An Edge Data Center Task Scheduling Method Based on Optical Interconnection

Abstract

The present invention relates to a task scheduling method for an edge data center based on optical interconnection. Based on a distributed edge data system, task scheduling is performed for each received optical packet task. The entire scheduling process can reduce the transmission bottleneck of electrical signals on high-speed data, and has the advantages of supporting bursty services, low latency, high bandwidth utilization, and large throughput, which is in line with the characteristics of large data volume and low latency in edge computing; placing a small edge data center at the network edge can transfer the latency-sensitive traffic of the optical transport network to the edge network, while sharing the computing, storage, and network transmission bandwidth pressure of the cloud data center, and can provide users with nearby service resources, thereby reducing the propagation latency of remote transmission to the cloud data center.

Description

Technical Field

[0001] The present invention relates to a task scheduling method for an edge data center based on optical interconnection, belonging to the technical field of edge computing. Background Art

[0002] Edge computing improves the aggregated high-latency service model of cloud computing, and has advantages such as high real-time performance, high stability, and high data security. The interconnection and data interaction between existing edge data centers mostly rely on wireless or wired electrical exchange methods, and their task scheduling problems have been widely studied, and the development of related theories, technologies, and algorithms is also relatively mature. However, the stability, throughput, transmission speed, latency, security, etc. of such edge computing systems are difficult to meet the requirements of real-time interconnection and interoperability of massive and heterogeneous data. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a task scheduling method for an edge data center based on optical interconnection, which ensures good blocking conditions of the entire system and balanced utilization of each edge node through an efficient resource allocation strategy and execution mechanism.

[0004] The present invention adopts the following technical solutions to solve the above technical problems: The present invention designs a task scheduling method for an edge data center based on optical interconnection, which is used to implement the scheduling and processing of each optical packet task received by a distributed edge data system; the distributed edge data system includes a main computing node ECN-M and at least two edge computing centers respectively connected and managed by it, and each edge computing center respectively includes an edge computing main node ECN-M 边 and at least one slave node respectively connected and managed by it; each slave node respectively includes an edge optical switching unit OSU including a Rec-FDL 边 and at least two edge computing nodes respectively connected to the edge optical switching unit OSU 边 ;

[0005] Based on the computing capabilities of the main computing node ECN-M, each edge computing main node ECN-M 边 and each edge computing node all meet the computing power requirements of each optical packet task received by the distributed edge data system, and the edge data center task scheduling method includes respectively performing the following steps for each slave node:

[0006] Step A. Patrol each optical packet task received by each edge computing node in the slave node to obtain the respective patrol end times of each optical packet task, and then concurrently execute the following steps B, C, and D;

[0007] Step B. Sort the end times of each round of polling from small to large, and then, for each optical packet task corresponding to each end time of the round of polling in sequence, schedule the optical packet task to the corresponding node for processing or schedule it to the Rec-FDL for circulation;

[0008] Step C. Based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, cyclically and sequentially process each optical packet task that circulates in the Rec-FDL until each optical packet task in the Rec-FDL is updated to the state of waiting for an idle node or discarded;

[0009] Step D. For each optical packet task in the Rec-FDL that is in the state of waiting for an idle node, based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, cyclically and sequentially process each optical packet task until each optical packet task is scheduled to the corresponding node for processing.

[0010] As a preferred technical solution of the present invention: in the above Step A, for each optical packet task received from each edge computing node in the slave node, according to the computing amount d(T i ) of each optical packet task, the length l(T i ) of each optical packet task, and the arrival rate λ(T i ) of the edge computing node corresponding to each optical packet task, the scheduling unit of the slave node performs polling for each optical packet task to obtain the end time of each round of polling corresponding to each optical packet task, and then enters Step B; where 1≤i≤I, I represents the total number of optical packet tasks received by each edge computing node in the slave node, and T i represents the i-th optical packet task.

[0011] As a preferred technical solution of the present invention: in the above Step B, based on the sorting of the end times of each round of polling from small to large, perform the following Steps B1 to B8 to schedule each optical packet task to the corresponding node for processing or schedule it to the Rec-FDL for circulation;

[0012] Step B1. Based on the sorting of the end times of each round of polling from small to large, sequentially select the first end time of the round of polling that has not participated in the processing of Steps B1 to B8 as the target end time of the round of polling, and enter Step B2;

[0013] Step B2. Determine whether there are other end times of the round of polling within a preset threshold time range after the target end time of the round of polling. If so, obtain the optical packet tasks corresponding to the target end time of the round of polling and each of these other end times of the round of polling, and enter Step B3; otherwise, enter Step B6;

[0014] Step B3. The edge computing node preset as the control unit among the slave nodes forms the priority sorting of each optical packet task according to the sorting of the corresponding preset attributes for each optical packet task obtained in the current iteration step B2, and proceeds to step B4;

[0015] Step B4. Determine whether there is a destination node in the header information of the optical packet task corresponding to the highest priority. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the highest priority to the destination node for processing, and transport the optical packet tasks of the remaining priorities to the Rec-FDL for circulation, and then proceed to step B8; if the destination node is a non-idle node, transport all the optical packet tasks of each priority to the Rec-FDL for circulation, and then proceed to step B8; otherwise, proceed to step B5;

[0016] Step B5. Based on the main computing nodes ECN-M, each edge computing main node ECN-M 边 , and each edge computing node in the preset order of each regional range, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the highest priority to one of the idle nodes for processing, and transport the optical packet tasks of the remaining priorities to the Rec-FDL for circulation, and then proceed to step B8; otherwise, transport all the optical packet tasks of each priority to the Rec-FDL for circulation, and then proceed to step B8;

[0017] Step B6. Determine whether there is a destination node in the header information of the optical packet task corresponding to the target polling end time. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the target polling end time to the destination node for processing, and then proceed to step B8; if the destination node is a non-idle node, transport the optical packet task corresponding to the target polling end time to the Rec-FDL for circulation, and then proceed to step B8; otherwise, proceed to step B7;

[0018] Step B7. Based on the main computing nodes ECN-M, each edge computing main node ECN-M 边 , and each edge computing node in the preset order of each regional range, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the target polling end time to one of the idle nodes for processing, and then proceed to step B8; otherwise, transport the optical packet task corresponding to the target polling end time to the Rec-FDL for circulation, and then proceed to step B8;

[0019] Step B8. Determine whether there is a polling end time that has not participated in the processing of steps B1 to B8 in the sorting of the polling end times from small to large. If so, return to step B1; otherwise, complete the scheduling of each optical packet task to the corresponding node for processing or scheduling to the Rec-FDL for circulation.

[0020] As a preferred technical solution of the present invention: in step C, based on the order of the time points when each optical packet task enters the Rec-FDL, loop through each optical packet task that circulates in the Rec-FDL in sequence, and execute the following steps C1 to C2 until each optical packet task in the Rec-FDL is updated to the waiting idle node state or discarded;

[0021] Step C1. Determine whether the length l(T i ) of the optical packet task is less than the average length l ave of all the optical packet tasks received by the node. If so, it means that the optical packet task can pass through the Rec-FDL without blocking, update the optical packet task to the waiting idle node state, and continue to circulate in the Rec-FDL; otherwise, enter step C2;

[0022] Step C2. Obtain the difference between the time point when the optical packet task enters the Rec-FDL and the time point when the adjacent subsequent optical packet task enters the Rec-FDL, and determine whether the difference is greater than (l(T i ) - l ave ) / V. If so, it means that the adjacent subsequent optical packet task does not conflict with the optical packet task, update the optical packet task to the waiting idle node state, and continue to circulate in the Rec-FDL; otherwise, it means that the adjacent subsequent optical packet task conflicts with the optical packet task, and discard the adjacent subsequent optical packet task; V represents the transmission speed of the optical packet task.

[0023] As a preferred technical solution of the present invention: in step D, for each optical packet task in the Rec-FDL that is in the waiting idle node state, based on the order of the time points when each optical packet task enters the Rec-FDL, loop through and process each optical packet task in sequence, and execute the following steps D1 to D2 until each optical packet task is scheduled to the corresponding node for processing;

[0024] Step D1. Determine whether there is a destination node in the header information of the optical packet task. If so, if the destination node is an idle node, schedule the optical packet task to the destination node for processing; if the destination node is a non-idle node, keep the optical packet task continuing to circulate in the Rec-FDL; otherwise, enter step D2;

[0025] Step D2. Based on the main computing nodes ECN-M and each edge computing main node ECN-M in each area range in the preset order 边, and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task to one of the idle nodes for processing; otherwise, keep the optical packet task continuing to loop in the Rec-FDL.

[0026] As a preferred technical solution of the present invention: the main computing node ECN-M under each regional range based on the preset order, each edge computing main node ECN-M 边 , and each edge computing node, in the process of determining whether there is an idle node among them:

[0027] First, based on each edge computing main node ECN-M within the range of the slave node where the optical packet task corresponding to the target polling end time is located 边 , determine whether there is an idle node among them. If so, an idle node is found. Otherwise, further based on the edge computing main node ECN-M of the edge computing center where the slave node is located 边 , and each edge computing node within the range of other slave nodes in the edge computing center, determine whether there is an idle node among them. If so, an idle node is found. Otherwise, further based on the main computing node ECN-M, the edge computing main nodes ECN-M of other edge computing centers 边 , and each edge computing node within the range of each slave node in other edge computing centers, determine whether there is an idle node among them. If so, an idle node is found. Otherwise, it means that there is no idle node.

[0028] As a preferred technical solution of the present invention: the main computing node ECN-M under each regional range based on the preset order, each edge computing main node ECN-M 边 , and each edge computing node, in the process of determining whether there is an idle node among them, if there are at least two idle nodes, then for the optical packet task to be scheduled, respectively for each idle node, assume that the optical packet task to be scheduled is scheduled to the idle node, and perform the following operations:

[0029] First, according to the optical fiber length L during the process of scheduling the optical packet task to be scheduled to the idle node fiber , and the amount of the optical packet task to be scheduled d(T i ), according to t trans = d(T i ) × L fiber , obtain the transmission time t trans ;

[0030] Then, according to the computing speed V of the idle node calculate , according to t calculate = d(T i ) / V calculate , obtain the computing time t calculate; Finally, according to the filtering times t of each filter TF through which the optical packet task to be scheduled passes to the idle node and t1; according to t s = t1 + t trans + t calculate , the service time t s corresponding to the process of scheduling the optical packet task to be scheduled to the idle node is obtained;

[0031] Furthermore, the service times t s corresponding to the processes of scheduling the optical packet task to be scheduled to each idle node are obtained respectively, and the idle node corresponding to the minimum service time t s is selected to schedule the optical packet task to be scheduled to the idle node for processing.

[0032] As a preferred technical solution of the present invention: during the parallel execution of step B, step C, and step D, the operation priority of scheduling the optical packet task to the corresponding node for processing in step B is higher than the operation priority of scheduling the optical packet task to the corresponding node for processing in step D.

[0033] As a preferred technical solution of the present invention: the header information of the optical packet task is obtained through the all-optical packet header extraction module HEM.

[0034] Compared with the prior art, the task scheduling method for the edge data center based on optical interconnection of the present invention has the following technical effects by adopting the above technical solutions:

[0035] (1) The task scheduling method for the edge data center based on optical interconnection designed by the present invention is a research on high-efficiency data balancing processing algorithms and edge-cloud collaborative optimization algorithms, which can minimize the occurrence of system conflict blocking and packet loss during the scheduling process, improve transmission reliability, and make the system blocking situation good and the node utilization rate balanced; and perform balanced allocation of services according to the geographical locations, computing and storage resources of each edge computing node in the system, as well as the task computing volume, priority, packet arrival rate, etc., to ensure the balanced utilization rate of each edge computing node; at the same time, set some optimization objectives to optimize the scheduling of the task allocation on the edge nodes;

[0036] (2) The task scheduling method for edge data centers based on optical interconnection designed by the present invention aims at the real-time requirements of data collection and transmission in application scenarios such as distributed data centers and industrial Internet of Things networks. It proposes the implementation principles and methods for interoperability between different communication protocols and connected devices in a distributed edge computing environment, proposes new scalability and expandability technologies for edge data centers, and proposes business balancing data processing algorithms and collaborative optimization algorithms that support the efficiency of data collection, exchange, and transmission. It can not only promote the application of edge computing technology in large scientific projects such as the industrial Internet and new infrastructure construction, but also drive the development of civilian edge computing technologies such as wireless edge, mobile edge, building edge, power edge, and energy edge, and contribute to China's Industry 4.0 and Intelligent Manufacturing 2025. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is the architecture diagram of a distributed edge data center system based on optical packet switching in the implementation of the design method of the present invention;

[0038] Figure 2 is the architecture diagram of the edge computing center in a distributed edge data center system based on optical packet switching in the implementation of the design method of the present invention;

[0039] Figure 3 is the schematic diagram of the task scheduling process of the edge computing layer based on optical interconnection designed by the present invention;

[0040] Figure 4 is the schematic diagram of the task scheduling method for edge data centers based on a conflict resolution mechanism designed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The following further describes in detail the specific embodiments of the present invention with reference to the accompanying drawings of the specification.

[0042] The task scheduling method for edge data centers based on optical interconnection designed by the present invention is used to implement the scheduling and processing of each optical packet task received by a distributed edge data system. In actual applications, as Figure 1 shown, the distributed edge data system includes a main computing node ECN-M and at least two edge computing centers. Among them, each edge computing center includes an edge computing main node ECN-M 边 , and at least one slave node; each slave node includes an edge optical switching unit OSU 边 , and at least two edge computing nodes. In the structure of each slave node: each edge computing node is respectively docked with the corresponding edge optical switching unit OSU 边 .

[0043] In the structure of each edge computing center: If the number of slave nodes is 1, then in the structure of this edge computing center, the optical switching unit OSU in this slave node 边 is docked with the corresponding edge computing master node ECN-M 边 ; if the number of slave nodes is at least 2, then this edge computing center further includes an intermediate optical switching unit OSU 中 . In the structure of this edge computing center, the optical switching unit OSU in each slave node 边 is respectively connected to this intermediate optical switching unit OSU 中 , and this intermediate optical switching unit OSU 中 is docked with the corresponding edge computing master node ECN-M 边 .

[0044] Taking the edge computing master node ECN-M of each edge computing center 边 as a leaf node, and applying each intermediate optical switching unit OSU 中 as a non-leaf node to construct a tree structure, and the intermediate optical switching unit OSU as the root node in the tree structure 中 is docked with the main computing node ECN-M

[0045] The main computing node ECN-M is connected and communicates with the cloud data center, and each slave node in each edge computing center is connected to the user-side device and application

[0046] In practical applications, the structures of each edge computing node are the same specifically. As Figure 2 shown, each edge computing node respectively includes a multi-protocol interface, a storage module, a data assembly module, an analysis and calculation module, an optical transceiver module, and a wireless transceiver module. In the structure of each edge computing node, the multi-protocol interface supports multiple communication protocols, provides flexible data access capabilities, and can realize the data upload and download of user-side applications; the multi-protocol interface is connected and communicates with the storage module, the storage module is connected and communicates with the analysis and calculation module, and the storage module and the analysis and calculation module have the ability to intelligently interpret data and real-time data analysis, can provide differentiated and easy-to-use development kits and libraries, and support the rapid development of data center applications; the analysis and calculation module is connected and communicates with the data assembly module, and the data assembly module is respectively connected and communicates with the optical transceiver module and the wireless transceiver module. The optical transceiver module is used to realize the data uplink output end TX and the data downlink output end RX of the edge computing node. In the application, the payload is stored in the packet header, and the packet header carries information such as the storage capacity and computing power required for the task. If there is a clear destination, it should also carry address information

[0047] And based on the number of slave nodes in each of the above-mentioned edge computing centers being at least 2, as Figure 2 shown, the structures of each slave node are as follows:

[0048] Edge Optical Switching Unit (OSU) 边 Including arrayed waveguide grating AWG, 1×2 splitter, all-optical packet header extraction module HEM, circular fiber delay line REC-FDL, Splitter, and high-speed optical switch OS, the number of edge computing nodes is equal to 1≤q≤Q, Q represents the number of edge computing centers, 1≤n q ≤N q , N q represents the number of slave nodes in the qth edge computing center, Indicates the number of edge computing nodes in the nth slave node in the qth edge computing center; the data uplink output terminal TX of each edge computing node is connected to the corresponding edge optical switching unit OSU 边 Each input end of the arrayed waveguide grating AWG in the middle and the output end of the arrayed waveguide grating AWG are connected to the single input end of the 1×2 splitter, one of the output ends on the 1×2 splitter is connected to the input end of the corresponding all-optical packet header extraction module HEM, the output end of the all-optical packet header extraction module HEM is connected to the feedback input end of each edge computing node, the other output end on the 1×2 splitter is connected to the input end of the corresponding circular optical fiber delay line REC-FDL, and the output end of the circular optical fiber delay line REC-FDL is connected The single input of the splitter, Each output end on the splitter is connected to each high-speed optical switch OS one by one, and the control end on each edge computing node is connected to each high-speed optical switch OS to control the state of each high-speed optical switch OS. The high-speed optical switches OS are further connected to the data downlink input port RX of each edge computing node one by one; the remaining high-speed optical switch OS is further connected to the intermediate optical switching unit OSU in the edge computing center to which it belongs. 中 , and the intermediate optical switching unit OSU 中 The feedback end is connected to the input end of the arrayed waveguide grating AWG in the slave node.

[0049] And for the edge optical switching unit OSU in each slave node 边 For the circular fiber delay line REC-FDL in Figure 2As shown in the figure, each loop optical fiber delay line REC-FDL respectively includes a first tunable filter TF1, a second tunable filter TF2, a third tunable filter TF3, a fourth tunable filter TF4, a first coupler A1, a second coupler A2, a third coupler A3, and a fourth coupler A4. Among them, the single-channel side end of the first coupler A1 constitutes the input end of the loop optical fiber delay line REC-FDL. One end of the two-channel side of the first coupler A1 is connected to one end of the two-channel side of the fourth coupler A4 through the third tunable filter TF3. The other end of the two-channel side of the first coupler A1 is connected to one end of the two-channel side of the second coupler A2 through the fourth tunable filter TF4. One end of the two-channel side of the third coupler A3 is connected to the other end of the two-channel side of the fourth coupler A4 through the first tunable filter TF1. The single-channel side end of the fourth coupler A4 is connected to the single-channel side end of the third coupler A3. The other end of the two-channel side of the third coupler A3 is connected to the other end of the two-channel side of the second coupler A2 through the second tunable filter TF2. The single-channel side end of the second coupler A2 is connected to form the output end of the loop optical fiber delay line REC-FDL.

[0050] In practical applications, the intermediate optical switching unit OSU in each edge computing center involved 中 , such as Figure 2 shown in the figure, is designed to respectively include an arrayed waveguide grating AWG, a 1×2 splitter, an all-optical packet header extraction module HEM, a 1×N q splitter, and N q tunable wavelength converters TWC, N q signal feedback paths. The structure of each intermediate optical switching unit OSU 中 is as follows:

[0051] N q tunable wavelength converters TWC are respectively in one-to-one correspondence with the edge optical switching units OSU of each slave node in the edge computing center to which the intermediate optical switching unit OSU 中 belongs. The input ends of each tunable wavelength converter TWC are respectively connected to the high-speed optical switches OS used to connect the intermediate optical switching unit OSU 边 in the corresponding edge optical switching unit OSU. The output ends of each tunable wavelength converter TWC are connected to the input ends of the corresponding arrayed waveguide grating AWG. The output end of the arrayed waveguide grating AWG is connected to the single-channel input end of the corresponding 1×2 splitter. One output end of the 1×2 splitter is connected to the input end of the corresponding all-optical packet header extraction module HEM. The output end of the all-optical packet header extraction module HEM is connected to the edge computing master node ECN-M in the edge computing center to which the intermediate optical switching unit OSU 边 belongs 中 中 ​边 , the other output end of the 1×2 splitter is connected to the corresponding single input end of the 1×N q splitter. Each output end of the 1×N q splitter is respectively and correspondingly connected to the input end of each signal feedback path. The structures of the signal feedback paths are the same as each other. Each signal feedback path respectively includes a tunable filter TF and a tunable wavelength converter TWC. The input end of the tunable filter TF in each signal feedback path constitutes the input end of the corresponding signal feedback path. The output end of the tunable filter TF is connected to the input end of the corresponding tunable wavelength converter TWC. The control end of the intermediate optical switching unit OSU 中 in the edge computing center to which it belongs, the edge computing master node ECN-M 边 respectively controls the tunable filter TF in each of the signal feedback paths. The output end of the tunable wavelength converter TWC constitutes the output end of the corresponding signal feedback path. The output ends of the respective signal feedback paths are respectively and correspondingly connected to the edge optical switching units OSU 中 of the slave nodes in the edge computing center to which the intermediate optical switching unit OSU 边 belongs. The output ends of the respective signal feedback paths are respectively and correspondingly connected to the corresponding edge optical switching units OSU 边 and are connected to the input ends of the arrayed waveguide gratings AWG therein.

[0052] Based on the above-designed distributed edge data system, the present invention designs a task scheduling method for an edge data center based on optical interconnection. Based on the computing capabilities of the main computing node ECN-M, each edge computing master node ECN-M 边 , and each edge computing node all meet the computing capability requirements of each optical packet task received by the distributed edge data system. The edge data center task scheduling method includes, for each slave node respectively, as Figure 3 shown, performing the following steps.

[0053] Step A. Poll the respective optical packet tasks received by each edge computing node in the slave node to obtain the respective polling end times corresponding to the respective optical packet tasks, and then concurrently perform the following steps B, C, and D.

[0054] In the above step A, for the respective optical packet tasks received by each edge computing node in the slave node, according to the computing amount d(T i ) of each optical packet task, the length l(T i ) of each optical packet task, and the arrival rate λ(T i), the scheduling unit SU of the slave node polls each optical packet task to obtain the polling end time corresponding to each optical packet task, and then proceeds to step B; where 1 ≤ i ≤ I, and I represents the total number of optical packet tasks received by each edge computing node in the slave node, T i represents the i-th optical packet task.

[0055] Step B. Based on the sorting of the polling end times from small to large, for each optical packet task corresponding to each polling end time in turn, the optical packet task is scheduled to the corresponding node for processing or scheduled to the Rec-FDL for circulation.

[0056] In practical applications, in the above step B, based on the sorting of the polling end times from small to large, the following steps B1 to B8 are executed to schedule each optical packet task to the corresponding node for processing or scheduled to the Rec-FDL for circulation.

[0057] Step B1. Based on the sorting of the polling end times from small to large, sequentially select the first polling end time that has not participated in the processing of steps B1 to B8 as the target polling end time, and enter step B2.

[0058] Step B2. Determine whether there are other polling end times within a preset threshold duration after the target polling end time. If so, obtain the optical packet tasks corresponding to the target polling end time and each of these other polling end times, and enter step B3; otherwise, enter step B6.

[0059] Step B3. The edge computing node preset as the control unit in the slave node sorts the optical packet tasks obtained in step B2 of this iteration according to the corresponding preset attributes to form the priority sorting of these optical packet tasks, and enter step B4.

[0060] Step B4. Determine whether there is a destination node in the header information of the optical packet task corresponding to the highest priority. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the highest priority to the destination node for processing through the 1→2 link, and select the remaining optical packet tasks of each priority through the TF on the 1→3 link and send them to the 4-port for transportation to the Rec-FDL for circulation, and then enter step B8; if the destination node is a non-idle node, transport all the optical packet tasks of each priority to the Rec-FDL for circulation, and then enter step B8; otherwise, enter step B5.

[0061] In the application, regarding the header information of the optical packet task, when the optical packet of the optical packet task passes through the AWG, it is split into two beams by a 1×2 optical splitter. One beam enters the all-optical packet header extraction module (HEM), and the HEM obtains the header information of the optical packet task. The other beam is scheduled, that is, it is scheduled to the corresponding node for processing or scheduled to the Rec-FDL for circulation.

[0062] Step B5. Based on the preset order, for each main computing node (ECN-M) and each edge computing main node (ECN-M) within each regional range 边 and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the highest priority to one of the idle nodes for processing via a 1→2 link, and select the optical packet tasks of the remaining priorities via the TF on the 1→3 link and send them to the 4-port for transportation to the Rec-FDL for circulation, then enter Step B8; otherwise, transport all the optical packet tasks of each priority to the Rec-FDL for circulation, then enter Step B8.

[0063] Step B6. Determine whether there is a destination node in the header information of the optical packet task corresponding to the target polling end time. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the target polling end time to the destination node for processing via a 1→2 link, then enter Step B8; if the destination node is a non-idle node, select the optical packet task corresponding to the target polling end time via the TF on the 1→3 link and send it to the 4-port for transportation to the Rec-FDL for circulation, then enter Step B8; otherwise, enter Step B7.

[0064] Step B7. Based on the preset order, for each main computing node (ECN-M) and each edge computing main node (ECN-M) within each regional range 边 and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the target polling end time to one of the idle nodes for processing via a 1→2 link, then enter Step B8; otherwise, select the optical packet task corresponding to the target polling end time via the TF on the 1→3 link and send it to the 4-port for transportation to the Rec-FDL for circulation, then enter Step B8.

[0065] Step B8. Determine whether there is a polling end time that has not participated in the processing from Step B1 to Step B8 in the sorting of each polling end time from small to large. If so, return to Step B1; otherwise, complete the scheduling of each optical packet task to the corresponding node for processing or scheduling to the Rec-FDL for circulation.

[0066] Step C. Based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, loop through each optical packet task that is cycled in the Rec-FDL in sequence, and execute the following Step C1 to Step C2 until each optical packet task in the Rec-FDL is updated to the waiting for idle node state or discarded.

[0067] Step C1. Determine whether the length l(T i ) of the optical packet task is less than the average length l of all the optical packet tasks received by the node. ave If so, it means that the optical packet task can pass through the Rec-FDL without blocking. Update the optical packet task to the waiting for idle node state and continue to loop in the Rec-FDL; otherwise, go to Step C2.

[0068] Step C2. Obtain the difference between the time point when the optical packet task enters the Rec-FDL and the time point when the adjacent next optical packet task enters the Rec-FDL, and determine whether this difference is greater than (l(T i ) - l ave ) / V. If so, it means that there is no conflict between the adjacent next optical packet task and this optical packet task. Update the optical packet task to the waiting for idle node state and continue to loop in the Rec-FDL; otherwise, it means that there is a conflict between the adjacent next optical packet task and this optical packet task, and discard the adjacent next optical packet task; V represents the transmission speed of the optical packet task.

[0069] Step D. For each optical packet task in the Rec-FDL that is in the waiting for idle node state, based on the order of the time points when each of these optical packet tasks enters the Rec-FDL in sequence, loop through and process each of these optical packet tasks in sequence, and execute the following Step D1 to Step D2 until each optical packet task is scheduled to the corresponding node for processing.

[0070] Step D1. Determine whether there is a destination node in the header information of the optical packet task. If so, if the destination node is an idle node, schedule the optical packet task to the destination node through port 2 for processing; if the destination node is a non-idle node, keep the optical packet task continuing to loop in the Rec-FDL; otherwise, go to Step D2.

[0071] Step D2. Based on the main computing nodes ECN-M in each area range in the preset order, each edge computing main node ECN-M 边 , and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task to one of the idle nodes through port 2 for processing; otherwise, keep the optical packet task continuing to loop in the Rec-FDL.

[0072] During the execution of the above-designed steps, for the main computing node ECN-M, each edge computing main node ECN-M within the range of each area based on the preset order 边 , and each edge computing node, during the process of determining whether there is an idle node among them:

[0073] First, for each edge computing main node ECN-M within the range of the slave nodes where the optical packet task corresponding to the target polling end time is located 边 , determine whether there is an idle node among them. If so, an idle node is found; otherwise, further based on the edge computing main node ECN-M of the edge computing center where the slave node is located 边 , and each edge computing node within the range of other slave nodes in this edge computing center, determine whether there is an idle node among them. If so, an idle node is found; otherwise, further based on the main computing node ECN-M, the edge computing main nodes ECN-M of other edge computing centers 边 , and each edge computing node within the range of each slave node in other edge computing centers, determine whether there is an idle node among them. If so, an idle node is found; otherwise, it means there is no idle node.

[0074] And during the process of determining whether there is an idle node among them, if there are at least two idle nodes, then for the optical packet task to be scheduled, for each idle node respectively, assume that the optical packet task to be scheduled is scheduled to the idle node, and perform the following operations:

[0075] First, according to the optical fiber length L during the process of scheduling the optical packet task to be scheduled to the idle node fiber , and the amount of the optical packet task to be scheduled d(T i ), calculate t trans = d(T i ) × L fiber , to obtain the transmission time t trans .

[0076] Then, according to the computing speed V of the idle node calculate , calculate t calculate = d(T i ) / V calculate , to obtain the computing time t calculate .

[0077] Finally, according to the sum of the filtering times t1 of each filter TF passed by the optical packet task to be scheduled when it is scheduled to the idle node; calculate t s = t1 + t trans + t calculate , to obtain the service time t corresponding to the process of scheduling the optical packet task to be scheduled to the idle node s .

[0078] Furthermore, the service time t corresponding to the process of scheduling the optical packet task to be scheduled to each idle node is obtained s , and the idle node corresponding to the minimum service time t is selected s for scheduling the optical packet task to be scheduled to the idle node for processing.

[0079] In practical applications, based on the above-designed method, task scheduling is performed for a single optical packet task, as Figure 4 shown. Regarding scheduling in practical applications, it is possible to further design the edge nodes to be sorted from high to low according to node efficiency during scheduling, and tasks are preferentially scheduled to the edge nodes with high node efficiency. Compared with tasks randomly entering idle edge nodes, the processing efficiency of the entire system can be improved, and the blocking state of the entire system will be greatly improved. At the same time, some optimization goals are set to optimize the scheduling of task allocation on edge nodes to achieve the optimal effect of one or more optimization goals. For example, a genetic algorithm is used for scheduling optimization, and the objective function is to minimize the total delay of completing all tasks. During the application process of the genetic algorithm, the population is first initialized, the fitness value of the initial population is calculated. If the termination condition is not reached, selection, crossover, and mutation operations are performed to update the population, and the population moves closer to the optimal solution during this process, and so on, until the termination condition is reached and the optimal solution is output.

[0080] And in practical applications, the above-mentioned scheduling of optical packet tasks to corresponding nodes for processing is involved in both step B and step D. When, at a certain time point, there are simultaneously optical packet tasks that need to execute step B to be scheduled to corresponding nodes for processing and optical packet tasks that need to execute step D to be scheduled to corresponding nodes for processing, it is designed that the operation priority of scheduling optical packet tasks to corresponding nodes for processing in step B is higher than the operation priority of scheduling optical packet tasks to corresponding nodes for processing in step D, that is, this operation in step B is executed first at this time, and then this operation in step D is executed.

[0081] The above-designed task scheduling method for edge data centers based on optical interconnection is a research on high-efficiency data balancing processing algorithms and edge-cloud collaborative optimization algorithms, which can minimize the occurrence of system conflict blocking and packet loss during the scheduling process, improve transmission reliability, and make the system blocking condition good and the node utilization rate balanced; and perform balanced allocation of services according to the geographical locations, computing and storage resources of each edge computing node in the system, as well as task computing volume, priority, packet arrival rate, etc., to ensure the balanced utilization rate of each edge computing node; at the same time, set some optimization goals to optimize the scheduling of task allocation on edge nodes;

[0082] And the designed task scheduling method for edge data centers based on optical interconnection, aiming at the real-time requirements of data collection and transmission in application scenarios such as distributed data centers and industrial Internet of Things networks, proposes the implementation principles and methods for interoperability between different communication protocols and connected devices in a distributed edge computing environment, proposes new edge data center scalability and expandability technologies, and proposes business-balanced data processing algorithms and collaborative optimization algorithms to support data collection, exchange, and transfer efficiency. It can not only promote the application of edge computing technology in large scientific projects such as the industrial Internet and new infrastructure construction, but also drive the development of civilian edge computing technologies such as wireless edge, mobile edge, building edge, power edge, and energy edge, contributing to China's Industry 4.0 and Intelligent Manufacturing 2025.

[0083] The designed task scheduling method for edge data centers based on optical interconnection in the entire design technical solution can meet the real-time collection, secure interaction, and predictive analysis requirements of multi-source, massive, and easily structured data in distributed edge computing. This edge computing system adopts a multi-level distributed management architecture and is located on the network edge side close to the data terminal, which can provide edge intelligent services nearby, solve the problems of insufficient real-time performance, insufficient bandwidth, high energy consumption, and insufficient data security of cloud computing, and meet the key requirements of the industry's digitalization in aspects such as real-time services, data aggregation, application intelligence, security, and privacy. This edge computing system transfers a large number of switching services to the optical domain, reduces the transmission bottleneck of electrical signals on high-speed data, can achieve the matching of switching capacity and the transmission capacity of wavelength division multiplexing, and through the combination with technologies such as new optical cross-connection and multi-protocol label switching, can realize the optimization of the network system and the rational utilization of resources; based on this edge computing system, a dynamic task scheduling method is designed to evenly allocate services according to the computing, storage resources, and network conditions of each edge computing node in the system, ensure the balanced utilization rate of each edge computing node, minimize system conflict blocking during the scheduling process, and improve transmission reliability;

[0084] Therefore, the business-balanced data processing algorithm and collaborative optimization algorithm proposed in the present invention to support data collection, exchange, and transfer efficiency can not only solve the theoretical and technical bottlenecks in the application of new optical interconnection, optical switching, and optoelectronic integration technologies in edge data centers, but also amplify the value of edge computing and cloud computing, and meet the basic requirements of the industry in aspects such as real-time services, application intelligence, security, and privacy protection.

[0085] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge scope of those of ordinary skill in the art.

Claims

1. A task scheduling method for an edge data center based on optical interconnection, which is used to implement the scheduling and processing of each received optical packet task by a distributed edge data system; characterized in that: The distributed edge data system includes a main computing node ECN-M and at least two edge computing centers respectively connected and managed by it. Each edge computing center includes an edge computing main node ECN-M 边 and at least one slave node respectively connected and managed by it; each slave node includes an edge optical switching unit OSU containing Rec-FDL 边 and at least two edge computing nodes respectively connected to the edge optical switching unit OSU 边 ​ Based on the main computing node ECN-M and each edge computing main node ECN-M 边 and the computing capabilities of each edge computing node all meet the computing capability requirements of each optical packet task received by the distributed edge data system, the edge data center task scheduling method includes performing the following steps for each slave node respectively: Step A. Poll each optical packet task received from each edge computing node in the slave node to obtain the polling end time corresponding to each optical packet task, and then execute the following steps B, C, and D in parallel; In the above step A, for each optical packet task received from each edge computing node in the slave node, according to the computing amount d(T i ), the length l(T i ), and the arrival rate λ(T i ) of each optical packet task corresponding to the received edge computing node, the scheduling unit of the slave node performs round-robin scheduling for each optical packet task to obtain the round-robin end time corresponding to each optical packet task, and then enters step B; where 1 ≤ i ≤ I, I represents the total number of optical packet tasks received by each edge computing node in the slave node, T i represents the i-th optical packet task; Step B. Sort based on the ascending order of each polling end time. For each optical packet task corresponding to each polling end time in sequence, through the search for idle nodes, schedule the optical packet task to the corresponding idle node for processing or schedule it to the Rec-FDL for circulation; Step C. Based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, loop through and execute the processing for each optical packet task circulating in the Rec-FDL until each optical packet task in the Rec-FDL is updated to the waiting idle node state or discarded; Step D. For each optical packet task in the Rec-FDL in the waiting idle node state, based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, loop through and execute the processing for each optical packet task until each optical packet task is scheduled to the corresponding node for processing.

2. The method for task scheduling of an edge data center based on optical interconnection according to claim 1, wherein: In the said Step B, sort based on the ascending order of each polling end time, and execute the following steps B1 to B8 to schedule each optical packet task to the corresponding node for processing or schedule it to the Rec-FDL for circulation; Step B1. Sort based on the ascending order of each polling end time, and sequentially select the first polling end time that has not participated in the processing of Steps B1 to B8 as the target polling end time, and enter Step B2; Step B2. Determine whether there are other polling end times within a preset threshold duration after the target polling end time. If so, obtain the optical packet tasks corresponding to the target polling end time and each of these other polling end times, and enter Step B3; Otherwise, enter Step B6; Step B3. The edge computing node preset as the control unit in the slave node sorts each optical packet task obtained in Step B2 of this iteration according to the corresponding preset attributes to form the priority sorting of each optical packet task, and enter Step B4; Step B4. Determine whether there is a destination node in the header information of the optical packet task corresponding to the highest priority. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the highest priority to the destination node for processing, and convey the remaining optical packet tasks with each priority to the Rec-FDL for circulation, and then enter Step B8; if the destination node is a non-idle node, convey all the optical packet tasks with each priority to the Rec-FDL for circulation, and then enter Step B8; Otherwise, enter Step B5; Step B5. Based on the main computing nodes ECN-M and each edge computing main node ECN-M within the range of each region in the preset order 边 and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the highest priority to one of the idle nodes for processing, and transport the optical packet tasks with the remaining priorities to the Rec-FDL for circulation, and then proceed to Step B8; otherwise, transport all the optical packet tasks with each priority to the Rec-FDL for circulation, and then proceed to Step B8; Step B6. Determine whether there is a destination node in the header information of the optical packet task corresponding to the target polling end time. If so, if the destination node is an idle node, schedule the optical packet task corresponding to the target polling end time to the destination node for processing, and then enter Step B8; if the destination node is a non-idle node, convey the optical packet task corresponding to the target polling end time to the Rec-FDL for circulation, and then enter Step B8; Otherwise, go to step B7; Step B7. Based on the main computing nodes ECN-M and each edge computing main node ECN-M within the range of each area in the preset order 边 , and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task corresponding to the target polling end time to one of the idle nodes for processing, and then proceed to Step B8; otherwise, convey the optical packet task corresponding to the target polling end time to the Rec-FDL for circulation, and then proceed to Step B8; Step B8. Determine whether there is a polling end time that has not participated in the processing from step B1 to step B8 in the sorting of the polling end times from small to large. If yes, return to step B1; otherwise, complete the scheduling of each optical packet task to the corresponding node for processing, or schedule it to the Rec-FDL for cycling.

3. The method for task scheduling of an edge data center based on optical interconnection according to claim 1, wherein: In the said step C, based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, loop through each optical packet task that cycles in the Rec-FDL in sequence, and execute the following steps C1 to C2 until each optical packet task in the Rec-FDL is updated to the waiting idle node state or discarded; Step C1. Determine whether the length l(T i ) of the optical packet task is less than the average length l of all the optical packet tasks received by the node ave . If yes, it means that the optical packet task can pass through the Rec-FDL without blocking. Update the optical packet task to the waiting idle node state and continue to loop in the Rec-FDL; otherwise, go to Step C2; Step C2. Obtain the difference between the time point when this optical packet task enters Rec-FDL and the time point when the next adjacent optical packet task enters Rec-FDL, and determine whether this difference is greater than (l(T i ) - l ave ) / V. If so, it means that the next adjacent optical packet task does not conflict with this optical packet task. Update this optical packet task to the waiting-for-idle-node state and continue to loop in Rec-FDL; otherwise, it means that the next adjacent optical packet task conflicts with this optical packet task, and discard the next adjacent optical packet task; V represents the transmission speed of the optical packet task.

4. The method for task scheduling of an edge data center based on optical interconnection according to claim 1, wherein: In the said step D, for each optical packet task in the waiting idle node state in the Rec-FDL, based on the order of the time points when each optical packet task enters the Rec-FDL in sequence, loop through and process each optical packet task in sequence, and execute the following steps D1 to D2 until each optical packet task is scheduled to the corresponding node for processing; Step D1. Determine whether there is a destination node in the header information of the optical packet task. If yes, if the destination node is an idle node, schedule the optical packet task to the destination node for processing; if the destination node is a non-idle node, keep the optical packet task continuing to cycle in the Rec-FDL; Otherwise, go to step D2; Step D2. Based on the main computing nodes ECN-M and each edge computing main node ECN-M within the range of each area in the preset order 边 , and each edge computing node, determine whether there is an idle node among them. If so, schedule the optical packet task to one of the idle nodes for processing; otherwise, keep the optical packet task continuing to loop in the Rec-FDL.

5. The task scheduling method for an edge data center based on optical interconnection according to claim 2 or 4, characterized in that: The main computing node ECN-M and each edge computing main node ECN-M under the range of each area based on the preset order 边 , and each edge computing node, in the process of determining whether there is an idle node among them: First, based on each edge computing master node ECN-M within the range of the slave nodes where the optical packet task corresponding to the target polling end time is located 边 , determine whether there is an idle node among them. If so, the idle node is found; otherwise, further based on the edge computing master node ECN-M of the edge computing center where the slave node is located 边 , and each edge computing node within the range of other slave nodes in the edge computing center, determine whether there is an idle node among them. If so, the idle node is found; otherwise, further based on the main computing node ECN-M, the edge computing master nodes ECN-M of other edge computing centers 边 , and each edge computing node within the range of each slave node in other edge computing centers, determine whether there is an idle node among them. If so, the idle node is found; otherwise, it means that there is no idle node.

6. The task scheduling method for an edge data center based on optical interconnection according to claim 2 or 4, characterized in that: The main computing node ECN-M and each edge computing main node ECN-M under the respective regional ranges based on the preset order 边 And each edge computing node, during the process of determining whether there are idle nodes among them, if there are at least two idle nodes, then for the optical packet task to be scheduled, respectively for each idle node, assume that the optical packet task to be scheduled is scheduled to the idle node, and perform the following operations: First, according to the length L of the optical fiber passed during the process of scheduling the to-be-scheduled optical packet task to an idle node fiber , and the to-be-scheduled optical packet task volume d(T i ), according to t trans = d(T i ) × L fiber , obtain the transmission time t trans ; Then calculate the speed V based on the idle nodes calculate , according to t calculate = d(T i ) / V calculate , obtain the calculation time t calculate ; Finally, according to the filtering time of each filter TF passed by the optical packet task to be scheduled to the idle node and t1; calculate t s = t1 + t trans + t calculate , and obtain the service time t s corresponding to the process of scheduling the optical packet task to be scheduled to the idle node; furthermore, obtain the service time t s corresponding to the process of scheduling the optical packet task to be scheduled to each idle node respectively, and select the idle node corresponding to the minimum service time t s for scheduling the optical packet task to be scheduled to this idle node for processing.

7. The method for task scheduling of an edge data center based on optical interconnection according to claim 1, characterized in that: During the parallel execution of step B, step C, and step D, the operation priority of scheduling the optical packet task to the corresponding node for processing in step B is higher than the operation priority of scheduling the optical packet task to the corresponding node for processing in step D.

8. The method for task scheduling of an edge data center based on optical interconnection according to claim 1, wherein: Obtain the header information of the optical packet task through the all-optical packet header extraction module HEM.

Images