A distributed communication system, data communication method and device, and storage medium

Abstract

The application discloses a distributed communication system, a data communication method, equipment and a storage medium, and relates to the technical field of quantum communication. The system comprises a host, a network forwarding device, a control endpoint and a quantum processing unit. The control endpoint is used for collecting quantum measurement and control data from the connected quantum processing unit, encapsulating a data frame and sending the data frame to the network forwarding device, receiving a feedback data frame, and performing corresponding control actions on the connected quantum processing unit according to the feedback result in the feedback data frame. The network forwarding device is used for receiving the data frame from the control endpoint and sending the data frame to the host, so that the quantum measurement and control data in the data frame is directly written into the memory of a graphic processing unit in the host, and forwarding the feedback data frame containing the feedback result to the control endpoint. The graphic processing unit is used for processing the quantum measurement and control data according to preset processing logic to obtain the feedback result. The application realizes low-delay direct-connection data transmission between the control endpoint and the graphic processing unit.

Description

Technical Field

[0001] This application relates to the field of quantum computing technology, and more specifically, to a distributed communication system, data communication method and device, and storage medium. Background Technology

[0002] During the execution of real-time feedback and closed-loop control in quantum computing, the control endpoint located on the quantum processing unit (QPU) side needs to send the collected quantum measurement and control data to the host's graphics processing unit (GPU) for real-time processing. The GPU generates feedback results and then returns them to the control endpoint to perform real-time control actions on the quantum processing unit.

[0003] Existing data communication schemes typically center on a central processing unit (CPU). Data collected by the control endpoint is first written to the host system memory, then processed by the CPU thread scheduling, protocol stack, and memory copy before being forwarded to the graphics processing unit. Feedback results also need to be relayed by the CPU packet by packet before being returned to the control endpoint. However, in real-time quantum operations with strictly limited feedback time windows, the CPU-involved packet relay introduces uncertainties such as memory copying, thread scheduling, system calls, and protocol stack processing, making it difficult to control end-to-end data transmission latency and jitter, and increasing the risk of feedback timeout.

[0004] Therefore, how to achieve low-latency direct data transmission between the control endpoint and the graphics processing unit is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this application is to provide a distributed communication system, data communication method and device and storage medium that realizes low-latency direct data transmission between the control endpoint and the graphics processing unit.

[0006] To achieve the above objectives, this application provides a distributed communication system, comprising: a host, a network forwarding device, at least one control endpoint, and at least one quantum processing unit connected to the control endpoint, wherein the host includes a graphics processing unit; The control endpoint is used to acquire quantum measurement and control data from the connected quantum processing unit, encapsulate data frames based on the quantum measurement and control data, and send the data frames to the network forwarding device; and to receive feedback data frames and execute corresponding control actions on the connected quantum processing unit according to the feedback results in the feedback data frames. The network forwarding device is configured to receive the data frame from the control endpoint, send the data frame to the host so that the quantum measurement and control data in the data frame is directly written into the memory of the graphics processing unit; and to forward the feedback data frame containing the feedback result to the control endpoint. The graphics processing unit is used to process the quantum measurement and control data according to a preset processing logic to obtain the feedback result, and then send the feedback data frame generated based on the feedback result to the control endpoint via the network forwarding device.

[0007] Optionally, a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device are pre-allocated in the memory of the graphics processing unit; The network forwarding device is further configured to: write the frame header of the data frame into the control state area, and write the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; The graphics processing unit is used to: after acquiring the data frame, read the quantum measurement and control data from the input buffer, process it according to the corresponding preset processing logic to obtain the feedback result, and write the feedback result into the output buffer; The network forwarding device is also used to: read the feedback result from the output buffer and send it to the control endpoint.

[0008] Optionally, the feedback data frame carries a feedback deadline; The control endpoint is further configured to: upon receiving the feedback data frame, determine whether the feedback result is within a valid time window based on the arrival time and the feedback deadline in the feedback data frame; if it is within the valid time window, execute the control action on the connected quantum processing unit; if it is not within the valid time window, execute a preset local protection strategy.

[0009] Optionally, the preset local protection strategy includes at least one of the following: maintaining the Pauli frame of the previous round, marking the quantum operation of the current round as invalid, or skipping the feedback operation of the current round.

[0010] Optionally, the frame header includes the real-time cutoff time of the quantum measurement and control data, and the network forwarding device is used to: parse the real-time cutoff time and forward the data frame to the host according to the real-time cutoff time.

[0011] Optionally, the frame header includes the priority of the quantum measurement and control data; The network forwarding device is configured to: after receiving the data frame, write the data frame into the corresponding priority queue according to the priority, and send the data frames in the priority queue to the host in order of priority from high to low.

[0012] Optionally, the system further includes a time synchronization module, which provides a unified time reference for the network forwarding device, the control endpoint, and the host.

[0013] Optionally, the control endpoint is specifically used for: acquiring raw quantum measurement and control data from the connected quantum processing unit, and performing a filtering operation on the raw quantum measurement and control data to obtain quantum measurement and control data; wherein the filtering operation includes at least one of threshold decision, effective window truncation, channel enable check, idle channel filtering, and duplicate data removal.

[0014] Optionally, the host is configured to: determine the endpoint identifier of at least one control endpoint participating in the communication, and send the endpoint identifier to the network forwarding device; The network forwarding device is also used to: establish a forwarding path based on the endpoint identifier as the data flow between the graphics processing unit and the control endpoint.

[0015] Optionally, the preset processing logic includes at least one of readout classification, syndrome decoding, error path estimation, and feedback decision calculation; The feedback result includes at least one of Pauli frame increment, correction pulse selection, active reset command, and subsequent gate operation branch.

[0016] Optionally, the host is further configured to: Monitoring data is acquired from the graphics processing unit at a preset frequency; wherein the monitoring data includes at least one of latency distribution, packet loss rate, number of feedback deadline defaults, time offset, and endpoint status; Adjust the effective time window for the feedback deadline of the next round or switch the decoding algorithm based on the monitoring data.

[0017] To achieve the above objectives, this application provides a data communication method applied to a network forwarding device in a distributed communication system as described above, the method comprising: Receive data frames from a control endpoint; wherein the control endpoint acquires quantum measurement and control data from a connected quantum processing unit and encapsulates the data frames based on the quantum measurement and control data; The data frame is sent to the host so that the quantum measurement and control data in the data frame is directly written into the memory of the graphics processing unit in the host. The system receives feedback results from the graphics processing unit; wherein the graphics processing unit processes the quantum measurement and control data according to a preset processing logic to obtain the feedback results. The feedback data frame containing the feedback result is forwarded to the control endpoint so that the control endpoint can perform corresponding control actions on the connected quantum processing unit based on the feedback result in the feedback data frame.

[0018] Optionally, a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device are pre-allocated in the memory of the graphics processing unit; Accordingly, sending the data frame to the host includes: Write the frame header of the data frame into the control state area, and write the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; Accordingly, receiving feedback results from the graphics processing unit includes: The feedback result is read from the output buffer.

[0019] Optionally, the data frame carries the real-time cutoff time of the quantum measurement and control data, and sending the data frame to the host includes: The real-time deadline is parsed, and the data frame is forwarded to the host according to the real-time deadline.

[0020] Optionally, the feedback data frame carries the feedback deadline of the quantum measurement and control data, and forwarding the feedback data frame containing the feedback result to the control endpoint includes: The feedback deadline is parsed, and a feedback data frame containing the feedback result is forwarded to the control endpoint according to the feedback deadline.

[0021] Optionally, it also includes: A forwarding path is established between the graphics processing unit and the control endpoint based on the endpoint identifier of the control endpoint participating in the communication.

[0022] To achieve the above objectives, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the data communication method described above.

[0023] To achieve the above objectives, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the data communication method described above.

[0024] The distributed communication system provided in this application establishes a direct data path between the control endpoint and the graphics processing unit of the host through a network forwarding device. This enables closed-loop transmission of quantum measurement and control data and feedback results directly between the control endpoint and the graphics processing unit without relying on CPU packet-by-packet relay. Specifically, the network forwarding device receives data frames encapsulated based on quantum measurement and control data from the control endpoint and sends the data frames to the host. This allows the quantum measurement and control data in the data frames to be directly written into the memory of the graphics processing unit. Since this process does not require CPU memory copying and protocol stack processing, it effectively reduces the latency and jitter of uplink data transmission. Subsequently, after the graphics processing unit processes the quantum measurement and control data and generates feedback results, the network forwarding device receives the feedback data frame carrying the feedback results returned by the host and forwards the feedback data frame to the control endpoint. This allows the control endpoint to execute control actions on the quantum processing unit, enabling the entire data closed loop to be completed through the relay of the network forwarding device without relying on CPU packet-by-packet relay. Therefore, this application achieves direct closed-loop data transmission between the control endpoint and the graphics processing unit through a network forwarding device, avoiding the latency and jitter introduced by CPU relay, significantly reducing end-to-end data transmission latency, and improving the timing determinism of the feedback link in real-time quantum operations. This application also discloses a data communication method, an electronic device, and a computer-readable storage medium, which can achieve the same technical effects.

[0025] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit this application. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. The drawings are used to provide a further understanding of this disclosure and constitute a part of the specification. They are used together with the following detailed description to explain this disclosure, but do not constitute a limitation of this disclosure. In the drawings: Figure 1 This is an architecture diagram of a quantum operating system in related technologies; Figure 2 This is an architectural diagram of a distributed communication system according to an exemplary embodiment; Figure 3 This is an architectural diagram of another distributed communication system according to an exemplary embodiment; Figure 4 This is a flowchart illustrating a data communication method according to an exemplary embodiment; Figure 5This is a structural diagram of an electronic device according to an exemplary embodiment. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of this application.

[0028] It should be noted that, in the description of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. The terms "first," "second," etc., in this application are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0029] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] To serve the real-time or low-latency closed-loop control requirements of large-scale quantum bit systems, such as quantum error correction and fault-tolerant quantum computing tasks, the following section first combines... Figure 1 This section describes the quantum operating system and communication architecture in related technologies. For example... Figure 1 As shown, the quantum operating system architecture in related technologies adopts a three-tier structure: the host computer running the quantum operating system serves as the control core, one or more field-programmable gate arrays (FPGAs) or radio frequency system-on-chip (RFSoCs) are responsible for pulse driving and signal acquisition of qubits, and the physical quantum processing unit serves as the target to be manipulated and measured. The quantum operating system is deployed on the host computer, responsible for creating and managing experimental sessions, configuring and distributing experimental parameters, compiling and scheduling quantum task programs, recording runtime logs, and managing experimental results. The FPGA or RFSoC is directly physically connected to the QPU (Quantum Processing Unit) via a measurement and control link, performing tasks such as playing pulse waveforms, reading out and acquiring qubit states, synchronously controlling multi-channel trigger signals, and locally processing some low-level feedback logic during experimental execution.

[0031] When quantum operations require GPU acceleration to perform real-time computational tasks such as readout classification, syndrome decoding, active reset decision-making, online calibration, or mid-course measurement feedback, existing systems use the GPU as a general-purpose computing peripheral for the host. Quantum measurement and control data acquired by the FPGA or RFSoC first enters the CPU memory via an external bus or network interface. Then, after undergoing operating system thread scheduling, network protocol stack unpacking, software layer callback processing, or message queue queuing, the CPU finally submits the data to the GPU for processing. After generating feedback results, the GPU also needs to return to the FPGA or RFSoC via the CPU memory path. In this path, each data packet must be handled and forwarded by the CPU packet by packet, adding multiple layers of software stack-introduced latency uncertainties to the closed-loop path from measurement to feedback. Therefore, this path is more suitable for scenarios that tolerate larger latency, such as non-real-time analysis, offline data post-processing, or batch processing acceleration, but it is difficult to stably control the end-to-end feedback latency within the microsecond range, posing a severe challenge to real-time error correction in large-scale quantum bit systems.

[0032] In terms of optimizing data transfer between external devices and GPUs, technologies such as RDMA, GPUDirect, and high-speed Ethernet offer some room for improvement. These technologies can reduce data transfer overhead by bypassing CPU memory copying or operating system kernel protocol stacks, but their application scenarios are usually concentrated in general sensor data aggregation, high-performance computing cluster interconnection, or point-to-point single-link communication verification. In the field of quantum computing measurement and control, real-time quantum operating systems place more specific requirements on communication links, such as distinguishing the data sources of different QPU endpoints, accurately aligning data with specific experimental sessions and error correction rounds, recording the transmission time and feedback deadline of each frame, identifying the quantum task type of the intra-frame payload (such as measurement frames, syndrome frames, or correction instruction frames), and supporting distributed expansion capabilities that allow multiple QPU endpoints to communicate with the GPU simultaneously. Existing RDMA or GPUDirect solutions do not provide native support for the semantic and architectural requirements unique to quantum measurement and control, and therefore do not yet have the conditions to form a complete distributed high-speed direct-connection communication architecture under the framework of a quantum operating system.

[0033] Therefore, the above schemes have the following shortcomings: First, the CPU-centralized data path introduces multi-layer software stack latency and scheduling jitter, making it difficult to reliably maintain end-to-end feedback latency within the strict deadline, increasing the risk of feedback timeout in large-scale quantum error correction. Second, existing direct connection schemes are all one-to-one point-to-point communication modes, lacking a distributed communication architecture for multiple QPU endpoints, and cannot support the requirements of parallel operation and unified scheduling of multiple measurement and control endpoints in large-scale quantum bit systems. Third, the protocol fields of ordinary network messages do not define the endpoint identifier, experimental session identifier, round identifier, timestamp, and real-time deadline required for quantum operations, making it impossible to directly distinguish the quantum task type and real-time requirements of data frames at the network level. Fourth, the retransmission confirmation mechanism adopted by traditional networks to ensure reliability may introduce unacceptable additional waiting under the microsecond-level feedback timescale of quantum operations, and the correct feedback result arriving after the timeout has actually lost its control effect on the quantum operation of that round. Therefore, a real-time validity judgment criterion and local degradation protection strategy that are different from general reliable communication protocols are needed.

[0034] To address the aforementioned shortcomings, the primary technical problem this application aims to solve is: within a real-time quantum operating system framework, how to enable the GPU as a real-time computing unit to communicate with multiple QPU-side FPGA control endpoints via a 100G or higher-speed distributed direct-connect network, while controlling the end-to-end latency and jitter of measurement and feedback data to within microseconds, and uniformly identifying, scheduling, forwarding, and verifying the integrity of information such as endpoint source, experimental session, error correction rounds, and real-time deadlines during transmission. Building upon this, a further problem to be solved is: how to decouple the quantum operating system's control and management plane from the GPU-QPU real-time data communication plane, allowing the CPU to only handle experimental session creation, parameter configuration, and asynchronous monitoring of operational status, while the real-time data plane, composed of the GPU, high-speed network card, high-speed switch, and QPU-side control endpoints, independently handles the packet-by-packet high-speed closed-loop transmission of measurement and feedback data, thereby achieving separation of responsibilities and performance decoupling between the control and management planes and the real-time communication plane.

[0035] This embodiment discloses a distributed communication system, such as Figure 2 As shown, it includes: a host 10, a network forwarding device 20 connected to the host 10, at least one control endpoint 30 connected to the network forwarding device 20, and at least one quantum processing unit 40 connected to the control endpoint 30. The host 10 includes a graphics processing unit 101. The distributed communication system refers to a system architecture for quantum computing measurement and control scenarios, which uses a network forwarding device 20 to achieve high-speed direct communication between the graphics processing unit 101 in the host 10 and multiple control endpoints 30. The host 10 is a high-performance computing server or real-time computing workstation running a quantum operating system (QOS) to generate experimental sessions, configure control endpoints, issue real-time computing tasks, and maintain endpoint mapping relationships. The graphics processing unit (GPU) 101 is a hardware acceleration unit within the host 10 used to perform real-time quantum computing processing. A direct data path can be established between the control endpoints 30 and the GPU 101 of the host 10 through the network forwarding device 20, which forwards data between the control endpoints 30 and the GPU 101. As a feasible implementation method, such as... Figure 3 As shown, the network forwarding device is a stand-alone switch 201, a network switching device that supports high-speed ports and real-time service scheduling. Its high-speed ports can be 100G, 200G, 400G, or 800G or higher speeds, and it supports dedicated network partitions for real-time services, priority queuing, and congestion control functions. The host 10 also includes a network interface card (NIC) 102 or a data processing unit (DPU) for accessing the memory of the graphics processing unit 101 and communicating with the switch 201. Alternatively, the network forwarding device 20 can be a NIC or a DPU, integrating the aforementioned data forwarding functions, dedicated network partitions for real-time services, priority queuing, and congestion control functions into the host's NIC or DPU. The control endpoint 30 can be a field-programmable gate array (FPGA), a radio frequency system-on-a-chip (RFSoC), a pulse processing unit, a measurement and control chassis, or other devices capable of generating and receiving quantum real-time data frames. The quantum processing unit (QPU) 40 is a physical quantum bit chip or a quantum processor, and the control endpoint 30 is connected to it via a measurement and control link.

[0036] The control endpoint 30 is used to: acquire quantum measurement and control data from the connected quantum processing unit, encapsulate data frames based on the quantum measurement and control data, and send the data frames to the network forwarding device 20; In practice, during each quantum operation round, the control endpoint 30 acquires raw quantum measurement and control data from the connected quantum processing unit 40's measurement and control link, including raw I / Q sample values, readout amplitudes, channel flags, sampling window information, and qubit state information. The control endpoint 30 organizes and encapsulates the acquired data to generate a data frame conforming to a preset data frame format. The data frame fields include frame type, QPU endpoint identifier, FPGA endpoint identifier, experimental session identifier, round identifier, sequence number, timestamp, real-time deadline, payload length, and quantum task data. After encapsulation, the control endpoint 30 sends the data frame through its high-speed network interface, which is then forwarded to the host 10 via network forwarding device 20.

[0037] The meanings and technical functions of each field in the data frame are as follows: The frame_type field is used to identify the type of data frame, including measurement frames, syndrome frames, feedback frames, correction frames, heartbeat frames, or status frames, so that the GPU can select the corresponding real-time processing function according to the frame type.

[0038] The qpu_id and endpoint_id fields are used to identify the QPU and the FPGA control endpoint on the QPU side. They support multi-endpoint distributed routing, enabling network forwarding devices to correctly forward data frames to the corresponding control endpoint or correctly route feedback data frames to the target control endpoint.

[0039] The session_id field is used to identify the current experimental session, and the round_id field is used to identify the current quantum operation round, so that the data is aligned with the specific experimental task and quantum operation round, making it easier for the GPU to associate and process data from the same session and the same round.

[0040] The sequence_number field is used to indicate the order of messages, and is used at the receiving end to detect packet loss, duplicate packets, or out-of-order packets, ensuring the integrity and order of data frames.

[0041] The timestamp field is used to record the sending time, sampling time, or receiving time of a data frame. It is used to calculate the end-to-end transmission delay and verify whether the feedback is within the valid time window.

[0042] The deadline field is used to indicate the latest valid time of the current feedback result, and is used to control whether the endpoint should trigger timeout processing or local protection policy when it receives a feedback data frame.

[0043] The payload field is used to carry the actual measurement values, syndromes, calibration results, or control parameters to complete the actual quantum real-time task processing.

[0044] In a preferred embodiment, the control endpoint 30 is specifically used to: acquire raw quantum measurement and control data from the connected quantum processing unit, and perform a filtering operation on the raw quantum measurement and control data to obtain quantum measurement and control data; wherein, the filtering operation includes at least one of threshold decision, effective window truncation, channel enable check, idle channel filtering, and duplicate data removal.

[0045] Threshold decision refers to comparing the amplitude or signal-to-noise ratio of the original readout signal against a threshold. Effective window truncation refers to extracting data within a preset sampling time window. Channel enable check refers to filtering out data from disabled channels based on their enable status in the experimental configuration. Idle channel filtering refers to identifying and discarding data from channels that are not connected to qubits or are not participating in the current experiment. Duplicate data removal refers to detecting and removing duplicate sampled data caused by hardware failure or signal reflection.

[0046] In practical implementation, during each error correction round or readout cycle, the control endpoint 30 acquires the raw I / Q sample values, readout amplitude, channel flags, and sampling window information from the connected quantum processing unit 40's measurement and control link, and sequentially performs the aforementioned filtering operations. Taking the real-time syndrome decoding scenario in a superconducting quantum bit system as an example, each control endpoint 30 corresponds to a set of readout channels. After each error correction round, the control endpoint 30 first performs initial filtering on the acquired data locally. Signals whose readout amplitude or signal-to-noise ratio does not reach a preset threshold are judged as invalid and filtered out through threshold decision. Data within a preset sampling time window is retained through the effective window. Data from disabled channels is filtered out through channel enable checks. Data from channels not connected to qubits or not participating in the current experiment is discarded through idle channel filtering. Duplicate data generated by hardware failures or signal reflections is removed through duplicate data removal. After filtering, the control endpoint 30 organizes the retained effective readout results, syndrome bits, confidence information, or anomaly flags into a compact quantum task payload, which is then encapsulated into a data frame as quantum measurement and control data. As can be seen, by performing the filtering operation locally at the control endpoint 30, only the valid information related to the current quantum feedback control is written into the data frame, which reduces the amount of useless information entering the direct communication link between the control endpoint 30 and the graphics processing unit 101, thereby reducing the link bandwidth usage and the processing overhead of the graphics processing unit 101.

[0047] The network forwarding device 20 is used to: receive data frames from the control endpoint and send the data frames to the host so that the quantum measurement and control data in the data frames is directly written into the memory of the graphics processing unit; In specific implementation, the network forwarding device 20 receives data frames from multiple control endpoints 30 through its high-speed port, and parses and forwards the data frames according to the pre-configured forwarding paths and flow table rules. Since a direct data transmission path is established between the network forwarding device 20 and the high-speed network card or data processing unit of the host 10, after the data frame arrives at the host 10, the quantum measurement and control data in it can be directly written to the pre-allocated memory area of ​​the graphics processing unit 101 without going through the packet-by-packet relay of the central processing unit (CPU), but through Remote Direct Memory Access (RDMA) or an equivalent direct transmission mechanism.

[0048] As one possible implementation, the data frame carries the real-time cutoff time of the quantum measurement and control data, and the network forwarding device is used to: parse the real-time cutoff time, and forward the data frame to the host according to the real-time cutoff time.

[0049] In practical implementation, when encapsulating data frames, the control endpoint 30 marks the real-time deadline for the quantum measurement and control data according to the timing requirements of the current quantum operation round, which is used to indicate the effective time of the quantum measurement and control data. After receiving the data frame, the network forwarding device 20 parses the real-time deadline field in the frame header and schedules the forwarding of data frames according to the proximity of the real-time deadline, giving priority to forwarding data frames with closer real-time deadlines, thereby ensuring that the quantum measurement and control data can reach the host 10 in a timely manner.

[0050] As another feasible implementation, the data frame carries the priority of the quantum measurement and control data; the network forwarding device is used to: after receiving the data frame, write the data frame into the corresponding priority queue according to the priority, and send the data frames in the priority queue to the host in order of priority from high to low.

[0051] In practical implementation, when encapsulating data frames, the control endpoint 30 assigns priority to each frame based on the type of service carried by the quantum measurement and control data. For example, data frames carrying measurement data required for quantum error correction syndrome decoding, active reset decision data, or real-time branch decision data can be assigned high priority, as the processing results of these data directly affect the timing correctness of subsequent quantum operations; data frames carrying periodic heartbeat status reporting, operation logs, or offline calibration data can be assigned normal or low priority. After receiving a data frame, the network forwarding device 20 parses the priority field in the frame header and arranges the data frame into the corresponding priority queue according to the priority value. During dequeueing, the network forwarding device 20 adopts a strict priority scheduling strategy: it first checks if there are any data frames to be sent in the high-priority queue, and if so, sends them first, until the high-priority queue is empty before checking the next priority queue. Through this scheduling mechanism, data frames carrying real-time closed-loop feedback services can obtain port bandwidth and forwarding resources at all times, avoiding head-blocking problems caused by low-real-time traffic queuing at the front, thus ensuring low-latency transmission of critical feedback data frames in scenarios where multiple control endpoints send data in parallel.

[0052] Of course, the network forwarding device 20 can also determine the forwarding order of data frames by considering multiple factors, and this embodiment does not impose specific limitations. For example, considering the priority and real-time deadline of the data frames, for each data frame to be forwarded, a priority score is calculated based on its priority (e.g., 3 points for high priority, 2 points for medium priority, and 1 point for low priority), and an urgency score is calculated based on its real-time deadline (the closer the real-time deadline is to the current time, the higher the urgency score). Then, a weighted total score is calculated according to a preset weight ratio (e.g., priority score weight is 0.6, urgency score weight is 0.4), and finally, the actual forwarding order of the data frames is determined according to the order of weighted total scores from high to low. This weighted scheduling method takes into account both the business importance dimension and the time urgency dimension: on the one hand, high-priority real-time feedback data frames usually receive a higher weighted total score and are forwarded first; on the other hand, when the real-time deadline of medium or low-priority data frames is very close, their weighted total score may also exceed that of some higher-priority but less urgent data frames, thereby avoiding some data frames from being stuck in the queue for timeout due to strict priority scheduling. As can be seen, this implementation method improves the timeliness guarantee capability of data frames in multi-priority mixed scenarios by introducing a weighted scheduling strategy, while ensuring the priority processing of real-time services.

[0053] The graphics processing unit 101 is used to: process quantum measurement and control data according to preset processing logic to obtain feedback results; In practice, the persistent real-time processing program running on the graphics processing unit 101 continuously polls the pre-allocated memory area. After detecting the arrival of new data, it processes the quantum measurement and control data according to the preset processing logic to obtain feedback results.

[0054] As a feasible implementation, the preset processing logic includes at least one of readout classification, syndrome decoding, error path estimation, and feedback decision calculation; the feedback result includes at least one of Pauli frame increment, correction pulse selection, active reset instruction, and subsequent gate operation branch.

[0055] In this context, readout classification refers to the classification and identification of qubit readout signals from different control endpoints 30 by the graphics processing unit 101. Syndrome decoding refers to the decoding calculation process in which the graphics processing unit 101 infers the error type and error location based on the syndrome measurement results of multiple qubits in a quantum error correction scenario. Error path estimation refers to the process of inferring the error propagation path of qubits based on the syndrome decoding results. Feedback decision calculation refers to the calculation process in which the graphics processing unit 101 determines the subsequent quantum operation strategy based on the processing results. Pauli frame increment is the correction information that needs to be updated in the current Pauli reference frame in quantum error correction. Correction pulse selection is the decision to select the corresponding correction pulse waveform based on the decoding results. Active reset instruction is the instruction that instructs the control endpoint to perform an active reset operation on a specific qubit. Subsequent gate operation branch is the branch decision to select the subsequent quantum gate operation path based on the feedback results.

[0056] Network forwarding device 20 is further configured to: forward the feedback data frame containing the feedback result to the control endpoint; In specific implementation, if the network forwarding device 20 is a stand-alone switch, the host's network card or DPU encapsulates the feedback data frame based on the feedback result and sends the feedback data frame to the network forwarding device 20, which then forwards it to the control endpoint 30. If the network forwarding device 20 is a network card or DPU, the network forwarding device 20 encapsulates the feedback data frame based on the feedback result and then forwards it to the control endpoint 30.

[0057] As a possible implementation, the feedback data frame carries the feedback deadline of the quantum measurement and control data, and the network forwarding device is used to: parse the feedback deadline and forward the feedback data frame to the control endpoint according to the feedback deadline.

[0058] In practical implementation, after processing the quantum measurement and control data to obtain the feedback result, the graphics processing unit 101 determines the feedback deadline corresponding to the feedback result according to the timing requirements of subsequent quantum operations, which is used to indicate the latest valid time of the feedback result. The network forwarding device 20 performs downlink forwarding scheduling of the feedback data frames according to the proximity of the feedback deadline, giving priority to forwarding feedback data frames with closer feedback deadlines, so as to ensure that the feedback result can reach the control endpoint 30 within the valid time window.

[0059] It should be noted that the real-time cutoff time in the uplink data frame and the feedback cutoff time carried in the feedback data frame are time limits set for different data frames and from different perspectives, and they are not necessarily the same. The real-time cutoff time is marked by the control endpoint 30 when encapsulating the uplink data frame, and is used to indicate the valid time of the quantum measurement and control data; the feedback cutoff time is marked by the graphics processing unit 101 according to the timing requirements of subsequent quantum operations, and is used to indicate the latest valid time of the feedback result. Since the real-time cutoff time is determined when the quantum measurement and control data is generated, and the feedback cutoff time is determined only after the graphics processing unit 101 has completed processing and obtained the feedback result, their specific values ​​may differ because they are generated at different times and by different entities.

[0060] The control endpoint 30 is also used to: receive feedback data frames and perform corresponding control actions on the connected quantum processing unit based on the feedback results in the feedback data frames.

[0061] In practice, after receiving the feedback data frame, the control endpoint 30 verifies whether the endpoint identifier, round identifier, sequence number, and timestamp match the current request. If the verification passes, the control endpoint 30 parses the feedback result and executes the corresponding control action, such as loading a correction pulse, updating the Pauli reference frame, or performing an active reset.

[0062] In a preferred embodiment, the feedback data frame carries a feedback deadline; the control endpoint is further configured to: after receiving the feedback data frame, determine whether the feedback result is within a valid time window based on the arrival time and the feedback deadline in the feedback data frame; if it is within the valid time window, execute the control action on the connected quantum processing unit; if it is not within the valid time window, execute a preset local protection strategy.

[0063] In practice, after receiving a feedback data frame, the control endpoint 30 compares the current local time with the feedback deadline. If the current local time is earlier than the feedback deadline, the feedback result is within the valid time window, and the control endpoint 30 executes the corresponding real-time control action; if it times out or is invalid, the feedback result is discarded, and a preset local protection strategy is executed, such as maintaining the Pauli frame of the previous round, marking the current round of quantum operation as invalid, or skipping at least one of the following:

[0064] As can be seen, the distributed communication system provided in this embodiment establishes a direct data path between the control endpoint 30 and the graphics processing unit 101 of the host 10 through the network forwarding device 20, so that quantum measurement and control data and feedback results can be transmitted directly between the control endpoint 30 and the graphics processing unit 101 in a closed loop without being relayed by the CPU packet by packet. This achieves end-to-end low-latency, low-jitter direct communication and significantly reduces the latency uncertainty of the feedback link in real-time quantum operations.

[0065] Based on the above implementation, as a feasible implementation method, a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device 20 are pre-allocated in the memory of the graphics processing unit 101; the network forwarding device 20 is further configured to: write the frame header of the data frame into the control state area, and write the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; the graphics processing unit 101 is configured to: after acquiring the data frame, read the quantum measurement and control data from the input buffer, process it according to the corresponding preset processing logic to obtain the feedback result, and write the feedback result into the output buffer; the network forwarding device 20 is further configured to read the feedback result from the output buffer.

[0066] In this embodiment, the data frame adopts a logical structure that separates the frame header and the payload. The frame header is divided into a fixed-length metadata area, which stores control information such as frame type, endpoint identifier, experimental session identifier, round identifier, sequence number, and timestamp, facilitating the graphics processing unit to quickly classify, sort, and schedule the data frames without reading the complete payload. The payload is a variable-length data area used to carry the actual quantum measurement and control data.

[0067] In a specific implementation, the graphics processing unit 101 pre-allocates a control state area, an input buffer, and an output buffer in the device memory, and registers these memory areas as directly accessible memory areas for the network forwarding device 20 through the CUDA GPUDirect RDMA registration interface. The input and output buffers here can specifically be in a ring structure. The network forwarding device 20, based on the pre-configured queue number, memory registration information, and endpoint mapping relationship, writes the frame header of the data frame to the GPU control state area via DMA, and writes the payload (i.e., quantum measurement and control data) to the input buffer of the corresponding endpoint via DMA. After writing is complete, the write pointer of the input buffer is updated, and the graphics processing unit 101 is notified of the arrival of new data via a doorbell mechanism or by completing a queue entry. The persistent real-time processing program on the graphics processing unit 101 continuously polls the write pointer of the input buffer. Upon detecting an update to the write pointer, it reads the quantum measurement and control data from the buffer, selects the preset processing logic corresponding to the quantum measurement and control data for subsequent processing, and writes the feedback result to the output buffer after processing. The network forwarding device 20 reads the feedback result from the output buffer and sends a feedback data frame containing the feedback result to the control endpoint 30.

[0068] It should be noted that if the network forwarding device 20 is a standalone switch 201, the network interface card (NIC) or data processor writes the frame header and payload to the control state area and input buffer of the graphics processing unit 101 respectively via DMA. The buffers are managed using a producer-consumer model. The NIC or data processor acts as the producer, updating the write pointer after each write and notifying the graphics processing unit of new data arrival via a doorbell mechanism or completion queue entry. The graphics processing unit acts as the consumer, reading the data frame from the input buffer after detecting the write pointer update for subsequent processing. After processing, the graphics processing unit, as the producer, writes the feedback result to the output buffer. The NIC or data processor, as the consumer, reads the feedback result from the output buffer and encapsulates a feedback data frame based on the feedback result, sending it to the switch 201.

[0069] As can be seen, the separate writing of the frame header and the payload enables the graphics processing unit to process the metadata parsing of the frame header and the data reading of the payload in parallel in a pipeline manner, thereby improving the throughput efficiency of real-time processing.

[0070] Based on the above implementation, as a feasible implementation method, the host 10 is used to: determine the endpoint identifier of at least one control endpoint participating in the communication, and send the endpoint identifier to the network forwarding device; the network forwarding device 20 is also used to: establish a forwarding path for the data flow between the graphics processing unit and the control endpoint according to the endpoint identifier.

[0071] In practice, the quantum operating system on host 10 determines the list of endpoint identifiers for the control endpoints 30 participating in this communication based on experimental requirements and generates a unique experimental session identifier. Host 10 sends the endpoint identifier and experimental session identifier to network forwarding device 20 through the control interface. After receiving the endpoint identifier and experimental session identifier, network forwarding device 20 establishes a dedicated forwarding path for the data flow between graphics processing unit 101 and each control endpoint 30 based on this information, and configures flow table entries, forwarding port mapping relationships, real-time service dedicated network partitions, priority queue parameters, and congestion control parameters.

[0072] In a preferred embodiment, the host 10 is further configured to: acquire monitoring data from the graphics processing unit at a preset frequency; wherein the monitoring data includes at least one of latency distribution, packet loss rate, number of feedback deadline defaults, time offset, and endpoint status; and adjust the effective time window of the feedback deadline for the next round or switch the decoding algorithm based on the monitoring data.

[0073] In this embodiment, the quantum operating system running on the host 10 reads monitoring data from the control state area of ​​the graphics processing unit 101 at low frequency. Delay distribution is used to statistically analyze the end-to-end delay from the time the quantum measurement and control data is sent from the control endpoint 30 to the time the graphics processing unit 101 completes processing. Packet loss rate is used to statistically analyze the proportion of data frames lost during transmission. Feedback deadline violation count is used to statistically analyze the number of times feedback data frames arrive after the feedback deadline. Time offset is used to detect the time synchronization deviation between each control endpoint 30 and the host 10. Endpoint status is used to record the online status, working status, and error information of each control endpoint 30. The host 10 analyzes the acquired monitoring data and executes corresponding strategies, which may include adjusting the effective time window of the feedback deadline for the next round, or switching to a different level of decoding algorithm. For example, if the timeout rate exceeds a preset threshold, the effective time window of the feedback deadline for the next round can be widened, or a lightweight decoding algorithm can be switched.

[0074] It should be noted that the decoding algorithm in this embodiment can be designed in various ways depending on the inference intensity and inference time, including rule-based decoding algorithms, big data-trained decoding algorithms, etc. Rule-based decoding algorithms perform inference according to preset quantum error correction rules, resulting in fast inference speed and short inference time, suitable for rounds with tight effective time windows. Big data-trained decoding algorithms perform inference based on models trained on large-scale training data, resulting in high inference intensity but longer inference time, suitable for rounds with wider effective time windows and higher requirements for decoding accuracy. The host 10 selects or switches between various decoding algorithms based on monitoring data, ensuring that the feedback result arrives within the effective time window while also considering decoding accuracy.

[0075] In a preferred embodiment, the system further includes a time synchronization module, which provides a unified time reference for the network forwarding device, the control endpoint, and the host.

[0076] In practical implementation, the time synchronization module can employ any of the following: Precision Time Protocol (PTP), White Rabbit (WR), a 10 MHz reference clock, or a Pulse Per Second (PPS) external reference clock. The time synchronization module provides a unified time base for the network forwarding device 20, control endpoint 30, and host 10, ensuring that the timestamps, real-time cutoff times, and feedback cutoff times carried in data frames and feedback data frames are generated, forwarded, and verified within the same time domain. This avoids inaccurate cutoff time determination due to discrepancies in the local clocks of different devices. For example, if there is a discrepancy between the local clock of the network forwarding device 20 and the target control endpoint 30, the control endpoint 30 may misjudge whether the feedback result is within the valid time window. A unified time base can eliminate such discrepancies.

[0077] As can be seen, this implementation method ensures that the generation and verification of the real-time deadline and the feedback deadline are carried out in the same time domain by using a unified time reference, thus providing time synchronization guarantee for low-latency and low-jitter real-time closed-loop communication.

[0078] This embodiment discloses a data communication method, in which the execution entity is a network forwarding device. See [link to documentation]. Figure 4 A flowchart illustrating a data communication method according to an exemplary embodiment, such as... Figure 4 As shown, it includes: S101: Receive a data frame from the control endpoint; wherein the control endpoint acquires quantum measurement and control data from the connected quantum processing unit and encapsulates the data frame based on the quantum measurement and control data; In this step, the network forwarding device continuously monitors the data streams from each control endpoint through its high-speed port. When a complete data frame arrives, it is temporarily stored in the port's receive buffer. The network forwarding device identifies the source control endpoint and the corresponding experimental session based on the frame header information, providing a basis for subsequent forwarding and scheduling. It should be noted that the network forwarding device only receives and temporarily stores data at this stage; it does not parse or modify the quantum measurement and control data in the payload, thus ensuring transparent forwarding of the data plane and minimizing processing latency.

[0079] S102: Send the data frame to the host so that the quantum measurement and control data in the data frame is directly written into the memory of the graphics processing unit in the host; In this step, the network forwarding device sends the data frame temporarily stored in the receive buffer through the corresponding transmit port according to the pre-established forwarding path, ultimately reaching the graphics processing unit. Since the graphics processing unit has pre-allocated and pre-registered its input buffer during the experimental preparation phase, the network forwarding device can directly write the quantum measurement and control data in the data frame into the input buffer of the graphics processing unit through direct memory access mechanisms such as RDMA or DMA, without requiring packet-by-packet memory copying and address mapping by the CPU. This direct write path ensures that the delay between the data sent from the control endpoint and the graphics processing unit being ready to process the data includes only network transmission delay and one DMA write delay. Compared to the traditional CPU relay path, this effectively eliminates uncertain delay sources such as operating system thread scheduling, user-mode and kernel-mode switching, and layer-by-layer packet unpacking of the protocol stack, significantly controlling the end-to-end latency and jitter of the uplink data path.

[0080] As a possible implementation, the data frame carries the real-time cutoff time of the quantum measurement and control data. Sending the data frame to the host includes: parsing the real-time cutoff time and forwarding the data frame to the host according to the real-time cutoff time.

[0081] In practice, when encapsulating data frames, the control endpoint marks the quantum measurement and control data with a real-time deadline according to the timing requirements of the current quantum operation round. This real-time deadline indicates the effective time of the quantum measurement and control data. After receiving the data frame, the network forwarding device parses the real-time deadline field in the frame header and schedules the forwarding of data frames based on their proximity to the real-time deadline, prioritizing the forwarding of data frames with closer real-time deadlines. This ensures that the quantum measurement and control data can reach the host in a timely manner.

[0082] As another feasible implementation, the data frame carries the priority of the quantum measurement and control data. After receiving the data frame from the control endpoint, the method further includes: writing the data frame into the corresponding priority queue according to the priority; correspondingly, sending the data frame to the host includes: sending the data frames in the priority queue to the host in order of priority from high to low.

[0083] In practical implementation, when encapsulating data frames, the control endpoints prioritize them according to the service type carried by the quantum measurement and control data. For example, data frames carrying measurement data required for quantum error correction syndrome decoding, active reset decision data, or real-time branch decision data can be marked as high priority because the processing results of these data directly affect the timing correctness of subsequent quantum operations; data frames carrying periodic heartbeat status reporting, operation logs, or offline calibration data can be marked as ordinary priority or low priority. After receiving a data frame, the network forwarding device parses the priority field in the frame header and arranges the data frame into the corresponding priority queue according to the priority value. When dequeuing for transmission, the network forwarding device adopts a strict priority scheduling strategy: it first checks whether there are any data frames to be sent in the high-priority queue, and if so, sends them first, until the high-priority queue is empty before checking the next priority queue. Through this scheduling mechanism, data frames carrying real-time closed-loop feedback services can obtain port bandwidth and forwarding resources first at all times, avoiding the head-blocking problem caused by low-real-time traffic queuing at the front, thus ensuring low-latency transmission of critical feedback data frames in scenarios where multiple control endpoints send data in parallel.

[0084] Of course, network forwarding devices can also determine the forwarding order of data frames by considering multiple factors, and this embodiment does not impose specific limitations. For example, considering the priority and real-time deadline of the data frames, for each data frame to be forwarded, a priority score is calculated based on its priority (e.g., 3 points for high priority, 2 points for medium priority, and 1 point for low priority), and an urgency score is calculated based on its real-time deadline (the closer the real-time deadline is to the current time, the higher the urgency score). Then, a weighted total score is calculated according to a preset weight ratio (e.g., priority score weight is 0.6, urgency score weight is 0.4), and finally, the actual forwarding order of the data frames is determined according to the order of weighted total scores from high to low. This weighted scheduling method takes into account both the dimensions of business importance and time urgency: on the one hand, high-priority real-time feedback data frames usually receive a higher weighted total score and are forwarded first; on the other hand, when the real-time deadline of medium or low-priority data frames is very close, their weighted total score may exceed that of some higher-priority but less urgent data frames, thereby avoiding some data frames from being stuck in the queue for timeout due to strict priority scheduling. As can be seen, this implementation method improves the timeliness guarantee capability of data frames in multi-priority mixed scenarios by introducing a weighted scheduling strategy, while ensuring the priority processing of real-time services.

[0085] S103: Receive feedback results from the graphics processing unit; wherein the graphics processing unit processes the quantum measurement and control data according to preset processing logic to obtain feedback results; In this step, after processing the quantum measurement and control data, the graphics processing unit (GPU) writes the feedback result into its output buffer. The network forwarding device actively reads the feedback result from the buffer. Since the feedback data frame also travels from the GPU to the network forwarding device via a direct data path, this process does not rely on CPU intervention, maintaining the low-latency characteristics of the feedback link.

[0086] S104: Forward the feedback data frame containing the feedback result to the control endpoint so that the control endpoint can perform corresponding control actions on the connected quantum processing unit according to the feedback result in the feedback data frame.

[0087] In this step, the network forwarding device encapsulates a feedback data frame based on the feedback result and, according to the endpoint identifier in the frame header, looks up a pre-configured port mapping table, then sends the feedback data frame to the control endpoint through the corresponding high-speed port. Upon receiving the feedback data frame, the control endpoint first verifies its integrity, endpoint identifier matching, and sequence number continuity. If the verification passes, the control endpoint reads the feedback result within the frame and converts it into the corresponding physical layer control signal based on the type of the feedback result. For example, if the feedback result is a correction pulse selection, the control endpoint loads the specified pulse waveform from the local pulse library and applies it to the quantum processing unit through the telemetry and control link; if the feedback result is a Pauli frame increment, the local Pauli reference frame record is updated, and the logic definition of the operation is adjusted accordingly in subsequent quantum gate operations.

[0088] As a possible implementation, the feedback data frame carries the feedback deadline of the quantum measurement and control data. Forwarding the feedback data frame to the control endpoint includes: parsing the feedback deadline and forwarding the feedback data frame to the control endpoint according to the feedback deadline.

[0089] In practical implementation, after processing the quantum measurement and control data to obtain the feedback result, the graphics processing unit determines the feedback deadline corresponding to the feedback result based on the timing requirements of subsequent quantum operations. This deadline indicates the latest valid time of the feedback result. The network forwarding device performs downlink forwarding scheduling of the feedback data frames based on the proximity of the feedback deadline, prioritizing the forwarding of feedback data frames with closer feedback deadlines to ensure that the feedback result can reach the control endpoint within the valid time window.

[0090] Based on the above embodiments, as a feasible implementation method, before receiving the data frame from the control endpoint, the method further includes: establishing a forwarding path for the data stream between the graphics processing unit and the control endpoint based on the endpoint identifier of the control endpoint participating in the communication.

[0091] In this embodiment, during the experimental preparation phase, the quantum operating system running on the host determines the list of control endpoints participating in communication based on experimental requirements, generates a unique experimental session identifier, and distributes the endpoint identifier list and experimental session identifier to the network forwarding device through the management interface. Upon receiving this information, the network forwarding device creates corresponding forwarding rules in its internal flow table. The forwarding rules at least include the following mapping relationships: for data frames whose source address matches a certain control endpoint port and whose frame header carries the experimental session identifier, they are forwarded to the uplink port connected to the host; for data frames whose source address comes from the host and whose frame header's destination endpoint identifier matches a certain control port, they are forwarded to the corresponding downlink port. By pre-establishing forwarding paths, the network forwarding device does not need to perform route lookups and path calculations packet by packet in subsequent data communication phases; it can directly implement forwarding based on the configured flow table rules, thereby reducing forwarding latency.

[0092] As a feasible implementation, a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device are pre-allocated in the memory of the graphics processing unit; correspondingly, sending the data frame to the host includes: writing the frame header of the data frame into the control state area, and writing the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; correspondingly, receiving feedback results from the graphics processing unit includes: reading the feedback results from the output buffer.

[0093] In this embodiment, the data frame adopts a logical structure that separates the frame header and the payload. The frame header is divided into a fixed-length metadata area, which stores control information such as frame type, endpoint identifier, experimental session identifier, round identifier, sequence number, and timestamp, facilitating the graphics processing unit to quickly classify, sort, and schedule the data frames without reading the complete payload. The payload is a variable-length data area used to carry the actual quantum measurement and control data.

[0094] In practice, the network forwarding device, based on a pre-configured memory region mapping, writes the frame header and payload of the data frame separately to two independent memory regions of the graphics processing unit (GPU) via DMA. The frame header is written to the control status area, which is organized in a structured table format. Each entry corresponds to the complete frame header information of a data frame. The real-time processing program on the GPU can quickly obtain the metadata of newly arriving data frames by scanning or polling this area. The payload is written to the input buffer of the corresponding control endpoint. The buffer is managed using a producer-consumer model. The network forwarding device, acting as the producer, updates the write pointer after each write and notifies the GPU of new data arrival via a doorbell mechanism or a completion queue entry. The GPU, acting as the consumer, reads the quantum measurement and control data from the buffer for subsequent processing after detecting the write pointer update. After processing, the GPU, as the producer, writes the feedback result to the output buffer, and the network forwarding device, as the consumer, reads the feedback result from the output buffer. This separate writing of the frame header and payload allows the GPU to process the metadata parsing of the frame header and the data reading of the payload in parallel in a pipelined manner, improving the throughput efficiency of real-time processing.

[0095] The data communication method provided in this application is applied to a network forwarding device in a distributed communication system. The network forwarding device establishes a direct data path between the control endpoint and the host's graphics processing unit, enabling closed-loop transmission of quantum measurement and control data and feedback results directly between the control endpoint and the graphics processing unit without relying on CPU packet-by-packet relay. Specifically, the network forwarding device receives data frames encapsulated based on quantum measurement and control data from the control endpoint and sends these data frames to the host. This allows the quantum measurement and control data in the data frames to be directly written into the memory of the graphics processing unit. Since this process does not require CPU memory copying and protocol stack processing, it effectively reduces uplink data transmission latency and jitter. Subsequently, after the graphics processing unit processes the quantum measurement and control data and generates feedback results, the network forwarding device receives a feedback data frame carrying the feedback results returned by the host and forwards it to the control endpoint. This allows the control endpoint to execute control actions on the quantum processing unit, ensuring that the entire data loop is completed through the relay of the network forwarding device without relying on CPU packet-by-packet relay. Therefore, the embodiments of this application realize direct closed-loop data transmission between the control endpoint and the graphics processing unit through the network forwarding device, avoiding the delay and jitter introduced by the CPU relay, significantly reducing the end-to-end data transmission delay, and improving the timing determinism of the feedback link in real-time quantum operations.

[0096] To implement the method of the embodiments of this application, the embodiments of this application also provide an electronic device. Figure 5 This is a structural diagram of an electronic device according to an exemplary embodiment, such as... Figure 5 As shown, the electronic device includes: Communication interface 1 enables information exchange with other devices, such as network devices; Processor 2 is connected to communication interface 1 to enable information interaction with other devices and to execute the data communication methods provided by one or more of the above-mentioned technical solutions when running computer programs. The computer program is stored in memory 3.

[0097] Of course, in practical applications, the various components in an electronic device are coupled together through bus system 4. It can be understood that bus system 4 is used to achieve communication and connection between these components. In addition to the data bus, bus system 4 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 5 The general will label all buses as Bus System 4.

[0098] The memory 3 in this embodiment is used to store various types of data to support the operation of the electronic device. Examples of such data include any computer program used to operate on the electronic device.

[0099] It is understood that memory 3 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 3 described in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0100] The methods disclosed in the embodiments of this application can be applied to processor 2, or implemented by processor 2. Processor 2 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 2 or by instructions in the form of software. The processor 2 may be a general-purpose processor, DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 2 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in memory 3. Processor 2 reads the program in memory 3 and completes the steps of the aforementioned method in combination with its hardware.

[0101] When processor 2 executes the program, it implements the corresponding processes in the various methods of the embodiments of this application. For the sake of brevity, these will not be described in detail here.

[0102] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 3 that stores a computer program, which can be executed by a processor 2 to complete the steps described in the aforementioned method. The computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0103] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0104] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, network device, etc.) to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A distributed communication system, characterized in that, include: The host, network forwarding device, at least one control endpoint, and at least one quantum processing unit connected to the control endpoint, wherein the host includes a graphics processing unit; The control endpoint is used to acquire quantum measurement and control data from the connected quantum processing unit, encapsulate data frames based on the quantum measurement and control data, and send the data frames to the network forwarding device; And for receiving feedback data frames, and for performing corresponding control actions on the connected quantum processing unit based on the feedback results in the feedback data frames; The network forwarding device is used to receive the data frame from the control endpoint and send the data frame to the host so that the quantum measurement and control data in the data frame is directly written into the memory of the graphics processing unit. And for forwarding the feedback data frame containing the feedback result to the control endpoint; The graphics processing unit is used to process the quantum measurement and control data according to a preset processing logic to obtain the feedback result, and then send the feedback data frame generated based on the feedback result to the control endpoint via the network forwarding device.

2. The distributed communication system according to claim 1, characterized in that, The memory of the graphics processing unit is pre-allocated with a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device; The network forwarding device is further configured to: write the frame header of the data frame into the control state area, and write the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; The graphics processing unit is used to: after acquiring the data frame, read the quantum measurement and control data from the input buffer, process it according to the corresponding preset processing logic to obtain the feedback result, and write the feedback result into the output buffer; The network forwarding device is also used to: read the feedback result from the output buffer.

3. The distributed communication system according to claim 1, characterized in that, The feedback data frame carries a feedback deadline; The control endpoint is also used to: after receiving the feedback data frame, determine whether the feedback result is within the valid time window based on the arrival time and the feedback deadline in the feedback data frame; If the control action is performed on the connected quantum processing unit within the effective time window, the control action is performed if the control action is not performed within the effective time window.

4. The distributed communication system according to claim 3, characterized in that, The preset local protection strategy includes at least one of the following: maintaining the Pauli frame of the previous round, marking the quantum operation of the current round as invalid, or skipping the feedback operation of the current round.

5. The distributed communication system according to claim 1, characterized in that, The data frame carries the real-time cutoff time of the quantum measurement and control data. The network forwarding device is used to: parse the real-time cutoff time and forward the data frame to the host according to the real-time cutoff time.

6. The distributed communication system according to claim 1, characterized in that, The data frame carries the priority of the quantum measurement and control data; The network forwarding device is configured to: after receiving the data frame, write the data frame into the corresponding priority queue according to the priority, and send the data frames in the priority queue to the host in order of priority from high to low.

7. The distributed communication system according to any one of claims 1-6, characterized in that, The system also includes a time synchronization module, which provides a unified time reference for the network forwarding device, the control endpoint, and the host.

8. The distributed communication system according to any one of claims 1-6, characterized in that, The control endpoint is specifically used for: acquiring raw quantum measurement and control data from the connected quantum processing unit, and performing a filtering operation on the raw quantum measurement and control data to obtain quantum measurement and control data; wherein, the filtering operation includes at least one of threshold decision, effective window truncation, channel enable check, idle channel filtering, and duplicate data removal.

9. The distributed communication system according to any one of claims 1-6, characterized in that, The host is configured to: determine the endpoint identifier of at least one control endpoint participating in the communication, and send the endpoint identifier to the network forwarding device; The network forwarding device is also used to: establish a forwarding path based on the endpoint identifier as the data flow between the graphics processing unit and the control endpoint.

10. The distributed communication system according to any one of claims 1-6, characterized in that, The preset processing logic includes at least one of readout classification, syndrome decoding, error path estimation, and feedback decision calculation. The feedback result includes at least one of Pauli frame increment, correction pulse selection, active reset command, and subsequent gate operation branch.

11. The distributed communication system according to any one of claims 1-6, characterized in that, The host is also used for: Monitoring data is acquired from the graphics processing unit at a preset frequency; wherein the monitoring data includes at least one of latency distribution, packet loss rate, number of feedback deadline defaults, time offset, and endpoint status; Adjust the effective time window for the feedback deadline of the next round or switch the decoding algorithm based on the monitoring data.

12. A data communication method, characterized in that, The method, applied to a network forwarding device in any one of claims 1 to 11, comprises: Receive data frames from a control endpoint; wherein the control endpoint acquires quantum measurement and control data from a connected quantum processing unit and encapsulates the data frames based on the quantum measurement and control data; The data frame is sent to the host so that the quantum measurement and control data in the data frame is directly written into the memory of the graphics processing unit in the host. The system receives feedback results from the graphics processing unit; wherein the graphics processing unit processes the quantum measurement and control data according to a preset processing logic to obtain the feedback results. The feedback data frame containing the feedback result is forwarded to the control endpoint so that the control endpoint can perform corresponding control actions on the connected quantum processing unit based on the feedback result in the feedback data frame.

13. The data communication method according to claim 12, characterized in that, The memory of the graphics processing unit is pre-allocated with a control state area, an input buffer, and an output buffer that can be directly accessed by the network forwarding device; Accordingly, sending the data frame to the host includes: Write the frame header of the data frame into the control state area, and write the quantum measurement and control data contained in the data frame into the input buffer to notify the graphics processing unit that new data has arrived; Accordingly, receiving feedback results from the graphics processing unit includes: The feedback result is read from the output buffer.

14. The data communication method according to claim 12, characterized in that, The data frame carries the real-time cutoff time of the quantum measurement and control data. Sending the data frame to the host includes: The real-time deadline is parsed, and the data frame is forwarded to the host according to the real-time deadline.

15. The data communication method according to claim 12, characterized in that, The feedback data frame carries the feedback deadline of the quantum measurement and control data. Forwarding the feedback data frame containing the feedback result to the control endpoint includes: The feedback deadline is parsed, and a feedback data frame containing the feedback result is forwarded to the control endpoint according to the feedback deadline.

16. The data communication method according to claim 12, characterized in that, Also includes: A forwarding path is established between the graphics processing unit and the control endpoint based on the endpoint identifier of the control endpoint participating in the communication.

17. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the data communication method as described in any one of claims 12 to 16.

18. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the steps of the data communication method as described in any one of claims 12 to 16.

Images