing resource waste, and increasing task processing efficiency.

[0081] Meanwhile, the resource scheduling device 300 of the aforementioned hybrid computing cluster is configured to execute the resource scheduling method of any of the aforementioned hybrid computing clusters. Therefore, the relevant modules in the scheduling device are also used to execute the corresponding operations in the resource scheduling method of any of the aforementioned hybrid computing clusters. Accordingly, it also possesses all the beneficial effects of the resource scheduling method of any of the aforementioned hybrid computing clusters, which will not be elaborated here.

[0082] It should be noted that the specific modules in the resource scheduling device of the above-mentioned hybrid computing cluster are defined mainly based on the corresponding operations performed, and are not intended to limit the specific modules.

[0083] This invention also provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this invention.

[0084] This invention also provides a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the methods of embodiments of this invention. The computer program product should be understood as a software product that primarily implements the methods of this invention through a computer program.

[0085] This invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of this invention.

[0086] refer to Figure 4 The present invention will now be described in the form of a structural block diagram of an electronic device that can serve as an embodiment of the present invention, which is an example of a hardware device that can be applied to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0087] like Figure 4 As shown, the electronic device includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. The RAM 403 may also store various programs and data required for the operation of the electronic device. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0088] Multiple components in the electronic device are connected to I / O interface 405, including: input unit 406, output unit 407, storage unit 408, and communication unit 409. Input unit 406 can be any type of device capable of inputting information into the electronic device. Input unit 406 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 407 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 408 may include, but is not limited to, disks and optical discs. Communication unit 409 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0089] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as a computer program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0090] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0091] In the context of embodiments of the present invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0092] It should be noted that the term "comprising" and its variations used in the embodiments of the present invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "multiple" mentioned in the embodiments of the present invention are illustrative and not restrictive. Those skilled in the art should understand that, unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0093] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in the embodiments of the present invention are all information and data that have been permitted by the user or have been fully agreed upon by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to agree or refuse.

[0094] The steps described in the method embodiments provided by this invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of this invention is not limited in this respect.

[0095] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places in the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0096] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A resource scheduling method for a hybrid computing cluster, characterized in that, The hybrid computing cluster includes quantum computing units and classical computing units, and the method includes: The system acquires a computing task to be processed and extracts the task feature parameters. Based on the task feature parameters, it estimates the quantum computing speedup ratio. If the quantum computing speedup ratio is greater than a preset speedup ratio threshold, the system assigns the computing task to the quantum computing unit for execution. The physical state parameters of each quantum computing unit in the hybrid computing cluster are acquired in real time, and a quantum resource health score is calculated based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the temporal degradation characteristics of quantum states; and the quantum resource health score is used to dynamically characterize the availability state of quantum resources. Based on the quantum resource health score and the decoherence time window determined by the remaining decoherence time, the computing tasks assigned to the quantum computing unit are divided into multiple quantum subtasks based on coherence time constraints. Before the execution of the quantum task, if it is determined that the expected execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than the preset health score threshold, an automatic rollback mechanism is triggered to schedule the corresponding quantum task to the classical computing unit for computation.

2. The method according to claim 1, characterized in that, The task segmentation steps based on coherent time constraints include: The maximum allowed circuit depth within the decoherence time window is calculated based on the decoherence time window, the gate operation time of the quantum computing unit performing a single quantum gate operation, and the fidelity information contained in the quantum resource health score. The computational task assigned to the quantum computing unit for execution is divided into multiple quantum subtasks, such that the circuit depth corresponding to each quantum subtask is less than or equal to the maximum circuit depth.

3. The method according to claim 1, characterized in that, Prior to the task segmentation step based on coherent time constraints, the method further includes: Based on the partitioning strategy corresponding to the computing task, the estimated number of quantum tasks generated after task partitioning is estimated, and based on the estimated number, the communication overhead required to merge and reconstruct the computing results of each quantum task in the classical computing unit is calculated. If the communication overhead exceeds a preset overhead threshold, then the computation task is prohibited from being split, and the computation task is allocated as a whole to the classic computing unit for execution.

4. The method according to claim 1, characterized in that, The step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time and calculating the quantum resource health score based on the physical state parameters includes: Multiple physical state parameters characterizing the operating state of the quantum computing unit are acquired in real time; wherein, the physical state parameters include the remaining decoherence time, gate fidelity, computational error rate, and number of available qubits; Based on the physical state parameters, determine the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the computation error rate to the preset error rate threshold, and the proportion of available qubits for the quantum computing unit. The quantum resource health score is calculated based on a weighted regression model; wherein the input variables of the weighted regression model include: the ratio of the remaining decoherence time to the maximum decoherence time, the current gate fidelity, the ratio of the calculation error rate to the preset error rate threshold, and the proportion of available qubits.

5. The method according to claim 1, characterized in that, The step of estimating the quantum computing speedup based on the task characteristic parameters includes: The problem size corresponding to the computing task, the connectivity characterizing the degree of interaction between qubits in the task, and the proportion of the computing task that can be quantized are obtained from the task characteristic parameters, as the quantizable proportion. The quantum computing speedup is obtained by calculating the problem size, the connectivity, and the quantizability ratio using a speedup prediction function.

6. The method according to claim 1, characterized in that, The step of acquiring the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time further includes: Based on time series analysis of historical operating data, interpolation is used to predict the remaining decoherence time of each quantum computing unit at the current moment.

7. The method according to claim 1, characterized in that, Also includes: The communication adaptation unit adaptively switches between blocking and non-blocking call modes based on the task type. Specifically, tasks assigned to the quantum computing unit are executed in a blocking mode, while tasks scheduled to the classical computing unit after triggering the automatic rollback mechanism are executed in a non-blocking mode.

8. The method according to claim 1, characterized in that, The preset acceleration ratio threshold ranges from [1.5, 5], and the preset health score threshold ranges from [0.1, 0.5].

9. A resource scheduling device for a hybrid computing cluster, characterized in that, The hybrid computing cluster includes quantum computing units and classical computing units, and the device includes: The first scheduling module is used to acquire computing tasks to be processed, extract task feature parameters of the computing tasks, estimate the quantum computing speedup based on the task feature parameters, and if the quantum computing speedup is greater than a preset speedup threshold, then the computing task is assigned to the quantum computing unit for execution. The computing module is used to acquire the physical state parameters of each quantum computing unit in the hybrid computing cluster in real time, and calculate the quantum resource health score based on the physical state parameters; wherein, the physical state parameters include the remaining decoherence time used to characterize the time degradation characteristics of quantum states; the quantum resource health score is used to dynamically characterize the availability state of quantum resources; The task segmentation module is used to segment the computing tasks assigned to the quantum computing unit based on coherence time constraints according to the quantum resource health score and the decoherence time window determined by the remaining decoherence time, thereby generating multiple quantum subtasks. The second scheduling module is used to, before the execution of the quantum task, if it is determined that the estimated execution time of any quantum task exceeds the corresponding decoherence time window, or the quantum resource health score is lower than a preset health score threshold, trigger an automatic rollback mechanism to schedule the corresponding quantum task to the classical computing unit for computation.

10. A non-transitory machine-readable medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 8.

Images