An intelligent data scheduling method and system based on target driving

Abstract

This invention belongs to the field of communication network and data scheduling technology, and discloses a target-driven intelligent data scheduling method and system. The method includes: Step 1, when the front-end device detects a high-value target, it inserts a priority transmission marker into the data stream. After the data plane network node identifies the marker, it dynamically generates a high-priority forwarding rule locally and reports the information to the SDN controller in real time; Step 2, the SDN controller performs target matching and path management based on the reported information. When the same target moves or reappears, it performs priority retention or release operations on the old path. This invention realizes closed-loop scheduling with data plane autonomously triggering priority promotion and control plane intelligently managing priority callbacks. It eliminates the need for the SDN controller to pre-issue flow tables, reduces the control plane load, avoids bandwidth waste caused by fixed QoS policies, and ensures continuous priority transmission of video streams when targets move across devices. It is suitable for scenarios such as security monitoring.

Description

Technical Field

[0001] This invention belongs to the field of communication network and data scheduling technology, and particularly relates to a target-driven intelligent data scheduling method and system. Background Technology

[0002] In core security monitoring operations, video streams from high-value targets (such as pre-defined key personnel) need to be prioritized for smooth transmission from the front-end smart locks to the back-end video analytics servers to support real-time target feature extraction and behavior analysis. Existing SDN architectures cannot effectively address these issues, primarily due to: firstly, the centralized control architecture of SDN is prone to performance bottlenecks, with frequent flow table rule issuance by the controller exacerbating the processing load; secondly, timing differences exist in multi-layer network policy coordination, leading to delays in forwarding path scheduling; and thirdly, dynamic updates to flow table rules can easily cause rule conflicts and exhaust switch resources, ultimately reducing the forwarding efficiency of the data plane.

[0003] The existing traditional SDN security video transmission solution works by pre-distributing fixed QoS policies globally to all switches in the network through the SDN controller. This configures high-priority forwarding flow table rules for the video streams of all front-end security devices, thereby ensuring the basic transmission quality of video services. The core process is as follows: the application layer sends a QoS adjustment request to the SDN controller; the SDN controller generates flow table rules and distributes them to all relevant switches in the data plane; and the switches prioritize forwarding the video streams according to the pre-distributed flow table rules.

[0004] The current solution suffers from three main flaws: 1. Severe waste of network bandwidth resources: The emergence of high-value targets is random, and fixed high-priority QoS policies continue to occupy priority bandwidth even when there are no targets, making it impossible to dynamically schedule network resources on demand, resulting in a large amount of idle bandwidth resources. 2. Excessive load on the SDN controller, easily forming a performance bottleneck: Frequent distribution of flow table rules to all network switches is required, significantly increasing the frequency of signaling interaction between the controller and switches, exacerbating the performance bottleneck of centralized control, and increasing transmission latency and packet loss risk. 3. Poor adaptability to dynamic scenarios, prone to forwarding interruptions: It cannot dynamically adapt the forwarding path in a timely manner according to the movement of high-value targets, and there is a delay in the distribution of QoS policies. When targets move across devices and switches, it is impossible to accurately identify the same target and complete the seamless path switching, which easily leads to the invalidation of flow table rules and interruption of critical bitstream transmission. Summary of the Invention

[0005] The purpose of this invention is to provide a target-driven intelligent data scheduling method and system to solve the core problems of poor video transmission, packet loss and high latency in existing SDN networks under peak traffic conditions. It overcomes the shortcomings of traditional fixed QoS policies such as bandwidth waste, excessive load on SDN controllers and poor adaptability of forwarding paths when targets move. It enables dynamic and precise priority transmission of high-value target video streams, and can ensure that the core security code is delivered smoothly to the backend video analysis server even under network congestion or abnormal traffic attacks.

[0006] To address the aforementioned technical problems, the present invention provides a specific technical solution for a target-driven intelligent data scheduling method and system as follows: A goal-driven intelligent data scheduling method includes the following steps: Step 1: When the front-end device detects a high-value target, it writes a priority transmission flag into the custom option field of the IP packet. After the access switch and aggregation switch detect the priority transmission flag, they extract the five-tuple information of the data stream, dynamically generate high-priority forwarding rules for the five-tuple information locally, and put them into the high-priority scheduling queue. At the same time, they report the five-tuple information, priority rules, target flag, and target feature information to the SDN controller through the control plane channel. The SDN controller generates a complete forwarding path according to the global topology and stores it in the filing database. Step 2: The SDN controller performs target matching and path management based on the reported information. When the same target appears in the field of view of new front-end devices under different access switches or multiple front-end devices under the same access switch, priority retention or release operations are performed on the old path.

[0007] Furthermore, in step 2, when the same target appears on a new front-end device under a different access switch, the SDN controller performs the following operations: Step 2.1A: Receive newly reported information and perform target matching; Step 2.2A: If the same target is matched and the 5-tuple information is different, a new record is generated and the old record is deleted; Step 2.3A: Send a priority deletion command to the network nodes on the original path to release high-priority bandwidth and flow table resources.

[0008] Furthermore, in step 2, when the same target appears in the field of view of multiple front-end devices under the same access switch, the SDN controller performs the following operations: Step 2.1B: Receive new filing information and complete target matching; Step 2.2B: If the same target is matched but the 5-tuple information is different, send a target existence query command to the original ingress switch; Step 2.3B: The ingress switch compares the target features in the current bitstream with the target features in the query command, and returns a response signal indicating whether to retain or delete the target. Step 2.4B: The SDN controller determines whether to delete the high-priority rule of the original path based on the response signaling; Step 2.5B: If deletion is required, a priority deletion command is sent to each network node in the original path.

[0009] Furthermore, the front-end device is a smart door lock or a network camera, and the high-value target is a security monitoring target that requires priority transmission of its video stream.

[0010] Furthermore, the priority transmission flag is written into the Option field of the IPv4 packet, and the network node triggers the local generation of high-priority forwarding rules by detecting the non-zero value of this field.

[0011] Furthermore, the high-priority forwarding rules are generated and take effect locally on the switch based on the five-tuple information of the data flow, without the need for the SDN controller to pre-issue flow tables or QoS policies.

[0012] This invention also discloses a target-driven intelligent data scheduling system for implementing the method, comprising: Front-end devices are used to detect high-value targets and insert priority transmission markers into the data stream; Access switches and aggregation switches are used to identify priority transmission markers, dynamically generate high-priority forwarding rules locally, and report the information. The SDN controller is used to receive reported information, perform target matching and path management, and issue priority deletion commands when necessary. The system implements closed-loop scheduling that includes data plane trigger priority boosting and control plane proactive management priority callback.

[0013] Furthermore, the front-end device is also used to extract facial feature vectors and structured information of high-value targets and send them to the ingress switch cache.

[0014] Furthermore, the SDN controller is also used to maintain a global registration database, record target characteristics, five-tuple information, priority rules and complete forwarding paths, and support intelligent path switching and priority release when the target moves across devices.

[0015] The target-driven intelligent data scheduling method and system of the present invention have the following advantages: 1. Precise bandwidth scheduling on demand, completely eliminating resource waste: High-priority bandwidth resources are allocated to the corresponding video stream only when a high-value target appears. When there is no target, normal priority forwarding is restored, realizing dynamic on-demand scheduling of network resources and completely solving the bandwidth waste problem of traditional fixed QoS policies; 2. Significantly reduce the load on the SDN controller and alleviate the performance bottleneck of centralized control: The switch locally identifies the packet identifier to generate high-priority forwarding rules, and only needs to report the path registration information. There is no need for the controller to frequently send out the full flow table, which reduces the frequency of signaling interaction between the controller and the switch and effectively alleviates the performance bottleneck of centralized control. 3. Improve forwarding efficiency and reduce transmission latency: Network nodes only need to listen to specific fields of the packet to complete the forwarding, without parsing the packet content, which greatly reduces the processing overhead of the device; 4. Intelligent path management to ensure business continuity: It accurately identifies the same target through feature vector matching, realizes fine-grained path updates and seamless switching for different target movement scenarios, and proactively cleans up expired priority rules. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the system structure of the present invention, in which the black line represents data plane communication and the blue dashed line represents control plane communication. Detailed Implementation

[0017] To better understand the purpose, structure, and function of this invention, the following detailed description of a target-driven intelligent data scheduling method and system based on the accompanying drawings is provided.

[0018] It should be noted that the "high-value targets" described below specifically refer to targets in security monitoring scenarios that require special attention and priority transmission of their video streams, such as pre-defined key personnel.

[0019] like Figure 1 As shown, the present invention discloses a target-driven intelligent data scheduling system, including a smart door lock / network camera (front-end device), access switches, aggregation switches (supporting Layer 3 forwarding), an SDN controller, and a video analytics server. All access switches and aggregation switches are centrally managed by the SDN controller via the OpenFlow protocol. The video streams collected by the front-end devices are transmitted to the back-end video analytics server to complete target behavior analysis. The back-end video analytics server uniformly issues control targets and numbers.

[0020] System operation logic: Only when the front-end device detects a high-value target will the corresponding video stream be guaranteed to be forwarded with high priority; when there is no high-value target, the video stream will be forwarded with normal priority, releasing priority bandwidth resources; this realizes that the data plane triggers priority upgrades and the control plane actively controls priority callbacks.

[0021] The present invention provides a target-driven intelligent data scheduling method, comprising the following steps: Step 1: Dynamic generation and reporting of high-priority paths for high-value video streams: After the front-end smart lock detects a high-value target, the generated video / image stream will have a priority transmission flag added to the IP packet's custom options field for identification by network nodes in the data plane. Network nodes along the path (ingress switches, transit switches, and egress switches) dynamically adjust the forwarding path of this stream to high priority in the data plane, assigning high priority to specific five-tuple information so that the stream / image stream can be transmitted to the server first. Network nodes simultaneously report this priority path and facial feature vector information to the SDN controller for record-keeping, without requiring the SDN controller to pre-issue flow table rules. The detailed process is described below: Step 1.1: The front-end smart lock detects a high-value target and writes a target marker into a specific field of the IP packet in the bitstream: When the front-end smart door lock detects a high-value target (such as a person of priority attention) through its built-in AI face recognition module, blacklist / whitelist database, and video encoding unit, it writes a predefined exclusive priority transmission flag into the custom option field (Option field) of the IP packet header of the corresponding video / image stream. At the same time, it also sends the high-value target's facial feature vector and structured information (color of top and bottom clothing, age, gender, etc.) to the access switch for caching and processing.

[0022] For example, after the smart door lock 1A (IP: 192.168.1.101) in apartment 101 on the first floor of the residential building detects Zhang, a person of high interest, it writes a non-zero flag, such as the target person number 0xAA55, into the Option field of the generated video stream IPv4 packet. This packet will be sent to the backend video analysis server (IP: 192.168.100.200, port 6000) via UDP protocol through source port 5800. At the same time, the door lock extracts Zhang's 128-dimensional facial feature vector and structured information, and synchronously sends the above information to the access switch of Unit 1 directly connected to the door lock. The access switch caches this information in preparation for subsequent reporting to the SDN controller. Step 1.2: Data plane network nodes detect the target marker of the bitstream and dynamically adjust the bitstream to a high-priority forwarding path. The network nodes along the entire path, from the ingress access switch, through the aggregation switch, to the egress switch, continuously listen to the custom option field in the IP packet header. When a network node detects a non-zero Option field in the packet, it indicates that this stream requires priority transmission. It immediately extracts the unique five-tuple information (source IP address, destination IP address, source port, destination port, transport layer protocol) corresponding to this data stream, dynamically configures a high-level forwarding priority for this five-tuple information locally, and updates the traffic scheduling queue, forwarding rules, and caching policies, causing the video / image stream to enter a dedicated high-priority forwarding queue. This forwarding rule is generated and takes effect directly on the switch, without waiting for any flow table or QoS policy instructions from the SDN controller. The network nodes along the entire path synchronously adjust their priorities by identifying the same priority marker, forming an end-to-end high-priority transmission path.

[0023] For example, a video stream packet with the Option field marked 0xAA55 sent by the smart door lock 1A first enters the entry node—the Unit 1 access switch (IP address: 192.168.1.1). The CPU of the Unit 1 access switch listens and recognizes the priority transmission marker 0xAA55 of the packet, immediately extracts the five-tuple information of the stream (source IP 192.168.1.101, destination IP 192.168.100.200, source port 5800, destination port 6000, protocol UDP), and sets the highest priority of 802.1p protocol 7 for the five-tuple information locally, and puts the packet into a dedicated high-priority scheduling queue. When the packet travels to the node it passes through—the aggregation switch of the cell (IP address: 192.168.100.1)—the aggregation switch CPU also recognizes the 0xAA55 tag and synchronously sets the highest forwarding priority for this five-tuple information, completing high-priority forwarding across network segments. This ensures that the packet is ultimately transmitted to the video analytics server with low latency and no packet loss. The entire process requires no commands from the SDN controller, and even during peak-hour network congestion in the cell, this high-value video stream still receives end-to-end priority transmission.

[0024] Step 1.3: Network nodes synchronously report information to the control plane SDN controller for record-keeping. After the high-priority forwarding rules take effect, each network node immediately reports the corresponding forwarding information to the SDN controller via a dedicated control plane channel. Specifically: Ingress access switch: Reports five-tuple information, local high-priority forwarding rules, target tags, facial feature vectors, and structured information.

[0025] Passing through aggregation switches and egress switches: report five-tuple information and local high-priority forwarding rules.

[0026] First, the SDN controller receives and caches all core information reported by network nodes along the entire path. Based on the network topology it maintains, it also generates a complete high-priority forwarding path corresponding to the five-tuple information. Second, it initiates a search in the global filing database based on the target marker, facial feature vector, and structured information in the received information. If no record with the same key value is found, a new record is created; otherwise, the record is updated. Finally, this complete forwarding path information is bound and archived as a record in the global filing database along with the five-tuple information, high-priority forwarding rules, target marker, facial feature vector, and structured information.

[0027] For example, the full-path network nodes that generate high-priority forwarding rules for the video stream of the surveillance personnel Zhang are synchronously reported to the cell SDN controller with the corresponding registration information: 1) Entry node: Unit 1 access switch, after the local high priority rule takes effect, immediately reports to the SDN controller: the specific five-tuple information corresponding to the bit stream (source IP 192.168.1.101, destination IP 192.168.100.200, source port 5800, destination port 6000, protocol UDP), the 7-level highest forwarding priority rule configured on the local machine for the five-tuple information, the target tag 0xAA55 generated by door lock 1A for Zhang, 128-dimensional face feature vector and structured information; 2) Through nodes: After the local high-priority rule takes effect, the aggregation switch immediately reports to the SDN controller: the same specific five-tuple information consistent with the ingress node, and the highest forwarding priority rule of level 7 configured on the local machine for the five-tuple information; 3) After receiving all the information reported by the above nodes, the SDN controller, based on the global topology of the cell SDN network it maintains, summarizes and generates a complete end-to-end high-priority forwarding path for the code stream: Door Lock 1A → Unit 1 Access Switch 1 → Cell Aggregation Switch → Video Analysis Server. The SDN controller binds this complete forwarding path information, five-tuple information, priority rules, and Zhang's target marker, 128-dimensional facial feature vector, and structured information, storing it as a record in the global deployment target filing database. This provides the core matching basis for intelligent path updates when Zhang moves across units and devices.

[0028] Step 1 of this invention differs from the traditional SDN serial process of "application layer request → controller pre-issues flow table → switch executes forwarding". Based on the real-time requirements of SDN and security monitoring, it realizes local triggering and dynamic generation of priority forwarding of data plane, without the need for the SDN controller to pre-issue flow table or QoS policy, thereby improving the transmission efficiency and response speed of high-value code streams and reducing control plane signaling interaction overhead.

[0029] Step S2: SDN controller intelligent path management and priority release: When the target appears in the field of view of another smart lock (such as...) Figure 1 In the illustrated mobile scenario ①, a high-value target moves from lock 1A under the access switch of unit 1 to lock 2A under the access switch of unit 2. The new lock will also trigger the dynamic generation and reporting process of a high-priority path for the high-value video stream. The SDN controller, based on global registration information, determines that the key target has been detected by other locks. At this point, not only must a registration record be generated for the new stream, but priority release must also be applied to the stream corresponding to the old record. This invention designs refined processing logic for different mobile scenarios, which will be described in detail below: Scenario A: The target moves to a new door lock under a different access switch (e.g., Figure 1 Scenario ①: The target moves from door lock 1A under the access switch of unit 1 to door lock 2A under the access switch of unit 2. Step 2.1A: Operations at the SDN controller level: Step a: After receiving the high-priority path registration information reported by the access switch, the SDN controller performs target matching and record generation with the registration database: Since the registration information reported by the new switch contains high-value target tags, facial feature vectors and structured information, the SDN controller will start target retrieval in the global registration database based on the reported information.

[0030] If the target does not exist, meaning it is identified as a different target after matching the target tag / feature vector / structured information (such as another monitored person, Li), the SDN controller continues to search the record database for the existence of the same five-tuple information. If it does not exist, a new record is created; if it exists, the old record is updated. The record fully stores the forwarding path, priority rules, feature vector, five-tuple information, etc., corresponding to the new target, providing data support for the subsequent movement management and path maintenance of the target.

[0031] If an old record is identified as the same target after its similarity to the target's label / feature vector / structured information, and the 5-tuple information is also identical, then SDN takes no action and does not need to process the old record further. Here, the same target only temporarily reappears after the original door lock disappeared.

[0032] If an existing record is identified as the same target after matching its target tag / feature vector / structured information similarity, but the 5-tuple information is different, a new record needs to be generated in the global record database. For example, if door lock 2A detects a high-value target, Zhang, and writes the same target tag into a specific field of the IP packet of the bitstream, allowing this bitstream to be forwarded with high priority, but the 5-tuple information of this bitstream is different, such as the source IP being 192.168.1.201 (the source IP of door lock 1A is 192.168.1.101), similarly, the SDN controller creates a new record in the record database, and its high-priority forwarding path is "door lock 2A → unit 2 access switch → aggregation switch → video analysis server"; Step b: For cases where the five-tuple information is different but the target is the same (e.g., still Zhang), the SDN controller also needs to delete the matched old record not only at the control layer but also simultaneously send signaling to the network nodes at the data layer to delete its priority forwarding rule. For example, in mobile scenario ①, the target Zhang moves from door lock 1A to door lock 2A, and the target has disappeared from the original door lock 1A's field of vision. This means that the old record of the five-tuple information originating from door lock 1A will be deleted from the global record database by the SDN. Step 2.2A: Data-level operations: The SDN controller synchronously sends instructions to each network node along the forwarding path, requiring them to delete the priority forwarding rules for the old specific 5-tuple information to avoid wasting network resources. For example, the SDN controller sends a flow table deletion instruction (containing specific 5-tuple information) to the access switch and aggregation switch of Unit 1 on the original forwarding path, requiring the two switches to delete the high-priority forwarding rule for the 5-tuple information (source IP 192.168.1.101, destination IP 192.168.100.200, source port 5800, destination port 6000, protocol UDP). If there is no high-value target transmission for this 5-tuple information stream, the two devices will immediately clear the corresponding priority rule upon receiving the instruction, releasing the switch flow table entry and high-priority bandwidth resources occupied by the rule. This way, even if the lock 1A does not actively stop writing priority transmission flags into packets, it cannot continue to occupy high-priority transmission resources, thus solving the problem of front-end devices illegally occupying bandwidth.

[0033] Note: When a front-end smart lock detects the disappearance of a high-value target, the typical business logic simply involves stopping the writing of priority transmission flags into the video stream's IP packets, causing the video stream to automatically revert to normal priority and releasing high-priority bandwidth. However, the reality is more complex. Some front-end device manufacturers, to ensure the quality of their own stream transmission, fail to promptly disable flags and release occupied high-priority bandwidth resources, resulting in invalid network bandwidth usage. Therefore, in addition to monitoring the default settings of the IP packet's Option field to restore normal forwarding, this invention designs the above scheme to enable the SDN controller to proactively manage intelligent switching and deletion of high-priority paths on the control plane.

[0034] Scenario B: The target appears simultaneously in the field of view of multiple door locks under the same switch (e.g.) Figure 1 Scenario ②: A high-value target moves from door lock 1A to door lock 1B under the same access switch.

[0035] In this scenario, the same high-value target appears simultaneously in the monitoring field of multiple adjacent door locks / cameras. The SDN controller identifies the same high-value target in both door lock 1A and door lock 2A through target feature matching. Therefore, the above solution needs further optimization to avoid accidentally deleting high-priority bitstream paths that are still transmitting the target. Step 2.1B: The SDN controller receives the new registration information and completes feature vector matching: When the SDN controller receives a new high-priority path registration information reported by the access switch of Unit 1, in addition to generating a new registration record, it also compares it with existing records in the registration database. The process of comparing the target information collected from door lock 1B in the global registration database is similar to step a in scenario A above, and will not be described again here; Step 2.2B: SDN controller processes queries that identify the same target: For matched old records (i.e., the same target but different 5-tuple information), the SDN controller needs to send the same target existence query command to the ingress switch. The command encapsulates: target marker, facial feature vector and structured information, and the original 5-tuple information. For example, if the SDN controller caches target information for Zhang from lock 1B and finds an old record with a source IP from lock 1A based on target comparison, the SDN controller will send a target existence query command to the ingress switch in the old record, i.e., the unit 1 access switch of the original path. The command fully encapsulates: the target marker corresponding to the new registration information, facial feature vector and structured information, and the specific 5-tuple information corresponding to lock 1A in the original path. The query signal is sent to the unit 1 access switch to determine whether the target of the specific 5-tuple information in the bitstream transmission and the target encapsulated in the query command are the same target. Step 2.3B: Design of processing logic and response signaling after the ingress switch receives the query command: After receiving the query signaling, the Unit 1 access switch first matches the parsed specific 5-tuple information with all 5-tuple information transmitted by the switch in high priority to find the corresponding bitstream; then, it compares whether the target transmitted in the bitstream and the target encapsulated in the query command are the same target; finally, based on the above determination results, the switch generates a high-priority response signal (containing the original specific 5-tuple information) to retain / delete this specific 5-tuple information and sends it to the SDN controller. For example, the Unit 1 access switch stores the 5-tuple information corresponding to each bitstream in its local high-priority forwarding rule buffer. After matching the original door lock 1A specific 5-tuple information carried in the query command, the Unit 1 access switch performs the following target extraction and comparison process: ① Real-time monitoring of the video data packets currently being prioritized for forwarding by this 5-tuple information, extracting the target marker / feature vector in the packet; ② Comparing the locally extracted target marker / face feature vector with the target face feature vector sent by the SDN controller's query command, and processing according to the matching results: If the target is successfully matched, it means that the same person was simultaneously detected by both door lock 1A and door lock 2A. The switch determines that this specific five-tuple information should be retained for high-priority forwarding. If the target matching fails, such as if a high-value target person cannot be extracted from the target bitstream but is still transmitted with high priority, it indicates that the bitstream priority setting of door lock 1A is not standardized and needs to be downgraded to a normal bitstream. Generate a high-priority retention / deletion response signal (containing the original specific 5-tuple information) for this specific 5-tuple information and send it to the SDN controller; Step 2.4B: Corresponding operations at the SDN control plane: Step a: If the SDN controller receives a reservation signal, i.e., the access switch of Unit 1 reports that the original five-tuple information stream is still transmitting the same high-value target Zhang, then the SDN controller does not need to delete the high-priority rule of the original path. At this time, there are two independent high-priority forwarding paths for the target in the global reporting database at the same time, ensuring that the same target is preferentially transmitted by different door lock video streams.

[0036] Step b: If the SDN controller receives a deletion signaling message: that is, the access switch of Unit 1 reports that the high-value target no longer exists but the bitstream is still being transmitted with high priority, the SDN controller first generates and sends a high-priority deletion instruction (containing the original five-tuple information) to the network nodes on the original forwarding path to clear the high-priority rules of the original forwarding path; then deletes the old record corresponding to the original five-tuple information in the global filing database on the SDN controller. Step 2.5B: Corresponding operations for each network node at the data layer: The SDN controller sends the high-priority deletion command to the network nodes (unit 1 access switch and aggregation switch) on the original forwarding path. Each network node matches the five-tuple information in the signaling with all the five-tuple information of the high-priority transmission in the local switch, finds the priority forwarding rule of the corresponding code stream, deletes it, and forcibly releases the occupied high-priority bandwidth and flow table resources.

[0037] Through steps 1 and 2 above, this invention realizes a complete closed-loop data scheduling method driven by objectives, from autonomous triggering of the data plane to intelligent management of the control plane.

[0038] It is understood that the present invention has been described through some embodiments, and those skilled in the art will recognize that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of the present invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of the present invention.

Claims

1. A goal-driven intelligent data scheduling method, characterized in that, Includes the following steps: Step 1: When the front-end device detects a high-value target, it writes a priority transmission flag into the custom option field of the IP packet. After the access switch and aggregation switch detect the priority transmission flag, they extract the five-tuple information of the data stream, dynamically generate high-priority forwarding rules for the five-tuple information locally, and put them into the high-priority scheduling queue. At the same time, they report the five-tuple information, priority rules, target flag, and target feature information to the SDN controller through the control plane channel. The SDN controller generates a complete forwarding path according to the global topology and stores it in the filing database. Step 2: The SDN controller performs target matching and path management based on the reported information. When the same target appears in the field of view of new front-end devices under different access switches or multiple front-end devices under the same access switch, priority retention or release operations are performed on the old path.

2. The target-driven intelligent data scheduling method according to claim 1, characterized in that, In step 2, when the same target appears on a new front-end device under a different access switch, the SDN controller performs the following operations: Step 2.1A: Receive newly reported information and perform target matching; Step 2.2A: If the same target is matched and the 5-tuple information is different, a new record is generated and the old record is deleted; Step 2.3A: Send a priority deletion command to the network nodes on the original path to release high-priority bandwidth and flow table resources.

3. The target-driven intelligent data scheduling method according to claim 1, characterized in that, In step 2, when the same target appears in the field of view of multiple front-end devices under the same access switch, the SDN controller performs the following operations: Step 2.1B: Receive new filing information and complete target matching; Step 2.2B: If the same target is matched but the 5-tuple information is different, send a target existence query command to the original ingress switch; Step 2.3B: The ingress switch compares the target features in the current bitstream with the target features in the query command, and returns a response signal indicating whether to retain or delete the target. Step 2.4B: The SDN controller determines whether to delete the high-priority rule of the original path based on the response signaling; Step 2.5B: If deletion is required, a priority deletion command is sent to each network node in the original path.

4. The target-driven intelligent data scheduling method according to claim 1, characterized in that, The front-end device is a smart door lock or a network camera, and the high-value target is a security monitoring target that requires priority transmission of its video stream.

5. The target-driven intelligent data scheduling method according to claim 1, characterized in that, The priority transmission flag is written into the Option field of the IPv4 packet, and the network node triggers the local generation of high-priority forwarding rules by detecting the non-zero value of this field.

6. The target-driven intelligent data scheduling method according to claim 1, characterized in that, The high-priority forwarding rules are generated and take effect locally on the switch based on the five-tuple information of the data flow, without the need for the SDN controller to pre-issue flow tables or QoS policies.

7. A goal-driven intelligent data scheduling system for implementing the method of any one of claims 1-6, characterized in that, include: Front-end devices are used to detect high-value targets and insert priority transmission markers into the data stream; Access switches and aggregation switches are used to identify priority transmission markers, dynamically generate high-priority forwarding rules locally, and report the information. The SDN controller is used to receive reported information, perform target matching and path management, and issue priority deletion commands when necessary. The system implements closed-loop scheduling that includes data plane trigger priority boosting and control plane proactive management priority callback.

8. The target-driven intelligent data scheduling system according to claim 7, characterized in that, The front-end device is also used to extract facial feature vectors and structured information of high-value targets and send them to the ingress switch cache.

9. The target-driven intelligent data scheduling system according to claim 7, characterized in that, The SDN controller is also used to maintain a global filing database, record target characteristics, five-tuple information, priority rules and complete forwarding paths, and support intelligent path switching and priority release when the target moves across devices.

Images