Integrated platform for time-sensitive computing and time-sensitive networking and industrial control systems

Abstract

This invention provides an integrated platform for time-sensitive computing (TSC) and time-sensitive networking (TSN), and an industrial control system based on this integrated platform. The integrated platform includes multiple computing units (CUCs) connected via a TSN network. The CUCs of the TSN network acquire the QoS (Quality of Service) of the time-sensitive computing tasks. The CNCs or CCCs utilize an algorithm model to determine the scheduled computing units to complete the tasks based on the QoS of the computing tasks, computing resources, and network resources. The total time of computation for each scheduled computing unit and the data transmission delay between scheduled computing units are equal to or less than the task completion time required by the QoS of the computing tasks. This invention solves the technical problem of computing units lacking high real-time performance design, achieving the goal of deterministic computing in deterministic networks. It enables computing units to participate in the industrial control system in real time, providing fundamental support for the dynamic adaptation of the industrial control system.

Description

Technical Field

[0001] This invention relates to the field of intelligent control, and in particular to an integrated platform for time-sensitive computing and time-sensitive networking, and an industrial control system. Background Technology

[0002] In existing industrial control platforms, control units typically interconnect with sensors and actuators via fieldbus or hardwired connections to periodically traverse the states of sensors and actuators in milliseconds for real-time status acquisition and control.

[0003] With the development of industrial intelligence and the Industrial Internet, an increasing number of intelligent computing units and information management units are being added to industrial systems. Existing problems include:

[0004] The computing unit lacks a high real-time design and cannot directly participate in the control execution system. Currently, it mainly focuses on collecting and calculating data to provide decision suggestions for the control unit or human-machine interaction, but it cannot participate in the system operation in real time and cannot achieve dynamic adaptive control capabilities.

[0005] There is a lack of unified network technology to support high-real-time computing and high-deterministic communication under intelligent industrial conditions. OPCUA only addresses the issue of unified modeling in the industrial field, but does not solve the communication problem for high-real-time and high-deterministic computing.

[0006] Under the conditions of industrial intelligence, there is a lack of a unified high-reliability redundancy design scheme for incremental computing systems.

[0007] Information processing systems (IT systems) and industrial control systems (OT) have a siloed structure, and there is a lack of effective technologies to achieve the integration of IT and OT. Summary of the Invention

[0008] In view of this, the embodiments of the present invention are used to solve the technical problems of the lack of high real-time design of computing units and their inability to directly participate in the control execution system. They achieve the purpose of computing units realizing deterministic computation in deterministic networks, enabling computing units to participate in industrial control systems in real time, and providing basic support for the dynamic adaptation of industrial control systems.

[0009] In a first aspect, embodiments of the present invention provide an integrated platform for time-sensitive computing and time-sensitive networking, comprising: computing units and a TSN network, each computing unit being connected to the TSN network and synchronized with the TSN network time; the CUC of the TSN network being used to acquire the QoS of time-sensitive computing tasks; wherein, the QoS includes the task completion time required by the computing task; the CNC of the TSN network being used to determine, using an algorithm model, the scheduled computing units to complete the computing task based on the QoS, computing resources, and network resources, wherein the total duration of the computing time of the scheduled computing units and the transmission delay of the data stream between the scheduled computing units is equal to or less than the task completion time required by the QoS.

[0010] Based on the above, by synchronizing the computing unit with the TSN network and by integrating the network resource scheduling and computing resource scheduling of CUC and CNC, a fusion architecture of time-sensitive computing and time-sensitive networking is realized. It also realizes the integration of real-time scheduling of computing resources for time-sensitive computing and deterministic transmission of its data stream, forming an integrated computing network platform with time determinism.

[0011] In one possible implementation of the first aspect, the QoS further includes the computing power required by the computing task and the completion time confidence, wherein the scheduled computing unit completes the computing task within the task completion time of the QoS, and the expected completion time confidence is greater than or equal to the completion time confidence of the QoS.

[0012] As described above, by including computing power and completion time confidence in the QoS of the computing task, the computing power of the scheduled computing unit is used to complete the computing task within the task completion time of the QoS, and the expected completion time confidence is greater than or equal to the completion time confidence of the QoS, so as to achieve confidence-based scheduling and avoid over-scheduling and wasting computing resources.

[0013] In one possible implementation of the first aspect, each type of computing task corresponds to an algorithm model. When the required QoS of the same type of computing task is different, the available computing units calculated by the algorithm model are different and the completion time confidence is different.

[0014] Based on the above, the algorithm model calculates different completion time confidence levels corresponding to different available computing units, which facilitates the selection of computing units that match the QoS of the computing task within the space of computing power and completion time confidence levels, thereby improving the utilization rate of computing units.

[0015] In one possible implementation of the first aspect, when the CNC or CCC determines the scheduled computing unit to complete the computing task based on the QoS, computing resources and network resources using the algorithm model, it preferentially determines the scheduled computing unit to complete the computing task from a physical node.

[0016] As described above, by prioritizing the determination of the scheduled computing unit that completes the computing task on a single physical node, the transmission latency of data streams between scheduled computing units is reduced.

[0017] In one possible implementation of the first aspect, the CNC or CCC is further configured to determine, sequentially, the type of computing unit required by the computing task and the available computing units within that type, based on the QoS, computing resources, and network resources, wherein the total duration of the computing time of the available computing units and the transmission delay of the data stream between the computing units is equal to or less than the task completion time required by the QoS. In some embodiments, when multiple groups of available computing units exist and at least one group of scheduled computing units originates from the same physical node, the available computing units within the same group that originate from the same physical node are retained; otherwise, the multiple groups of available computing units are retained; the available computing unit whose total duration is closest to the task completion time required by the QoS of the computing task is selected as the final scheduled computing unit, thereby achieving priority scheduling from one physical node.

[0018] As described above, by first determining the type of computing unit required for the computing task and the available computing units within that type, the type and computing power of the available computing units can be matched with the computing task.

[0019] In one possible implementation of the first aspect, the CNC or CCC is further configured to, when multiple groups of scheduled computing units meet the QoS requirements, select the group of scheduled computing units corresponding to the point with the smallest distance from the QoS's computing power and completion time confidence level from the space composed of the computing power of the scheduled computing units and the expected completion time confidence level. In some embodiments, when at least one group of scheduled computing units originates from the same physical node, before selecting the final scheduled computing unit, the groups of scheduled computing units originating from the same physical node within the same group are retained; otherwise, the multiple groups of scheduled computing units are retained, thereby achieving priority scheduling from one physical node.

[0020] Based on the above, by selecting a group of scheduled computing units corresponding to the points with the smallest distance from the computing power and expected completion time confidence of the computing task QoS in the space composed of computing power and completion probability, the most suitable computing units for scheduling the computing task are obtained.

[0021] In one possible implementation of the first aspect, the CNC or CCC is further configured to set a priority for the computing task based on the QoS of the computing task, with the computing task having the highest priority being the computing unit that schedules the computing task.

[0022] As shown above, when multiple computing tasks arrive simultaneously, the priority is set according to the QoS of the computing tasks, and the computing units are scheduled for the higher priority computing tasks first, so that important computing tasks are scheduled in advance.

[0023] In one possible implementation of the first aspect, when there are scheduling configuration parameters for data flows between scheduled computing units in the TSN network, the CNC is further used to generate computing configuration parameters and distribute them to the scheduled computing units; the computing configuration parameters include the priority level and transmission delay of the data flow of the computing task.

[0024] As described above, by configuring the priority level and transmission delay of the data stream between the scheduled computing units through calculation configuration parameters, the deterministic transmission of the data stream is achieved in the TSN network according to the existing scheduling strategy, so that the total time for the computing unit to complete the calculation and data stream transmission meets the QoS requirements of the computing task.

[0025] In one possible implementation of the first aspect, when there are no scheduling configuration parameters for the data flow between scheduled computing units in the TSN network, the CNC is further configured to generate scheduling configuration parameters for the data flow and distribute them to the TSN switches between the scheduled computing units; the scheduling configuration parameters include gating parameters and traffic shaping parameters; the TSN switches perform deterministic scheduling of the data flow according to the scheduling configuration parameters, so that the data flow completes transmission within its corresponding deterministic duration.

[0026] As described above, by configuring the TSN switches between the scheduled computing units through scheduling configuration parameters, the data flow between the scheduled computing units achieves deterministic transmission of the data flow according to the new scheduling strategy, thereby ensuring that the total time for the computing unit to complete the computation and data flow transmission meets the QoS requirements of the computing task.

[0027] In one possible implementation of the first aspect, the process of completing the computing task includes: deterministic transmission of a data stream between a scheduled computing unit completing the corresponding computation and its corresponding network resources completing the task, wherein the data stream includes a data stream between computing units that have a logical relationship in computation and are located on different physical nodes. In some embodiments, when multiple computing tasks are scheduled simultaneously, and the multiple tasks have a logical relationship in computation and the scheduled computing units are located on different physical nodes, the data stream also includes a corresponding data stream between these scheduled computing units.

[0028] In summary, by integrating the scheduled computing units to complete the corresponding calculations and their corresponding network resources to complete the deterministic transmission of the data flow for the computing tasks, the computing units and the TSN network are integrated to meet the QoS requirements of the computing tasks.

[0029] In one possible implementation of the first aspect, the scheduled computing unit further sets up a cache queue according to the computing scheduling parameters, the size of which is set according to the data volume of the data stream and the bandwidth of the network resources between the scheduled computing units; when the scheduled computing unit performs computing, it caches the data of the data stream in the cache queue and timestamps the data for transmission to the TSN network.

[0030] As described above, deterministic transmission can be achieved within the bandwidth of the TSN network by setting up a cache queue in the computing unit and timestamping the cached data.

[0031] In one possible implementation of the first aspect, when the computing task includes multiple serially computed subtasks, when the computing unit running the current subtask detects a computing anomaly, it rolls back to the computing unit corresponding to the previous subtask and recalculates according to the timestamp of the subtask.

[0032] As described above, the rollback calculation of subtasks with abnormal calculations is achieved by synchronizing the time of the computing unit and using the timestamps of the corresponding subtasks.

[0033] In one possible implementation of the first aspect, when a subtask of the computation task is completed, its corresponding computation unit is released asynchronously. The asynchronous nature means that the computation unit that completes the subtask is released upon completion of the subtask, without waiting for the entire computation process to finish.

[0034] As described above, by asynchronously releasing the computing units that have completed the subtasks, the utilization rate of the computing units is improved.

[0035] In one possible implementation of the first aspect, both the events of the computing task and the events of the computing unit are timestamped at the time of generation.

[0036] As described above, by using the timestamps of events in computing tasks and events in computing units, the computing tasks and the management of computing units are integrated with the TSN network, enabling the deterministic management of computing tasks and computing units.

[0037] In one possible implementation of the first aspect, each computing unit includes a minimum scheduling unit of at least one computing power type on a physical node, and the CNC is further configured to allocate a corresponding number of computing cores of the corresponding computing power type to each minimum scheduling unit.

[0038] As described above, by encapsulating the smallest scheduling unit, the scheduling of computing units becomes flexible and matches the computing business requirements.

[0039] In one possible implementation of the first aspect, according to the computing type of the computing task, a plurality of computing units and the network resources between the plurality of computing units form a resource cluster, each resource cluster being used to complete a computing task of a certain type; the network resources between the scheduled computing unit and therebetween are preferentially scheduled from the resource cluster corresponding to the computing type of the computing task.

[0040] As described above, by prioritizing scheduling within the resource cluster corresponding to the computing type of the computing task, the scheduling computing unit and the network resources between it are matched with the computing task. This not only improves the time determinism of the computing task, but also improves the utilization rate of the computing unit and the network resources between it.

[0041] In one possible implementation of the first aspect, each resource cluster is configured with a backup of computing unit resources and network resources between computing units in the backup computing unit resources.

[0042] As described above, the backup computing unit not only enables highly reliable backup of computing resources for computing tasks, but also enables highly reliable transmission of data streams between computing resources, thereby achieving highly reliable backup of resources to complete computing tasks.

[0043] In one possible implementation of the first aspect, the CNC is further configured to schedule a set of redundant computing units for the task, wherein the redundant computing units and the network resources therebetween are preferentially scheduled from the resource cluster corresponding to the computing type of the computing task; the CNC is further configured to replace the scheduled computing units or the network resources therebetween with the redundant computing units and the network resources therebetween when the scheduled computing units or the network resources therebetween are unavailable.

[0044] As described above, by using redundant computing units, not only is highly reliable backup of computing resources for computing tasks achieved, but also highly reliable transmission of data streams between computing resources is achieved, thereby also achieving highly reliable backup of resources for completing computing tasks.

[0045] In one possible implementation of the first aspect, the algorithm model is a state machine pre-defined according to the type of various computational tasks and / or a pre-trained artificial intelligence model. In some embodiments, the algorithm model is obtained by training the artificial intelligence model with sample data obtained from the state machine.

[0046] Therefore, the accuracy of algorithm model scheduling can be improved by using state machines and / or artificial intelligence models.

[0047] Secondly, embodiments of the present invention provide an industrial control system, comprising: an integrated platform, a control unit, and a human-machine unit as described in any embodiment of the first aspect; the control unit being connected to the TSN network of the integrated platform; the control unit being used to collect data from sensors in real time and / or send control commands to actuators in real time via the TSN network; the human-machine unit being used for human-machine interaction with the industrial control platform; and the control unit or the human-machine unit being further used to make decisions based on the calculation results of the computing unit of the integrated platform.

[0048] In summary, by integrating time-sensitive computing and other information processing systems (IT systems) with industrial control systems (OT) through an integrated platform and using a TSN network to determine data transmission, the convergence of IT and OT is achieved, forming an integrated industrial control platform for real-time sensing, computing, decision-making, and control. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of a first embodiment of the time-sensitive computing and time-sensitive networking integrated platform of the present invention;

[0050] Figure 2 This is a schematic diagram of a variant of an embodiment of the time-sensitive computing and time-sensitive networking integrated platform of the present invention;

[0051] Figure 3 This is a schematic diagram of the structure of a second embodiment of the time-sensitive computing and time-sensitive networking integrated platform of the present invention on an edge server;

[0052] Figure 4 This is a schematic diagram illustrating scheduling from a resource cluster, according to a second embodiment of the time-sensitive computing and time-sensitive networking integrated platform of the present invention.

[0053] Figure 5 This is a schematic diagram of an embodiment of an industrial control system according to the present invention. Detailed Implementation

[0054] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0055] In the following description, the terms “first, second, third, etc.” or module A, module B, module C, etc. are used only to distinguish similar objects or different embodiments and do not represent a specific ordering of objects. It is understood that a specific order or sequence may be interchanged where permitted so that the embodiments of the invention described herein can be implemented in an order other than that illustrated or described herein.

[0056] In the following description, the labels of the steps, such as S110, S120, etc., do not necessarily mean that the steps will be executed in this way. The order of the steps can be interchanged or executed simultaneously if permitted.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0058] Let's combine the following... Figures 1 to 4 An embodiment of an integrated platform for time-sensitive computing and time-sensitive networking is introduced.

[0059] An embodiment of an integrated time-sensitive computing and time-sensitive networking platform includes: computing units and a TSN network, each computing unit being connected to the TSN network and synchronized with the TSN network time. The TSN network's CUC (Centralized User Configuration) is also used to obtain the QoS of time-sensitive computing tasks.

[0060] The scheduled computing unit can be determined by using the CNC (Centralized Network Configuration) in the TSN network, or it can be determined by setting the CCC (Centralized Computing Configuration) of a network element separately. Figure 1 The illustrated embodiment uses CNC to determine the scheduled computing unit as an example. The CNC is used to determine the scheduled computing unit to complete the computing task based on the QoS of the computing task, computing resources, and network resources using an algorithm model. The total duration of the computing time of the scheduled computing unit and the transmission delay of the data stream between the scheduled computing units is equal to or less than the task completion time required by the QoS of the computing task.

[0061] Figure 1 The structure of an embodiment of a time-sensitive computing and time-sensitive networking integrated platform is shown, including: several computing units 110 and a TSN network, each computing unit 110 being connected to the TSN network and synchronized with the TSN network time.

[0062] The TSN network includes several TSN network switches 20, TSN network CUC30, and TSN network CNC40. CUC30 and CNC40 perform functions such as configuring and distributing existing policies for the TSN network's CUC and CNC, and also integrate time-sensitive computing (TSC) related functions. Through the synchronization of the computing unit 110 with the TSN network and the integration of network resource scheduling and computing resource scheduling functions in CUC30 and CNC40, an integrated platform of TSC and TSN is achieved.

[0063] Each computing unit 110 is connected to a corresponding switch 20 to connect to the TSN network. The computing units 110 constitute computing resources, and each transmission path or TSN time slice of the TSN network constitutes network resources.

[0064] For example, Figure 1 The system includes four computing units 110, but in real-world scenarios, it can include any number of computing units 110; the TSN network includes four TSN network switches 20, but in real-world scenarios, it can include any number of switches 20. The computing units 110 are located at the corresponding physical nodes ( Figure 1 On (not shown), each computing unit 110 is connected to the corresponding switch 20 through its physical node, and communication between computing units 110 on the same physical node does not go through the TSN network.

[0065] CUC30 is used to obtain the QoS of time-sensitive computing tasks and send it to CNC40. The QoS of computing tasks includes at least the task completion time required by the computing task.

[0066] In some embodiments, the CUC30 directly obtains the QoS of the computing task from the control node of the TSN network or a non-real-time system that issues tasks to the real-time node. In other embodiments, the CUC30 directly obtains the input conditions of the computing task from other nodes or other systems in the TSN network and obtains the QoS of the computing task based on the input conditions.

[0067] It should be noted that the task completion time for calculating the QoS of the calculation task does not include the transmission delay of the initial input data and the final result of the calculation task. These two delays have been determined according to the existing scheduling policy of the TSN network.

[0068] In some embodiments, the QoS of a computing task also includes the computing power required by the computing task and the confidence level of the required completion time. The attributes of the computing power include combinations of computing power types and the computing capabilities of each type; for example, 1CPU+3GPU represents the computing power of a combination of one CPU and three GPUs.

[0069] In some embodiments, the CUC30 acquires the QoS of multiple computing tasks simultaneously, and the relationship between these computing tasks is not limited. For example, these tasks may be computing tasks with different priorities or computing tasks with the same priority, computing tasks that are completed serially or in parallel in time, or computing tasks that have no order of completion time.

[0070] In some embodiments, CUC30 is also used to collect information on available computing resources in the TSN network, including idle computing units. Specifically, when a computing unit 110 is idle, it reports to CUC30, which then collects the information on available computing resources in the TSN network and sends it to CNC40 for scheduling.

[0071] The CNC40 uses an algorithm model to determine the scheduled computing units to complete the computing task based on the QoS of the computing task, computing resources, and network resources. The total duration of the computing time of the scheduled computing unit and the transmission delay of the data stream between the scheduled computing units is equal to or less than the task completion time required by the QoS of the computing task.

[0072] In this context, the network resources in the TSN network between scheduled computing units constitute the data streams transmitted between the scheduled computing units. In some embodiments, scheduled network resources are further determined from the network resources in the TSN network between the scheduled computing units. These scheduled network resources employ a different scheduling strategy than those in the current TSN network, and this scheduling strategy determines the transmission delay of the data streams between the scheduled computing units.

[0073] In some embodiments, the scheduled network resources are reserved TSN transmission paths; in other embodiments, the scheduled network resources are TSN time slices determined according to a scheduling policy. The scheduling policy may include determination based on the priority of data flows between scheduled computing units and required latency.

[0074] In some embodiments, when the computation task QoS also includes the computational power required by the computation task and the required completion time confidence, the scheduled computation unit completes the computation task within the task completion time of the computation task QoS, and the expected completion time confidence is greater than or equal to the completion time confidence required in the computation task QoS.

[0075] In some embodiments, each type of computational task corresponds to a specific algorithm model. When the required QoS for the same type of computational task differs, the available computing units calculated by the algorithm model will differ, and the completion time confidence levels will also differ. For example, for a certain computational task, the computing power of C1+G1 reaches 90% time confidence, while the computing power of C2+G4 reaches 98% time confidence. Here, C1 and C2 represent the general computing power of two different CPUs, and G1 and C2 represent the accelerated computing power of two different GPUs. The types of computational tasks are categorized according to the on-site computational scenario, and for example, include: image recognition, image segmentation, speech recognition, semantic recognition, data clustering, data fitting, and logical computation.

[0076] In some embodiments, during scheduling, computing units are first selected based on the computing power of each computing unit 110, and then the total time to complete the computing task is determined based on the network resources between the selected computing units to see if the requirements are met. When selecting network resources between computing units, an existing scheduling strategy is first selected. If the existing scheduling strategy does not meet the requirements, a new scheduling strategy with lower deterministic latency is set to satisfy the data transmission between computing units.

[0077] In some embodiments, the algorithm model is a state machine pre-set according to the type of various computing tasks and / or a pre-trained artificial intelligence model. In other embodiments, the algorithm model for each computing type is a state machine first set according to human experience for various computing task types. After working for a period of time, the QoS, computing power and time to complete each computing task of that computing type are collected. The algorithm model for that computing type is then trained through an artificial intelligence model to improve the accuracy of the algorithm model for that computing type.

[0078] In some embodiments, when the CNC40 determines the scheduled computing unit to complete the computing task based on the computing task QoS, computing resources, and network resources using the algorithm model, it preferentially determines the scheduled computing unit to complete the computing task from a physical node.

[0079] In some embodiments, based on the QoS of the computing task, computing resources, and network resources, the type of computing unit required by the computing task and the available computing units within that type are determined. The total duration of the computing time of the available computing units and the transmission delay of the data stream between computing units is equal to or less than the task completion time required by the QoS of the computing task. In some embodiments, when multiple groups of available computing units exist, if at least one group of scheduled computing units comes from the same physical node, the available computing units within the same group whose computing units come from the same physical node are retained; otherwise, the multiple groups of available computing units are retained, thus prioritizing scheduling one physical node. The available computing unit whose total duration is closest to the task completion time required by the QoS of the computing task, or the available computing unit whose total duration is less than the task completion time required by the QoS of the computing task and has the lowest computing power, is selected as the final scheduled computing unit, thereby improving the utilization rate of computing units and avoiding over-scheduling of computing units.

[0080] In some embodiments, when multiple groups of scheduled computing units meet the QoS requirements of the computing task, the group of scheduled computing units corresponding to the point with the smallest distance from the computing power and completion time confidence of the computing task QoS requirements is selected as the final scheduled computing unit from the space composed of the computing power of the scheduled computing units and the expected completion time confidence. This maximizes the resource utilization of the computing units and avoids over-scheduling. In some embodiments, when at least one group of scheduled computing units comes from the same physical node, before selecting the final scheduled computing unit, the groups of scheduled computing units from the same physical node within the same group are retained; otherwise, the multiple groups of scheduled computing units are retained, thereby prioritizing scheduling from one physical node.

[0081] In some embodiments, the CNC40 is further configured to set priorities for computing tasks based on their QoS, prioritizing computing units for higher-priority computing tasks, thereby achieving priority scheduling when multiple computing tasks arrive simultaneously. In some embodiments, network resources among the scheduling computing units for higher-priority computing tasks are allocated through a high-priority scheduling strategy, for example, a path-reserving or resource preemption scheduling strategy.

[0082] In some embodiments, CNC40 is further configured to generate the priority level and transmission delay of the data stream of the computing task according to the transmission delay requirements of the computing task. The data stream of the computing task is the data stream between the scheduled computing units. If scheduling configuration parameters for the data stream of the computing task exist in the TSN network (i.e., the TSN network currently has a scheduling policy that satisfies the priority level and transmission delay of the data stream of the computing task), CNC40 is further configured to generate computing configuration parameters and distribute them to the scheduled computing units. This distribution method is the same as the method by which CNC40 distributes scheduling configuration parameters to TSN switch 20, and this method is reused. The computing configuration parameters include at least the priority level and transmission delay of the data stream of the computing task. The scheduled computing unit uses the computing configuration parameters to configure the priority level and transmission delay requirements of its emitted data stream. TSN switch 20 utilizes the existing scheduling policy to complete the deterministic transmission of the data stream of the computing task according to the priority level and transmission delay requirements of the data stream of the computing task, thereby enabling the scheduled computing unit and the network resources between them to complete the computing task according to the QoS requirements of the computing task.

[0083] In some embodiments, when there are no scheduling configuration parameters for the data flow of computing tasks in the TSN network (i.e., there is currently no scheduling strategy in the TSN network that meets the priority level and transmission delay requirements of the data flow of computing tasks), the CNC40 is further configured to generate scheduling configuration parameters for the data flow of computing tasks based on the priority level and transmission delay requirements of the data flow of computing tasks, and distribute them to the corresponding TSN switches 20 between the scheduled computing units. When there are multiple TSN switches 20 between the scheduled computing units, the parameters need to be distributed to each TSN switch 20. The method of distribution is the same as the method by which the CNC40 distributes existing scheduling configuration parameters to the TSN switches 20, and this method is reused. The scheduling configuration parameters include at least gating parameters and traffic shaping parameters. The TSN switches 20 perform deterministic scheduling of the data flow between the scheduled computing units based on the scheduling configuration parameters, so that the data flow completes transmission within its corresponding deterministic duration, thereby enabling the scheduled computing units and the network resources between them to complete the computing tasks according to the QoS requirements of the computing tasks. In some embodiments, the scheduling configuration parameters for the data stream generated by the CNC40 for computing tasks also take into account the amount of data in the computing tasks. For example, a larger TSN time slice is required in the gating parameters for data streams with larger amounts of data (such as GPU-accelerated computing).

[0084] The scheduled computing units are used to complete computing tasks. The process of completing a computing task includes: deterministic transmission of data streams between the scheduled computing units to complete the corresponding computations and the network resources between them to complete the computing tasks. These data streams include data streams between computing units that have logical computational relationships and are located on different physical nodes. When a computing task is divided into multiple subtasks and the subtasks are assigned to scheduled computing units on different physical nodes, the data stream includes data streams between these subtasks. In some embodiments, when multiple computing tasks are scheduled simultaneously, and these multiple tasks have logical computational relationships and the scheduled computing units are located on different physical nodes, the data stream also includes corresponding data streams between these scheduled computing units.

[0085] In some embodiments, the scheduled computing unit also sets up a buffer queue according to the computing scheduling parameters issued by the CNC. The bandwidth of the network resources between the current scheduled computing unit and the next scheduled computing unit is not infinite. The data sent by the current scheduled computing unit to the next scheduled computing unit is sent with deterministic transmission delay, but it is not sent as soon as it is generated. It needs to be buffered in the buffer queue. The size of the buffer queue is set according to the data volume of the corresponding data stream and the bandwidth of the network resources between the scheduled computing units. When the scheduled computing unit generates data that has not yet been sent during the calculation, the data that has not yet been sent is buffered in the buffer queue, and a timestamp is added to the data, along with the corresponding priority level and transmission delay requirements. When the data is sent to the TSN network, the TSN switch 20 uses the timestamp and the corresponding priority level and transmission delay requirements to complete the deterministic transmission according to the corresponding scheduling strategy.

[0086] In some embodiments, when a computation task includes multiple serially computed subtasks, if a computational unit running the current subtask detects a computational anomaly, it rolls back to the computational unit corresponding to the previous subtask and recalculates based on the timestamp of the corresponding subtask. Because all computational units 110 are time-synchronized, the computational state at that time can be accurately rolled back based on the timestamp of the data stream data of the computational task.

[0087] In some embodiments, when a subtask of a computing task is completed, its corresponding computing unit is released asynchronously so that it can be used by other computing tasks without waiting for the entire computing task to be completed.

[0088] In some embodiments, events of both the computing task and the computing unit 110 are timestamped upon generation. Events of the computing task are timestamped when generated by other tasks or other physical nodes. Events of the computing unit 110 include various status events of the computing unit 110, such as: availability, task received, computation started, settlement completed, task terminated, rollback, and data stream sent. These timestamps are used for collaboration between the computing unit 110 and the TSN network. By using the timestamps of events of the computing task and the computing unit, the computation of the computing task and the management of the computing unit are integrated with the TSN network, enabling deterministic management of the computing tasks and computing units.

[0089] Figure 2 A variant structure of an embodiment of a time-sensitive computing and time-sensitive networking integrated platform is shown, in which CNC40 is divided into CNC41 and CCC42. CNC41 completes the configuration function of the traditional TSN network data flow scheduling strategy through a time-based network resource scheduling table, and CCC42 completes the computing task-related functions described in this application through a time-based computing resource scheduling table.

[0090] In summary, in Embodiment 1 of an integrated platform for time-sensitive computing and time-sensitive networking, the Computational Controller (CUC) is used to acquire the QoS of time-sensitive computing tasks. The Computational Controller (CNC) is also used to determine, using an algorithm model based on the QoS of the computing task, computing resources, and network resources, the scheduled computing units and scheduled network resources for completing the computing task. The total duration of the computing time and data stream transmission delay of the scheduled computing units is equal to or less than the task completion time required by the QoS of the computing task. This integrated platform, through the synchronization of computing units with the TSN network and the fusion of network resource scheduling and computing resource scheduling by CUC and CNC, achieves an integrated platform for time-sensitive computing and time-sensitive networking. It integrates real-time scheduling of computing resources for time-sensitive computing with deterministic transmission of its data stream, forming an integrated computing and network platform with time determinism, thus improving the real-time performance of computing decisions.

[0091] Let's combine the following... Figure 3 and Figure 4 This paper introduces a second embodiment of an integrated platform for time-sensitive computing and time-sensitive networking.

[0092] The computing unit of the integrated time-sensitive computing and time-sensitive networking platform in Embodiment 2 is implemented on an edge server. It has all the advantages of the integrated time-sensitive computing and time-sensitive networking platform in Embodiment 1. Each edge server can be elastically expanded to join the integrated time-sensitive computing and time-sensitive networking platform, realizing online expansion of time-sensitive computing.

[0093] Figure 3 The illustration shows an embodiment of a time-sensitive computing and time-sensitive networking integrated platform architecture on an edge server, which includes several elastically scalable edge servers 100. Each edge server 100 is connected to a TSN network.

[0094] Each edge server 100 includes several computing units 110 at the application layer (three computing units are shown in the example, but more may be included in practice). Each computing unit 110 runs on the industrial operating system 210 of the edge server 100. The industrial operating system 210 provides a runtime environment for each computing unit 110, which runs on physical hardware consisting of several computing cores and TSN ports 410, etc. For example, the computing core includes at least one of the following computing power types: CPU 310, GPU 320, and DPU 330. In some embodiments, the computing core also includes at least one of the following computing power types: FPGA, TPU, and NPU.

[0095] The computing unit 110 is connected to the TSN network via TSN port 410, forming the integrated platform described in Embodiment 2 of the integrated platform for time-sensitive computing and time-sensitive networking. The CNC41, CUC30, and CCC42 of the TSN network can be established on other service servers or on one of the edge servers 100.

[0096] In some embodiments, each computing unit 110 includes at least one minimum scheduling unit of computing power type, and CCC42 is further configured to allocate a corresponding number of computing cores of the corresponding computing power type to each minimum scheduling unit. The computing cores of the minimum scheduling unit for each computing power type are encapsulated according to computing service requirements. Through the encapsulated minimum scheduling units, the scheduling of computing units becomes flexible and matches computing service requirements.

[0097] In some embodiments, the CCC42 uses the industrial operating system 210 to encapsulate the minimum scheduling unit for each computing unit 110 using real-time partitioning or real-time containers.

[0098] In some embodiments, an edge server 100 is a physical node, and the computing units 110 of each edge server 100 are connected through a TSN network to achieve deterministic transmission between them. The computing units 110 of an edge server 100 communicate with each other in real time through an industrial operating system 210, and the transmission delay between them is negligible relative to the computing time of the computing unit 110.

[0099] In some embodiments, a plurality of computing units 110 and the network resources of the plurality of computing units form a resource cluster according to the computing type of the computing task, and a resource cluster corresponds to a computing task of a certain computing type.

[0100] In some embodiments, the various computing service types of the time-sensitive computing of the integrated platform are analyzed in advance to determine the combination of computing units 110 for each computing service, which correspond to the respective resource clusters.

[0101] In some embodiments, when an edge server 100 (i.e., a physical node) is elastically scaled up, the computing units 110 and their network resources in the edge server 100 are registered to the respective resource clusters accordingly.

[0102] In some embodiments, the scheduled computing unit and the network resources between it are preferentially determined from the resource cluster corresponding to the computing type of the computing task. When the idle computing unit 110 in the corresponding resource cluster cannot meet the QoS requirements of the computing task, cross-resource cluster scheduling is then performed. When preferentially determining the scheduled computing unit for the computing task from the resource cluster corresponding to the computing type of the computing task, the computing unit 110 and the corresponding network resources are also preferentially scheduled from an edge server 100 (i.e., a physical node).

[0103] In this way, by prioritizing scheduling from the resource cluster corresponding to the computing type of the computing task, the scheduling computing unit and the network resources between them are matched with the computing task, which not only improves the time determinism of the computing task, but also improves the utilization rate of the computing unit and the network resources between them.

[0104] Figure 4 This is a schematic diagram illustrating scheduling from resource clusters in Embodiment 2 of a time-sensitive computing and time-sensitive networking integrated platform. The diagram shows two resource clusters: Task 1 is scheduled within one cluster, and Task 2 is scheduled within the other. Task 1 acquires its QoS during the time period t2. The algorithm model involves scheduling computing resources and corresponding network resources for Task 1. Network resources transmit input data during the time periods t2+T and t2+2T, computing resources perform calculations during the time periods t2+4T and t2+5T, and transmit the calculation results during the time period t2+7T.

[0105] In some embodiments, each resource cluster is configured with a backup computing unit and network resources between the computing units in the backup computing unit, and the backup resources are used to replace the scheduled computing unit or the scheduled network resources when they are unavailable.

[0106] In some embodiments, CCC42 is further configured to schedule a set of redundant computing units and network resources between the computing units in the redundant computing units for a computing task. The redundant computing units and their network resources are preferentially scheduled from within the resource cluster corresponding to the computing type of the computing task. When the scheduled computing unit or its network resources are unavailable, the redundant computing unit and its network resources are used as substitutes.

[0107] Among them, the redundant or backup computing unit not only realizes the highly reliable backup of computing resources for computing tasks, but also realizes the highly reliable transmission of data flow between computing resources, thereby achieving highly reliable backup of resources to complete computing tasks.

[0108] In summary, the computing unit of the second embodiment of the integrated time-sensitive computing and time-sensitive networking platform is implemented on an edge server. Each edge server can be added to the integrated time-sensitive computing and time-sensitive networking platform through elastic expansion, realizing online expansion of time-sensitive computing. Furthermore, highly reliable backup of resources to complete computing tasks is achieved through redundant or backup computing units and network resources between computing units.

[0109] Let's combine the following... Figure 5 An example of an industrial control system is described.

[0110] An embodiment of an industrial control platform is deployed on the integrated platform described in Embodiment 2 of a time-sensitive computing and time-sensitive networking integrated platform. It has all the advantages of the integrated platform and adds a control unit and a human-machine unit. The control unit is connected to the TSN network and collects data from sensors and / or sends control commands to actuators through the TSN network. The human-machine unit is used for human-machine interaction of the industrial control platform. The control unit or the human-machine unit is also used to make decisions based on the calculation results of the computing unit.

[0111] Figure 5 The structure of an embodiment of an industrial control system is shown, which includes an edge server 200, a sensor 510, and an actuator 520 connected to a TSN network.

[0112] The edge server 200 can be flexibly expanded to multiple as needed. Each edge server 200 has all the structures of the edge server 100 in Embodiment 2 of the Time-Sensitive Computing and Time-Sensitive Networking Integrated Platform. Compared with the edge server 100 in Embodiment 2 of the Time-Sensitive Computing and Time-Sensitive Networking Integrated Platform, it adds several control units 120 and human-machine units 130.

[0113] Each computing unit 110 is connected to the TSN network through the TSN port 410 of its respective edge server 200. The CNC, CUC, and CCC for resource allocation for time-sensitive computing in the TSN network are not shown in the figure.

[0114] The control unit 120 connects to the TSN network via the TSN port 410 of the edge server 200, and collects data from the sensor 510 and / or sends control commands to the actuator 520 through the TSN network. The human-machine unit 130 is used for human-machine interaction on the industrial control platform.

[0115] The control unit 120 sends the calculation task to the scheduled calculation unit 110. The scheduled calculation unit 110 performs time-sensitive calculation to obtain the calculation result. The calculation result is sent to the corresponding control unit 120 or human-machine unit 130 through the TSN network for decision-making. The decision is sent to the control unit 120 to generate control commands.

[0116] When the scheduled computing units 110 come from different edge servers 200, the total computing time includes the transmission latency of the corresponding TSN network resources between the scheduled computing units 110.

[0117] In some embodiments, the computing task is generated by the control unit 120 and sent to the CUC of the TSN network; in other embodiments, the computing task is generated by the human-machine unit 130 and sent to the CUC.

[0118] In summary, one embodiment of an industrial control platform combines time-sensitive computing and other information processing systems (IT systems) with industrial control systems (OT) through the integrated platform provided in the above embodiment. By determining the data transmission through a TSN network, the convergence of IT and OT is achieved, forming an integrated industrial control system for real-time sensing, computing, decision-making, and control.

[0119] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A time-sensitive computing and time-sensitive network integration platform, comprising: include: The computing unit and the TSN network: each computing unit is located on a physical node, connected to the TSN network through its physical node, and synchronized with the TSN network time. The CUC of the TSN network is used to obtain the QoS of time-sensitive computing tasks; wherein, the QoS includes the task completion time required by the computing task; The centralized network configuration (CNC) or centralized computing configuration (CCC) of the TSN network is used to determine the scheduled computing unit to complete the computing task based on the QoS, computing resources, and network resources using an algorithm model. The total duration of the computing time of the scheduled computing unit and the transmission delay of the data stream between the scheduled computing units is equal to or less than the task completion time required by the QoS. The QoS also includes the computing power and completion time confidence required for the computing task; the computing unit scheduled by the CNC or the CCC completes the computing task within the task completion time of the QoS, and the expected completion time confidence is greater than or equal to the completion time confidence of the QoS.

2. The integrated platform of claim 1, wherein, Each type of computation task corresponds to an algorithm model. When the required QoS of the same type of computation task is different, the available computing units calculated by the algorithm model are different and the completion time confidence is different.

3. The integrated platform of claim 1, wherein, When the CNC or CCC uses the algorithm model to determine the scheduled computing unit to complete the computing task based on the QoS, computing resources and network resources, it prioritizes determining the scheduled computing unit to complete the computing task from a physical node.

4. The integrated platform according to claim 1, characterized in that, The CNC or CCC is further configured to determine the type of computing unit required for the computing task and the available computing units in that type based on the QoS, computing resources, and network resources, wherein the total duration of the computing time of the available computing units and the transmission delay of the data stream between the computing units is equal to or less than the task completion time required by the QoS.

5. The integrated platform according to claim 1, characterized in that, The CNC or CCC is further configured to, when multiple groups of scheduled computing units meet the QoS requirements, select, from the space composed of the computing power of the scheduled computing units and the expected completion time confidence, the group of scheduled computing units corresponding to the point with the smallest distance from the computing power and completion time confidence of the QoS requirements as the final scheduled computing units.

6. The integrated platform according to claim 1, characterized in that, The CNC or CCC is also used to set a priority for the computing task according to the QoS of the computing task, and the computing task with the highest priority is scheduled to be a computing unit.

7. The integrated platform according to claim 1, characterized in that, In the case of scheduling configuration parameters for data flows between scheduled computing units in a TSN network, the CNC is also used to generate computing configuration parameters and send them to the scheduled computing units; the computing configuration parameters include the priority level and transmission delay of the data flow of the computing task.

8. The integrated platform according to claim 1, characterized in that, In the absence of scheduling configuration parameters for the data flow between scheduled computing units in the TSN network, the CNC is also used to generate scheduling configuration parameters for the data flow and distribute them to the TSN switches between the scheduled computing units; the scheduling configuration parameters include gating parameters and traffic shaping parameters; The TSN switch performs deterministic scheduling of the data stream according to the scheduling configuration parameters, so that the data stream completes transmission within its corresponding deterministic duration.

9. The integrated platform according to claim 1, characterized in that, The process of completing the computing task includes: the scheduled computing unit completing the corresponding calculation and the deterministic transmission of the data stream of the task through the network resources corresponding to the scheduled unit. The data stream includes the data stream between computing units that have logical relationships in computation and are located on different physical nodes.

10. The integrated platform according to claim 1, characterized in that, The scheduled computing unit also sets up a cache queue according to the computing scheduling parameters. The size of the cache queue is set according to the data volume of the data stream and the bandwidth of the network resources between the scheduled computing units. When the scheduled computing unit performs calculations, the data of the data stream is cached in a cache queue and timestamped for transmission to the TSN network.

11. The integrated platform according to claim 1, characterized in that, When the computation task includes multiple serially computed subtasks, if the computation unit running the current subtask detects a computational anomaly, it rolls back to the computation unit corresponding to the previous subtask and recalculates based on the timestamp of the current subtask.

12. The integrated platform according to claim 1, characterized in that, When a subtask of the computation task is completed, its corresponding computation unit is released asynchronously.

13. The integrated platform according to claim 1, characterized in that, Both the events of the computing task and the events of the computing unit are timestamped when they are generated.

14. The integrated platform according to claim 1, characterized in that, Each computing unit includes a minimum scheduling unit of at least one computing power type on a physical node, and the CNC or CCC is also used to allocate a corresponding number of computing cores of the corresponding computing power type to each minimum scheduling unit.

15. The integrated platform according to claim 1, characterized in that, According to the computing type of the computing task, a number of computing units and the network resources between the computing units form a resource cluster, and each resource cluster is used to complete a computing task of a certain type. The network resources between the scheduled computing unit and the network resources between them are preferentially scheduled from the resource cluster corresponding to the computing type of the computing task.

16. The integrated platform according to claim 15, characterized in that, Each resource cluster is configured with a backup of the compute unit resources and the network resources between the compute units in that backup compute unit resources.

17. The integrated platform according to claim 15, characterized in that, The CNC or CCC is also used to schedule a set of redundant computing units for the task, and the redundant computing units and the network resources between them are preferentially scheduled from the resource cluster corresponding to the computing type of the computing task. The CNC or CCC is also used to replace the redundant computing units and network resources between them when the scheduled computing units or the network resources between them are unavailable.

18. The integrated platform according to claim 1, characterized in that, The algorithm model is a state machine and / or an artificial intelligence model that is pre-set according to the type of various computing tasks.

19. An industrial control system, characterized in that, include: The integrated platform, control unit, and human-machine unit according to any one of claims 1 to 18, wherein the control unit is connected to the TSN network of the integrated platform; The control unit is used to collect data from the sensor in real time and / or send control commands to the actuator in real time through the TSN network; The human-machine unit is used for human-machine interaction in the industrial control platform; The control unit or the human-machine unit is also used to make decisions based on the calculation results of the computing unit of the integrated platform.

Images