External computing box data interaction acceleration method adaptive to low-delay protocol

Abstract

The invention relates to the technical field of data interaction acceleration, and provides an external computing box data interaction acceleration method adaptive to a low-delay protocol, which comprises the step of cooperatively realizing data interaction acceleration through a plurality of external computing boxes in a local network, and specifically comprises the following steps: all computing boxes are associated with edge equipment and a server; after a data packet is captured, protocol features and session time sequence features are extracted, and a broadcast channel is monitored to collect protocol observation and network states of other devices to form a neighbor set. A protocol structure with high consensus degree of multiple devices is identified, a data packet carrying the structure is simplified, only a verification key instruction is extracted, and task description is packaged and broadcasted. And the receiver generates intermediate result feedback by using a preset standardization function, and when the sender subsequently captures the same session data packet, the sender synthesizes the session data packet with the intermediate result, fills the reserved field, marks the session data packet ready and then sends the session data packet to the server. Single-device processing redundancy is reduced, data circulation efficiency is improved, and data interaction acceleration of the external computing box is realized.

Description

Technical Field

[0001] This invention relates to the field of data interaction acceleration technology, and more specifically, to a method for accelerating data interaction on an external computing box with low-latency protocol adaptation. Background Technology

[0002] The content in this section only provides background information related to this invention and may not constitute prior art.

[0003] With the rapid development of edge computing, distributed computing systems, with their multi-node collaborative working characteristics, have become an important support for improving data processing efficiency. The core advantage of edge computing lies in shortening the data transmission path and reducing latency by placing node devices close to the data source, thereby meeting various real-time data interaction needs.

[0004] In the prior art, such as Chinese Patent No. CN112596892A, a data interaction method and system for multi-node edge computing devices are disclosed. The design revolves around the data interaction of multi-node edge computing devices. It filters connectable nodes by obtaining the real-time busy / idle status of node servers, adjusts the interaction order based on historical task allocation, and adjusts the duration in combination with changes in data interaction rate, thereby improving the operating efficiency of the multi-node server system.

[0005] However, its practical application has significant shortcomings, namely, the lack of an effective acceleration mechanism designed for the data interaction process itself. In edge computing scenarios, data transmission often faces complex requirements such as high concurrency and multi-protocol compatibility. The parsing, verification, and processing of data packets all consume time. The aforementioned solutions only focus on the rational allocation of node resources and the optimization of interaction parameters, failing to accelerate data interaction from the perspectives of data processing flow and inter-node collaboration modes. Relying solely on adjusting the interaction order and duration cannot shorten the time consumption, nor does it consider improving data processing efficiency. As a result, when faced with a large number of high-frequency data packet transmissions, there is still a problem of high data interaction latency, making it difficult to meet the core requirements of real-time performance and low latency in edge computing scenarios. This limits its applicability in edge computing applications with stringent requirements for data transmission speed. Therefore, there is an urgent need for related technologies to compensate for the shortcomings of existing solutions in accelerating data interaction. Summary of the Invention

[0006] The purpose of this invention is to provide a method for accelerating data interaction with an external computing box that adapts to low-latency protocols, thereby improving the aforementioned problems. To achieve this objective, the technical solution adopted by this invention is as follows: This application provides a method for accelerating data interaction between external computing boxes with low-latency protocol adaptation. The method includes deploying multiple external computing boxes within the same local network (i.e., an edge network), with each external computing box connected to both an edge device and an edge server for edge computing. The method includes: Any external computing box captures data packets from the input data stream, parses the first preset number of fixed bytes of the application layer payload in each data packet, and extracts the payload structure signature located at the first predetermined offset and the instruction pattern encoding located at the second predetermined offset as private protocol features. Record the arrival time of each data packet, and calculate the inter-packet delay within the session based on the arrival time difference between the current data packet and the previous data packet of its session, which is used as the session correlation strength in the time series features; It continuously monitors the local network broadcast channel and receives periodic broadcast messages from other external computing boxes. The messages contain a list of load structure signatures recently observed by other external computing boxes and the frequency of occurrence of each signature, as well as network jitter observations. All broadcast messages are collected to form a neighbor observation set. The system compares the extracted payload structure signature with the signature lists in the neighbor observation set, identifies payload structure signatures that appear more frequently than a set value in lists exceeding a first threshold, and marks them as high consensus protocol structures. For data packets carrying high consensus protocol structures, the external computing box only extracts and verifies the instruction pattern encoding. Then, it encapsulates the arrival time of the current data packet, session correlation strength, and instruction pattern encoding into a computing task descriptor and sends it through the local network broadcast channel. The first external computing box, acting as the sender, listens to the local network broadcast channel to receive the computing result response message after sending the computing task descriptor; the second external computing box, acting as the receiver, after listening to the computing task descriptor, uses its pre-loaded standardized processing function library, which corresponds to the high consensus protocol structure, to generate intermediate computing results based on the parameters in the descriptor, and encapsulates the intermediate computing results in the computing result response message for broadcast. When the first external computing box captures the next data packet belonging to the same session, it combines it with the intermediate computing result obtained from the computing result response message. The combination process includes filling the application layer reserved field of the new data packet with the intermediate computing result and setting a specific result ready flag; and sending the combined data packet to the edge server.

[0007] Furthermore, the step of parsing the first preset number of fixed bytes of the application layer payload of each data packet specifically includes: The application layer start position is determined based on the network layer and transport layer header information of the data packet. A continuous block of bytes of fixed length is read from this start position. During the reading process, if the end marker of the data packet is encountered earlier than the fixed byte length, the reading stops at the actual end position and this abnormal situation is recorded.

[0008] Furthermore, the steps for calculating the in-session packet latency specifically include: For each arriving data packet, generate an arrival record containing the current timestamp; based on the session identifier carried by the data packet, find and associate the arrival record of the previous data packet of the session to which it belongs locally; calculate the difference between the timestamp of the current record and the timestamp of the previous record, and use the difference as the association strength.

[0009] Furthermore, the step of aggregating all broadcast messages to form a set of neighbor observations specifically includes: After receiving broadcast messages from other external computing boxes, the message integrity is first checked. The check includes the validity of the message identifier in the message header and the integrity of the fields in the message body. After verification, the payload structure signature list, the frequency of each signature, and network jitter observations are extracted from the message. The extracted payload structure signature list is deduplicated, and the frequency of the same payload structure signature in different messages is accumulated. At the same time, a receiving timestamp is added to each broadcast message, and broadcast messages that exceed the preset time limit are removed from the neighbor observation set to ensure the timeliness of the data in the set.

[0010] Furthermore, after performing integrity checks on the message, the process also includes: Verify whether the frequency of broadcast messages is within a reasonable range. If the frequency of messages received from the same external computing box is abnormally higher than the network's agreed value, it is determined to be potential interference or an attack. The messages sent by the external computing box are then isolated and not included in the neighbor observation set.

[0011] Furthermore, after isolating messages emitted by the external computing box and excluding them from the neighbor observation set, it also includes: Send anomaly alarm information to the edge server. The anomaly alarm information includes the alarm identifier, the location of the external computing box, and its identity information.

[0012] Furthermore, the steps of encapsulating the data into a computation task descriptor and transmitting it via a local network broadcast channel specifically include: The computation task descriptor replaces the complete packet payload and is distributed via a local network broadcast channel to at least one other external computation box with the lowest network jitter observation in the neighbor observation set.

[0013] Furthermore, utilizing its pre-loaded standardized processing function library corresponding to the high consensus protocol structure, intermediate computation results are generated based on the parameters in the descriptor, specifically including: Based on the instruction pattern encoding in the computation task descriptor, the corresponding preset processing function is searched and selected in the standardized processing function library; The arrival time and session relevance strength in the task descriptor are used as input parameters to call and execute the preset processing function; The preset processing function calculates the timing offset of the current data packet relative to the start time of its session or the previous preset reference time based on the arrival time, and adjusts the weighting factor of the timing offset based on the session correlation strength; according to the adjusted timing offset and the operation type indicated by the instruction mode encoding, it determines the corresponding output data segment from the predefined session state mapping table as an intermediate calculation result.

[0014] Furthermore, the weighting factor for adjusting the temporal offset based on the strength of session relevance is as follows: Get the average session association strength of the current session over the first few packets in history; Compare the session association strength of the current data packet to the average value; If the current value is lower than the average value, it is determined that the session stability is on the rise, and the weight factor is adjusted towards 1. If the current value is higher than the average value, it is determined that the session stability is declining, and the weight factor is adjusted towards 0.

[0015] Furthermore, after sending the synthesized data packet to the edge server, it also includes: When the edge server detects the result ready flag during parsing, it directly reads and uses the valid calculation results in the application layer reserved fields, skipping the repeated execution of the same calculation tasks implicit in the original payload of the data packet.

[0016] The beneficial effects of this invention are as follows: This invention deploys multiple external computing boxes within the same local network and associates them with edge devices and servers. It captures and parses data packets to extract protocol-related features and simultaneously calculates intra-session temporal correlation features. By monitoring network broadcast channels, it aggregates protocol feature observations and network status information from other external computing boxes, forming a neighbor observation set. Through comparison, it identifies protocol structures with high consensus among multiple devices, simplifies the processing of data packets carrying such structures, performing only the extraction and verification of key instructions, encapsulating computation task description information, and broadcasting it. The external computing box sending the task description information listens for the response, and other external computing boxes receiving this information use pre-defined standardized processing functions to generate intermediate computation results and broadcast them back. When capturing data packets from the same session subsequently, it combines these results with the acquired intermediate computation results, fills in reserved fields, marks the results as ready, and then sends them to the edge server. Through this series of collaborative identification, simplified processing, distributed computing, and result reuse methods, it reduces the processing redundancy of individual external computing boxes, improves the efficiency of data processing and transmission, and accelerates data interaction between external computing boxes. Attached Figure Description

[0017] Figure 1 A flowchart illustrating a method for accelerating data interaction with an external computing box that adapts to low-latency protocols, provided by the present invention; Figure 2 This is a system architecture diagram of the present invention; Figure 3 This is a detailed diagram of data packet parsing and feature extraction in this invention; Figure 4 This is a logic diagram for constructing the neighbor observation set and determining consensus in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] This invention proposes a method for accelerating data interaction between external computing boxes with low-latency protocol adaptation, comprising deploying multiple external computing boxes within the same local network, such as... Figure 2As shown, each external computing box connects to both the edge device and the edge server. The principle behind this is that the local area network (LAN) possesses the characteristics of short links and low transmission loss. Deploying multiple external computing boxes within the same LAN can build a distributed data processing node cluster, enabling parallel and localized data interaction. Each box connects to the edge device and the server, thus constructing a three-tiered data transmission and processing link: "edge device - external computing box - edge server," establishing a direct interaction channel between the data acquisition end and the core processing end. Technically, the LAN preferentially uses a low-latency Ethernet network. During deployment, a distributed node layout ensures coverage of the edge device's access range. The connection uses an adaptive-rate wired / wireless dual-mode interface, supporting dynamic switching based on data volume.

[0020] like Figure 1 As shown, the method includes: S101, any external computing box captures data packets from the input data stream, parses the first preset number of fixed bytes of the application layer payload in each data packet, and extracts the payload structure signature located at the first predetermined offset and the instruction pattern encoding located at the second predetermined offset as private protocol features.

[0021] Specifically, such as Figure 3 As shown, the core features of private protocols in this step are typically concentrated in a fixed area of ​​the application layer payload header. By locating feature fields at specific offsets, the protocol type can be quickly identified, providing a basis for subsequent differentiated processing. In specific implementation, packet capture uses a high-speed packet capture module based on DPDK to acquire target packets in the input data stream. The preset fixed byte length is pre-configured according to the header feature lengths of common private protocols (such as 32 bytes or 64 bytes). The first and second predetermined offsets are determined through protocol pre-analysis and stored in a local configuration table. During the parsing of the first preset number of fixed bytes of the application layer payload in each packet, the starting position of the application layer is determined based on the network layer and transport layer header information of the packet. A continuous block of bytes of fixed length is read from this starting position. The principle is that the lengths of the network layer (IP) and transport layer (TCP / UDP) headers can be determined through header field parsing. After superimposed calculation, the starting point of the application layer payload can be accurately located, ensuring the accuracy of parsing. Technically, the IP header length is obtained by parsing the IHL field and the TCP header length is obtained by parsing the TCP header data offset field. The sum of the two is the application layer start offset. During the reading process, packet end markers (such as the TCP FIN flag and packet length field) are monitored in real time. If a packet end marker is encountered earlier than the fixed byte length, the actual end position is used as the endpoint, and this exception is recorded. Since packets may be fragmented or corrupted during transmission, real-time exception recording ensures the reliability of subsequent data processing. The exception record includes information such as packet ID, capture time, and missing byte length, and is stored in a local exception log.

[0022] S102, record the arrival time of each data packet, and calculate the inter-packet delay within the session based on the arrival time difference between the current data packet and the previous data packet of its session, as the session correlation strength in the timing features.

[0023] Specifically, each external computing box is equipped with a high-precision clock synchronization module, which uses Network Time Protocol (NTP) to periodically synchronize with the edge server clock, ensuring the consistency of timestamps between different external computing boxes and within the same box. The synchronization period can be dynamically configured according to the network environment, usually set to once every 100ms to balance synchronization accuracy and network overhead. When a data packet arrives, the clock module generates a timestamp with nanosecond precision in real time. This timestamp is bound to the unique identifier of the data packet (such as a session identifier generated based on IP address, port number, and protocol type + data packet sequence number) to form a complete arrival record. The arrival record is stored in a structured data format, which, in addition to the timestamp and the unique identifier of the data packet, also includes the capture port information of the data packet for subsequent data source tracing.

[0024] Then, based on the arrival time difference between the current data packet and the previous data packet in its session, the inter-packet delay within the session is calculated as the session correlation strength in the timing features. The core principle is that data packets within the same data session (such as the interaction process corresponding to the same TCP connection or UDP data stream) usually have continuous transmission logic. The arrival time difference between the previous data packet and the current data packet (i.e., the inter-packet delay within the session) can directly reflect the data transmission rhythm of the session: the smaller the delay and the smoother the fluctuation, the stronger the session transmission stability and the higher the correlation between data packets; conversely, it indicates that the session may have network congestion and transmission interruption risks, and the data packet correlation is low. Therefore, using this delay value as the core timing feature to characterize the session correlation strength can provide a basis for optimizing the data packet processing strategy corresponding to the subsequent high consensus protocol structure.

[0025] From the perspective of the specific implementation process, the operation of calculating the inter-packet latency within a session needs to be carried out in an orderly manner based on the session management mechanism, which includes three key steps: First, an arrival record containing the current timestamp is generated for each arriving data packet. That is, for each data packet that arrives and passes the link layer verification, the session management module of the external computing box immediately triggers the arrival record generation process, encapsulates information such as nanosecond-level timestamp, unique data packet identifier, and session identifier into an arrival record, and stores it in the local session arrival record cache. This cache adopts a hash table structure, with the session identifier as the key and the data packet sequence number as the secondary index, to ensure the efficiency of subsequent queries. The storage duration of the cache can be dynamically configured according to the session timeout time, and by default, the arrival records within the last 5 minutes are retained to avoid cache overflow. Second, based on the session identifier carried by the data packet, the system searches locally for and associates the arrival record of the previous data packet belonging to its session. That is, the external computing box extracts the session identifier by parsing the network layer and transport layer header information of the data packet (for the TCP protocol, the session identifier is composed of the source IP address, destination IP address, source port number, and destination port number; for the UDP protocol, a data type field is added to the above four-tuple to distinguish different sessions). The session management module queries the local session arrival record cache based on the session identifier, locates the record set corresponding to the current session, and searches for the arrival record corresponding to the previous data packet of the current data packet in the set through the data packet sequence number. If the arrival record of the previous data packet is not found (e.g., the current data packet is the session start data packet, the arrival record of the previous data packet has expired or been lost), the current data packet is marked as the "session timing start point", and the inter-packet delay within the session is not calculated for the time being. Only its arrival time is recorded, and the delay is calculated when the next data packet of the same session is received later.

[0026] Calculate the difference between the timestamp of the current record and the timestamp of the previous record, and use this difference as the correlation strength. Third, the difference between the timestamp of the current record and the timestamp of the previous record is calculated, and this difference is used as the correlation strength. That is, after obtaining the arrival timestamps of the current data packet and the previous data packet, the time difference between the two is calculated by the high-precision time difference calculation module. This time difference is the intra-session packet delay and is directly used as the session correlation strength value in the timing features. The specific implementation is as follows: the time difference calculation uses 64-bit integer arithmetic to avoid precision loss and ensure accurate calculation of nanosecond-level delay. At the same time, a delay threshold filtering mechanism is set. If the calculated delay value exceeds the preset threshold (this threshold is preset according to the transmission performance of the local network, such as 10ms, and can be dynamically adjusted by the edge server), the current timing feature is marked as "abnormal correlation", and the abnormal information (including session identifier, current data packet identifier, delay value, and calculation time) is written to the local abnormal log to provide a basis for subsequent network quality optimization and data processing strategy adjustment.

[0027] S103 continuously monitors the local network broadcast channel, receiving periodic broadcast messages from other external computing boxes. The messages include a list of load structure signatures recently observed by other external computing boxes and the frequency of occurrence of each signature, as well as network jitter observations. All broadcast messages are collected to form a neighbor observation set.

[0028] Specifically, continuous monitoring of the local network broadcast channel is based on a preset broadcast channel port (e.g., using UDP broadcast port 255.255.255.255:8888). The external computing box has a dedicated built-in monitoring thread, employing a non-blocking I / O model to avoid consuming main thread resources and ensure the parallelism of the data processing flow. During monitoring, non-target broadcast messages are filtered in real time, and only messages carrying preset message header identifiers are received and processed, effectively reducing interference from invalid data. The payload structure signature list and the frequency of each signature in the message are core observation data for other nodes, used for subsequent multi-node consensus verification. Network jitter observations are obtained through the jitter detection module built into each external computing box, specifically the standard deviation of data packet transmission delay per unit time, reflecting the stability of the transmission link between nodes.

[0029] The subsequent step of aggregating all broadcast messages to form a neighbor observation set includes: after receiving broadcast messages from other external computing boxes, performing integrity checks on the messages. These checks include verifying the validity of the message identifier in the message header and the integrity of the fields in the message body. The principle behind this integrity check is to ensure the reliability of the received data and avoid errors in the observation data caused by message tampering or loss. Specifically, the message identifier validity check uses a preset key to decrypt and verify the identifier field in the message header; only messages that pass this verification are considered legitimately sent by a node. The message body field integrity check is implemented using the CRC-32 checksum algorithm. The checksum value of the received message body is calculated and compared with the checksum value carried at the end of the message; if they match, the field is considered intact.

[0030] After successful verification, the payload structure signature list, the frequency of each signature, and network jitter observations are extracted from the messages. The extracted payload structure signature list is deduplicated, and the frequency of the same payload structure signature appearing in different messages is accumulated. Simultaneously, a receiving timestamp is added to each broadcast message. Broadcast messages exceeding the preset timeout are removed from the neighbor observation set to ensure the timeliness of the data within the set. The deduplication process uses a hash table storage structure with the payload structure signature as the key to avoid storing the same signature repeatedly, improving the efficiency of subsequent comparisons. Frequency accumulation statistics are implemented using atomic operations to prevent statistical bias during multi-threaded concurrent processing, ensuring the accuracy of the frequency data. The preset timeout is set to 500ms by default, matching the broadcast period and can be dynamically adjusted according to the network environment. Removing expired messages avoids interference from outdated data in consensus judgments, ensuring the real-time nature of the neighbor observation set.

[0031] After verifying message integrity, the process also includes: verifying whether the broadcast message sending frequency is within a reasonable range. If the frequency of messages received from the same external computing box is abnormally higher than the network's agreed-upon value, it is determined to be potential interference or an attack, and the messages sent by the external computing box are isolated and not included in the neighbor observation set. The principle of this design is to prevent the abnormal broadcasting behavior of malicious or faulty nodes from interfering with the entire collaborative system and to ensure the data quality of the neighbor observation set. The reasonable range is preset by the edge server, with a default value of 3 times the agreed-upon broadcast period (i.e., a maximum of 3 messages from the same node are received within 600ms). The sending frequency verification is achieved by counting the number of messages from the same node received per unit time. If the range is exceeded, an isolation mechanism is triggered. Isolation adopts a message blacklist strategy, adding the abnormal node ID to a temporary blacklist.

[0032] After isolating messages from the external computing box and excluding them from the neighbor observation set, the process further includes: sending anomaly alarm information to the edge server. This alarm information includes an alarm identifier, the location of the external computing box, and its identification information. The principle behind this step is to enable timely tracing and handling of abnormal nodes, ensuring the stability of the distributed system. Alarm information is sent via an encrypted transmission channel to prevent tampering. Location information is obtained through the deployment location code of the external computing box, and the identification information is the box's unique device serial number. Upon receiving the alarm, the edge server can trigger subsequent processing procedures (such as remote diagnostics or restarting the abnormal node).

[0033] S104, such as Figure 4 As shown, the payload structure signature extracted by itself is compared with the signature lists in the neighbor observation set. Payload structure signatures that appear more frequently than a set value in the list exceeding the first threshold are identified and marked as high consensus protocol structures. For data packets carrying high consensus protocol structures, the external computing box only completes the extraction and verification of instruction pattern encoding. Then, the arrival time of the current data packet, session correlation strength, and instruction pattern encoding are encapsulated into a computing task descriptor and sent through the local network broadcast channel.

[0034] Specifically, in the process of comparing the signature lists with those in the neighbor observation set, a dual verification mechanism combining hash comparison and exact string matching is adopted. First, candidate matching signatures are quickly filtered by hash value, and then the candidate signatures are matched byte by byte to ensure the accuracy of the comparison results. At the same time, parallel comparison threads are used to improve processing efficiency and avoid blocking the data reception process during comparison operations. The first quantity threshold is set based on the total number of nodes of the external computing boxes in the local network, and is preset to 2 / 3 of the total number of nodes. It can be adjusted in real time by the edge server according to the dynamic changes of network nodes. The purpose of setting this threshold is to ensure a high consensus protocol structure. To ensure the universality of the signature and avoid the impact of abnormal observations from a few nodes on the overall judgment, the frequency of signature occurrence is set based on the average packet volume of network data transmission per unit time, for example, 5% of the average packet volume per unit time. This is used to filter out occasional non-core protocol signatures and ensure that the data interaction corresponding to the high consensus protocol structure is continuous and important. After the above dual screening, the payload structure signatures that meet the requirements are marked as high consensus protocol structures. The marking operation is specifically to add a high consensus identifier field to the corresponding signature in the local protocol feature table and associate and store the corresponding number and frequency information of consensus nodes to facilitate rapid matching of subsequent data.

[0035] For data packets carrying a high consensus protocol structure, the external computing box only extracts and verifies the instruction pattern encoding. The principle is that the high consensus protocol structure has already passed multi-node consensus verification, and its protocol format and interaction logic are stable. There is no need to repeatedly parse the complete application layer payload, simplifying the processing flow and reducing latency. In specific implementation, the extraction of the instruction pattern encoding follows the second predetermined offset preset in step S101. After extraction, verification is performed using preset encoding verification rules (such as verification based on a predefined encoding bitmask). If verification fails, the data packet is marked as abnormal and written to the local log. After successful verification, the packet proceeds to the subsequent encapsulation process. The arrival time, session association strength, and instruction mode of the current data packet are then encoded and encapsulated into a computation task descriptor. The encapsulation adopts a structured binary format, including a field identifier area, a data length area, and a data content area. The field identifier area is used to distinguish different data fields, the data length area ensures that the receiver can accurately parse the data of each field, and the data content area uses compression encoding (such as LZ4 compression algorithm) to reduce the descriptor size and reduce transmission overhead. After encapsulation, it is sent through a local network broadcast channel. During the transmission process, a transmission buffer and retransmission mechanism are enabled. If the receiver's acknowledgment response is not received within a preset timeout period (such as 5ms), a retransmission is triggered to ensure reliable transmission of the task descriptor.

[0036] The step of encapsulating a computation task descriptor and sending it via a local network broadcast channel is as follows: The computation task descriptor replaces the complete data packet payload and is distributed via the local network broadcast channel to at least one other external computing box with the lowest network jitter observation value in the neighbor observation set. The principle is that the node with the lowest network jitter observation value represents the most stable transmission link, which can ensure low latency and high reliability of task descriptor transmission. In specific implementation, the network jitter observation values ​​in the neighbor observation set are first sorted, and nodes with jitter values ​​lower than a preset jitter threshold (such as 2ms) are selected to form a candidate node set. Then, 1-2 nodes with the lowest jitter values ​​are selected as receivers. If the candidate node set is empty, the node with the lowest jitter value is selected. The distribution adopts a directional UDP sending mode, carrying the receiver device identifier to avoid irrelevant nodes receiving the data, thereby further reducing network interference.

[0037] S105, as the first external computing box that sends the computing task descriptor, listens to the local network broadcast channel to receive the computing result response message after sending the computing task descriptor; as the second external computing box that receives the computing task descriptor, after listening to the computing task descriptor, it uses its pre-loaded standardized processing function library corresponding to the high consensus protocol structure to generate intermediate computing results based on the parameters in the descriptor, and encapsulates the intermediate computing results in the computing result response message and broadcasts them.

[0038] Specifically, after the first external computing box sends out a computing task descriptor, it continuously listens to the local network broadcast channel. The principle behind this is to ensure the closed-loop execution of the computing task, ensuring that the sender can accurately retrieve the processing results of the corresponding computing task, and avoiding duplicate calculations or incomplete data processing due to result loss. This listening process reuses the broadcast channel port preset in step S103 (such as UDP broadcast port 255.255.255.255:8888), adopts a dedicated listening thread and a non-blocking I / O model, and adds a result response message filtering rule. By parsing the task association identifier in the message header (bound to the unique identifier of the sent computing task descriptor), it only processes response messages that match the task it sent, effectively avoiding interference from irrelevant messages and ensuring listening efficiency. Its beneficial effect is that it can avoid occupying the core data receiving and parsing resources of the first external computing box, ensuring that the main thread data processing flow is not blocked, and improving the accuracy of result reception.

[0039] For the receiving external computing box, its core logic is based on the stability of the high consensus protocol structure. It achieves rapid response and efficient execution of computing tasks by pre-loading a standardized processing function library. This "pre-loaded standardized processing function library corresponding to the high consensus protocol structure" works because the high consensus protocol structure, verified by multiple nodes, has a fixed format and unified interaction logic. Its corresponding processing logic can be pre-encapsulated as standardized functions and loaded into the external computing box's local storage module, avoiding the time-consuming process of real-time protocol parsing and subsequent computational logic construction. This function library is uniformly compiled and distributed to each external computing box by the edge server, using read-only storage to prevent tampering. Simultaneously, the edge server dynamically pushes updated function library packages based on network protocol structure updates, ensuring the library's timeliness. The function library internally uses an index table structure, with the payload structure signature of the high consensus protocol structure as the index key, enabling rapid mapping between the function and the corresponding processing function.

[0040] The receiver uses this function library to generate intermediate calculation results based on the parameters in the descriptor, specifically including: First, based on the instruction pattern encoding in the computation task descriptor, the corresponding preset processing function is searched and selected in the standardized processing function library. The principle is that the instruction pattern encoding directly associates with the specific operation type (such as data verification, timing analysis, instruction parsing, etc.) in the high consensus protocol structure, with different operation types corresponding to unique standardized processing logic. A bidirectional mapping table of "instruction pattern encoding - function index" is used for fast lookup. The mapping table is pre-stored in the function library header. The lookup process uses 64-bit hash operations to quickly locate the function entry address. If no matching function is found, it is marked as abnormal and an alarm is sent to the edge server, ensuring the accuracy and efficiency of processing function matching and avoiding unnecessary computational overhead. Second, the arrival time and session correlation strength in the computation task descriptor are used as input parameters to call and execute the preset processing function. The principle is that the arrival time is the core benchmark data for timing-related computations, and the session correlation strength reflects the stability of data transmission. Using both as inputs allows the computation result to adapt to the current session's transmission state, improving the relevance of the result. The parameter passing adopts a direct memory mapping method to avoid the delay caused by data copying. During the execution process, the calculation status monitoring module is enabled to detect the function execution progress and resource usage in real time, ensuring the stability of the calculation process, thereby reducing parameter passing delay, ensuring the reliability of the calculation process, and avoiding task interruption due to resource overflow.

[0041] Finally, the preset processing function calculates the timing offset of the current data packet relative to the start time of its session or the previous preset reference time based on the arrival time, and adjusts the weighting factor of the timing offset based on the session correlation strength. Based on the adjusted timing offset and the operation type indicated by the instruction mode encoding, the corresponding output data segment is determined from the predefined session state mapping table as an intermediate calculation result. The principle of timing offset calculation is to determine the relative position of the data packet in the session timeline, providing a timing reference for subsequent data synthesis and ensuring the timing consistency of multi-node processing results. Technically, the session start time or the previous preset reference time is stored in a local session state cache and quickly retrieved using the session identifier. The calculation employs a high-precision time difference calculation module, ensuring nanosecond-level accuracy. The beneficial effect of this design is to provide accurate timing basis for subsequent data synthesis. The core principle of adjusting the weighting factor of timing offset based on session correlation strength is that session correlation strength directly reflects the transmission stability of the current session. When the stability varies, the weighting of the timing offset on the final calculation result should be dynamically adapted to improve the accuracy of intermediate results. The specific implementation method is as follows: First, obtain the average session correlation strength of the first few data packets in the history of the current session. That is, first, retrieve the session correlation strength data of the first N data packets in the history of the current session (N is preset to 10 and can be dynamically adjusted by the edge server) through the local session cache, and calculate the average value using a moving average algorithm to ensure that the average value can reflect the recent session transmission status. Then, the weighting factor of the current data packet... The session correlation strength is compared with the average value. That is, the current data packet's session correlation strength is compared with the average value using a comparator. If the current value is lower than the average value, it is determined that the session stability is increasing. At this time, the weight factor of the timing offset is adjusted towards 1 (the adjustment step size is preset to 0.1 to ensure smooth adjustment) to enhance the impact of the timing offset on the calculation result. If the current value is higher than the average value, it is determined that the session stability is decreasing. The weight factor is adjusted towards 0 to reduce the influence weight of the timing offset, so that the intermediate calculation results can dynamically adapt to changes in session transmission status, improve the reliability and applicability of the results, and provide a guarantee for the accuracy of subsequent data synthesis.

[0042] S106, when the first external computing box captures the next data packet belonging to the same session, it combines it with the intermediate computing result obtained from the computing result response message. The combination process includes filling the application layer reserved field of the new data packet with the intermediate computing result and setting a specific result ready flag; and sending the combined data packet to the edge server.

[0043] Specifically, when the first external computing box captures the next data packet belonging to the same session, it first parses the network layer (IP) and transport layer (TCP / UDP) header information of the data packet through the local session management module, extracts the session identifier (a four-tuple of source IP, destination IP, source port, and destination port for TCP, and an additional data type field for UDP), and matches and verifies it with the session identifier corresponding to the task to be synthesized in the local cache. If the verification is successful, it confirms that the current data packet is a time-series continuation data of the same session. This matching process uses a hash table precise indexing mechanism. In practice, the data packet sequence number is simultaneously verified when matching the session identifier to ensure the time sequence accuracy of the synthesized object and avoid erroneous integration of cross-session or out-of-order data packets. The data packet is then combined with the intermediate calculation results obtained from the calculation result response message. The principle of the combination process is to utilize the redundancy carrying capacity of the application layer reserved fields to effectively carry the pre-calculated results without compromising the compatibility of the original protocol structure. In specific implementation, the preset offset and length of the application layer reserved fields are pre-configured by the edge server and stored in the local protocol configuration table (e.g., the preset offset is 64 bytes after the start of the application layer payload, and the length is 32 bytes). Before filling, the intermediate calculation results are verified for integrity using the CRC-32 check algorithm. After the verification is passed, the filling is completed by directly writing the binary stream to ensure the reliability of data transmission. At the same time, a specific result ready flag is set. This flag is set to the 8th binary bit in the application layer header. The setting logic is to switch this bit from the default 0 state to the 1 state. The principle is to provide the edge server with an intuitive processing status identification signal, so as to quickly determine whether the data packet carries a valid pre-calculated result without parsing the complete payload. The flag position status is synchronously recorded to the local transmission log for subsequent fault tracing. After the data is synthesized, the first external computing box sends the synthesized data packet to the edge server through the adaptive rate wired / wireless dual-mode interface. During the transmission process, a dynamic rate adaptation mechanism is enabled to adjust the transmission rate according to the real-time transmission bandwidth of the local network. At the same time, the transmission buffer and retransmission mechanism is used. If no acknowledgment response is received from the edge server within the preset 5ms timeout period, a retransmission is triggered to ensure that the data packet is delivered completely.

[0044] Furthermore, after the synthesized data packet is sent to the edge server, the edge server detects the result-ready flag during parsing and directly reads and uses the valid calculation results from the application-layer reserved fields, skipping the repeated execution of the same calculation tasks implicit in the original payload of the data packet. The principle of this step is to avoid redundant consumption of computing power through "pre-calculation result reuse". The edge server parsing module has a preset flag detection logic, which prioritizes checking the flag status and then selectively reads the reserved field data, without having to start the original payload parsing and calculation process. This significantly reduces the computing power load and data processing latency of the edge server, improves the overall data throughput of the system, and ensures the consistency of data processing results. Because the intermediate calculation results in the reserved fields have passed multi-node consensus verification and integrity verification, they can be directly used as valid data, balancing efficiency and reliability.

[0045] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for accelerating data interaction on an external computing box with low-latency protocol adaptation, characterized in that, Multiple external computing boxes are deployed within the same local network, and each external computing box is connected to an edge device and an edge server, respectively; the method includes: Any external computing box captures data packets from the input data stream, parses the first preset number of fixed bytes of the application layer payload in each data packet, and extracts the payload structure signature located at a first predetermined offset and the instruction pattern encoding located at a second predetermined offset as private protocol features; Record the arrival time of each data packet, and calculate the inter-packet delay within the session based on the arrival time difference between the current data packet and the previous data packet of its session, which is used as the session correlation strength in the time series features; The system continuously monitors the local network broadcast channel and receives periodic broadcast messages from other external computing boxes. These messages include a list of load structure signatures recently observed by other external computing boxes, the frequency of each signature, and network jitter observations. All broadcast messages are then aggregated to form a neighbor observation set. The external computing box compares the extracted payload structure signature with the signature lists in the neighbor observation set, identifies payload structure signatures that appear more frequently than a set value in lists exceeding a first threshold, and marks them as high consensus protocol structures. For data packets carrying high consensus protocol structures, the external computing box only extracts and verifies the instruction pattern encoding, and then encapsulates the arrival time of the current data packet, session correlation strength, and instruction pattern encoding into a computing task descriptor, and sends it through the local network broadcast channel. The first external computing box, acting as the sender, listens to the local network broadcast channel to receive the computing result response message after sending the computing task descriptor; the second external computing box, acting as the receiver, after listening to the computing task descriptor, uses its pre-loaded standardized processing function library corresponding to the high consensus protocol structure to generate intermediate computing results based on the parameters in the descriptor, and encapsulates the intermediate computing results in the computing result response message for broadcast. When the first external computing box captures the next data packet belonging to the same session, it combines it with the intermediate computing result obtained from the computing result response message. The combination process includes filling the application layer reserved field of the new data packet with the intermediate computing result and setting a specific result ready flag; and sending the combined data packet to the edge server.

2. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, The step of parsing the first preset number of fixed bytes of the application layer payload of each data packet specifically includes: The application layer start position is determined based on the network layer and transport layer header information of the data packet. A continuous block of bytes of fixed length is read from this start position. During the reading process, if the end marker of the data packet is encountered earlier than the fixed byte length, the reading stops at the actual end position, and this abnormal situation is recorded.

3. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, The steps for calculating the inter-packet latency within a session specifically include: For each arriving data packet, generate an arrival record containing the current timestamp; based on the session identifier carried by the data packet, find and associate the arrival record of the previous data packet of the session to which it belongs locally; calculate the difference between the timestamp of the current record and the timestamp of the previous record, and use the difference as the association strength.

4. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, The steps for compiling all broadcast messages to form a set of neighbor observations include: After receiving broadcast messages from other external computing boxes, the message integrity is first checked. The check includes the validity of the message identifier in the message header and the integrity of the fields in the message body. After verification, the payload structure signature list, the frequency of each signature, and network jitter observations are extracted from the message. The extracted payload structure signature list is deduplicated, and the frequency of the same payload structure signature in different messages is accumulated. At the same time, a receiving timestamp is added to each broadcast message, and broadcast messages that exceed the preset time limit are removed from the neighbor observation set to ensure the timeliness of the data in the set.

5. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 4, characterized in that, After performing integrity verification on the message, the process also includes: Verify whether the frequency of broadcast messages is within a reasonable range. If the frequency of messages received from the same external computing box is abnormally higher than the network's agreed value, it is determined to be potential interference or an attack. The messages sent by the external computing box are then isolated and not included in the neighbor observation set.

6. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 5, characterized in that, After isolating the messages emitted by the external computing box and excluding them from the neighbor observation set, the method further includes: An abnormal alarm message is sent to the edge server. The abnormal alarm message includes an alarm identifier, the location of the external computing box, and its identity information.

7. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, The step of encapsulating the data into a computation task descriptor and transmitting it through the local network broadcast channel specifically includes: The computation task descriptor replaces the complete packet payload and is distributed via the local network broadcast channel to at least one other external computing box with the lowest network jitter observation in the neighbor observation set.

8. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, The process of generating intermediate calculation results based on the parameters in the descriptor by utilizing a pre-loaded standardized processing function library corresponding to the high consensus protocol structure includes: Based on the instruction pattern encoding in the computation task descriptor, the corresponding preset processing function is searched and selected in the standardized processing function library; The arrival time and session association strength in the computation task descriptor are used as input parameters to call and execute the preset processing function; The preset processing function calculates the timing offset of the current data packet relative to the start time of its session or the previous preset reference time based on the arrival time, and adjusts the weight factor of the timing offset based on the session correlation strength; according to the adjusted timing offset and the operation type indicated by the instruction mode encoding, it determines the corresponding output data segment from the predefined session state mapping table as an intermediate calculation result.

9. A method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 8, characterized in that, The weighting factor for adjusting the temporal offset based on the session relevance strength is specifically as follows: Get the average session association strength of the current session over the first few packets in history; Compare the session association strength of the current data packet with the average value; If the current value is lower than the average value, it is determined that the session stability is on an improving trend, and the weight factor is adjusted towards 1; If the current value is higher than the average value, it is determined that the session stability is declining, and the weight factor is adjusted towards 0.

10. The method for accelerating data interaction of an external computing box with low-latency protocol adaptation according to claim 1, characterized in that, After sending the synthesized data packet to the edge server, the process also includes: When the edge server detects the result ready flag during parsing, it directly reads and uses the valid calculation result in the application layer reserved field, skipping the repeated execution of the same calculation task implicit in the original payload of the data packet.

Images