Dragonfly all-optical data center network service deployment method and system and optical channel

Abstract

The application relates to a Dragonfly all-optical data center network service deployment method and system and an optical system, and comprises the following steps: constructing an all-optical data center network; obtaining service requirements, generating a service sequence needing to be deployed based on the all-optical center network; and deploying the service sequence by using an optoelectronic optical algorithm, which comprises the following steps: when inter-group service is blocked, the final deployment time is calculated first, then all wavelengths of the wavelength selection switch outside the group where the source server and the target server are located are scanned to obtain all ToRs connected by the wavelength selection switch, signal reception is performed on the ToRs first, and then a transmitter is called to perform retransmission; the ToRs are traversed, the shortest route r1 from the source server to the intermediate-level ToR and the shortest route r2 from the intermediate-level ToR to the target server are calculated, and the final deployment time under the new route taking r1 and r2 as the new route is recorded; and the route with the earliest final deployment time is selected for service deployment. The application has the advantages of low delay, large port scale, flexible port configuration and low energy consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical communication technology, in particular to a service deployment method and system of a Dragonfly all-optical data center network and an optical system. BACKGROUND

[0002] With the development of resource-intensive and latency-sensitive emerging applications such as Internet of Things, Internet of Vehicles, virtual reality and augmented reality, the traffic of data centers is growing explosively, which puts higher requirements on the storage, processing and communication of data centers. High-performance computing is an application type that combines multiple servers to complete computationally intensive tasks, but as high-performance computing technology is in the key stage of development from 100P to E level, its network connection method is the key research direction. Research also shows that high-speed interconnection network is an important part of high-performance computing system and data center, which determines the upper limit of network performance. Therefore, researchers have studied various networks based on electrical switching and optical switching, and deployed services on these networks.

[0003] Fat tree architecture based on electrical switching is the earliest high-performance interconnection network topology, which inherits the low diameter advantage of butterfly structure network and can effectively avoid the disadvantages of unique path and load imbalance of butterfly network. C-Through architecture is an optical and electrical hybrid high-performance interconnection network, which uses electronic switches to process packet-based data and uses MEMS switches to transmit data with large bandwidth requirements. OSA topology is an all-optical interconnection structure, which can realize the establishment and reconstruction of the topology by configuring MEMS, WSS and ToR devices. Dragonfly structure proposed by Kim et al. is a typical representative of large-scale, low-diameter high-performance computing network, which can better utilize the port diversity of high-base routers by virtually grouping a certain amount of routers.

[0004] The expansion scale of electrical switching networks such as Fat tree is theoretically limited by the number of ports of core layer switches, which cannot well support One-to-All and All-to-All network communication modes, and is not conducive to deploying high-performance distributed services. Topologies such as C-Through use a single optical interconnection device, which has limited scalability and MEMS switches can cause reconstruction delay. In addition, in order to optimize the mixed use of electrical switching systems and optical switching systems, complex real-time traffic management is required. OSA architecture uses multiple optical switching devices, but still uses MEMS switches as top-level switches, and does not have the need for electrical switching to smooth delay-sensitive services. In general, electrical switching-based interconnection networks have high switching flexibility, but their low capacity, poor scalability and high energy consumption cannot meet the needs of high-performance computing. SUMMARY

[0005] To this end, the technical problem to be solved by the present application is to overcome the technical defects of low capacity, poor scalability and high energy consumption of the interconnection network based on electrical switching in the prior art.

[0006] To solve the above technical problems, the present application provides a service deployment method of a Dragonfly all-optical data center network, comprising the following steps:

[0007] S1, constructing an all-optical data center network;

[0008] S2, obtaining service requirements, and generating a service sequence to be deployed based on the all-optical data center network;

[0009] S3, deploying the service sequence using an optoelectronic-optical algorithm, comprising:

[0010] When the intra-group service and the inter-group non-blocking service are used, Dijkstra algorithm and port awareness are used for deployment;

[0011] When the inter-group service is blocked, the final deployment time is calculated first, then all the ToRs connected by the wavelength selection switches outside the groups where the source server s and the target server d are located are obtained, the signals are received on the ToRs first, and then the transmitter is called for retransmission;

[0012] The ToRs are traversed, the shortest route r1 from the source server s to the intermediate ToR and the shortest route r2 from the intermediate ToR to the target server d are calculated, and the final deployment time under the new route with r1 and r2 is recorded;

[0013] The route with the earliest final deployment time is selected for service deployment, wherein the ToR is the top of the rack.

[0014] As a preferred, the S1 comprises:

[0015] The all-optical data center network comprises servers and all-optical switches, each switch is composed of N N WSSs composed of N 1xN WSSs, wherein WSS represents a wavelength selection switch, and 1xN WSS represents a WSS containing one inlet end and N outlet ends;

[0016] In each N N WSS, p ports are directly connected to servers, a-1 ports are connected to other WSSs and form a group with a WSSs, and the remaining h ports are connected to WSSs of other different groups, wherein p+a-1+h=N;

[0017] Each WSS port is connected to only one server, and each server can be configured with multiple tunable optical transceivers.

[0018] As preferred, the S2 comprises:

[0019] A set of high-performance computing service requirements are obtained, each service requirement being represented as (s, d, T, r, t);

[0020] Wherein, s represents the source server, d represents the target server, T represents the time required for service transmission, t represents the time at which service scheduling deployment begins, time slots are used as the measurement unit of time, all data are randomly generated, and r represents the initial routing of the service.

[0021] As preferred, the S3 comprises:

[0022] S31, the shortest routing r of the source server s and the target server d is obtained according to the Dijkstra algorithm, and flag = 0 is set, wherein the flag is used to mark whether the tf constraint is considered in the algorithm process;

[0023] S32, the service scheduling start time is initially set as the service arrival time; when the time slot t is judged, it is judged whether the transceivers are available; if not, t = t + 1 is set, and the first available time tf of the service transceiver is recorded;

[0024] S33, the final wavelength configuration set Wu of the ports related to the routing passed by the current service is obtained by scanning the previously established services, and the final wavelength configuration set Wu is removed from the available wavelength set of the current service to update the available wavelength set;

[0025] For inter-group services, the available wavelength set is updated again according to the tf constraint when flag = 0; wherein the tf constraint is that when t-tf>T / W, the available wavelength set is empty, and the wavelength is selected according to the least use principle, wherein W is the maximum available wavelength of a single link; when no available wavelength is found, t = t + 1 and flag = 1 are updated;

[0026] S34, the configuration information of the WSS ports passed by the shortest routing r is scanned, and it is determined whether reconfiguration is needed according to whether the wavelengths are configured between the ports;

[0027] If reconfiguration is needed, the earliest available time slot tθ of the WSS port after reconfiguration is calculated, t = tθ is set, and step S32 is returned; if reconfiguration is not needed, the information under this routing is recorded, and the service is directly deployed;

[0028] S35, for service requirements with flag = 1 and inter-group, all WSSs not in the groups where the source server s and the target server d are located are scanned to obtain, and all ToRs connected by the intermediate WSSs are traversed;

[0029] According to the Dijkstra algorithm, the route r1 of the source server s to the intermediate ToR, the route r2 of the intermediate ToR to d is calculated, whether there is a continuous r1 and r2 corresponding scheduling time data is the same at this time is checked; if not, t=t+1 is updated, r is updated as r1 and r2 respectively, and is brought into step S32, and the scheduling time under different routes after the photoelectric-optical conversion of the intermediate level is recorded; if there is, the intermediate level iteration ends this time;

[0030] The service start scheduling slot t under different intermediate levels introducing photoelectric-optical is compared, and the route and wavelength corresponding to the minimum value are selected for service deployment: when the minimum value is the same as the initial record, it is indicated that the service does not need to be operated photoelectric-optically, and the shortest route of the source server s and the target server d is directly used for establishing an optical channel.

[0031] As preferred, in the S3, for each service sequence, the default service start scheduling deployment time is a service arrival time slot, and whether there is corresponding transceiver, wavelength and WSS port configuration in the all-optical data center network in the next T continuous TSs for service deployment is scanned;

[0032] If there is, the optical channel is established between the corresponding WSS using the wavelength of the service, and the current service deployment is successful; otherwise, the next time slot is judged again until the service deployment is successful and the algorithm is exited.

[0033] As preferred, the S3 further comprises:

[0034] S4, after all service sequences are processed, the performance indicators of all successfully deployed services are obtained to evaluate the algorithm performance.

[0035] As preferred, the performance indicators include one or more of the following: total time slot, service average waiting delay and WSS average configuration times.

[0036] The application discloses a service deployment system of a Dragonfly all-optical data center network, comprising:

[0037] A network construction module is used for constructing an all-optical data center network;

[0038] A service sequence generation module is used for obtaining service requirements, and generating service sequences to be deployed based on the all-optical data center network;

[0039] A service sequence deployment module deploys service sequences using an electro-optical algorithm, including: when intra-group services and inter-group non-blocking services are used, Dijkstra algorithm and port awareness are used for deployment; when inter-group services are blocked, the final deployment time is calculated first, then all wavelengths of the wavelength selective switch outside the group where the source server s and the target server d are located are scanned to obtain all ToRs connected by the wavelength selective switch, and the ToRs are used for signal receiving first, and then a transmitter is called for retransmission; the ToRs are traversed, the shortest route r1 from the source server s to the intermediate ToR and the shortest route r2 from the intermediate ToR to the target server d are calculated, and the final deployment time under the new route of r1 and r2 is recorded; the route with the earliest final deployment time is selected for service deployment, and the ToR is the top of the rack.

[0040] As preferred, it further comprises:

[0041] An evaluation module is used to obtain the performance indicators of all successfully deployed services to evaluate the algorithm performance after all service sequences are completely processed.

[0042] The application discloses an optical channel of an all-optical data center network, which is obtained based on the above service deployment method.

[0043] Compared with the prior art, the above technical scheme of the application has the following advantages:

[0044] 1. Based on the low-diameter and high-bandwidth Dragonfly topology, the application uses WSS to construct an all-optical switching node, and further constructs an all-optical data center network. The topology retains the low-diameter feature and further improves the network bandwidth, and the application has the advantages of low delay, large port scale, flexible port configuration and low energy consumption.

[0045] 2. The application proposes a service deployment method of a Dragonfly all-optical data center network based on a reconfigurable WSS. The method schedules services according to the service location in the network, the configuration information of the WSS, the constraints of the transceiver and the wavelength, and can effectively reduce the total time slot required for successful service deployment, the average waiting delay of services and the average configuration times of the WSS. DETAILED DESCRIPTION

[0046] Figure 1 A flowchart of the service deployment method of the Dragonfly all-optical data center network in the application;

[0047] Figure 2 A flowchart of the electro-optical algorithm based on port awareness;

[0048] Figure 3 A schematic diagram of the all-optical data center network with Dragonfly (2, 2, 1). DETAILED DESCRIPTION

[0049] The application will be further described below in connection with the drawings and specific embodiments so that those skilled in the art can better understand and implement the application, but the embodiments are not intended to limit the application.

[0050] Referring to Figure 1 As shown in the drawings, the application discloses a service deployment method of a Dragonfly all-optical data center network, comprising the following steps:

[0051] S1, constructing an all-optical data center network.

[0052] The all-optical data center network comprises servers and all-optical switches, each switch is composed of N×N WSSs of N 1×N WSSs, wherein WSS represents a wavelength selective switch, and 1×N WSS represents a WSS comprising one inlet end and N outlet ends.

[0053] In each N×N WSS, p ports are directly connected to servers, a-1 ports are connected to other WSSs and form a group with a WSSs, and the remaining h ports are connected to WSSs of other different groups, wherein p+a-1+h=N.

[0054] Each WSS port is connected to only one server, and each server can be configured with multiple tunable optical transceivers.

[0055] S2, obtaining service requirements and generating a service sequence to be deployed based on the all-optical data center network.

[0056] Specifically, a group of high-performance computing service requirements are obtained, and each service requirement is represented as (s, d, T, r, t); wherein s represents a source server, d represents a target server, T represents the time required for service transmission, t represents the time at which service scheduling and deployment starts, time slots are used as the time measurement unit, all data are randomly generated, and r represents the initial route of the service.

[0057] S3, deploying the service sequence using an optoelectronic algorithm, comprising:

[0058] When the intra-group service and the inter-group non-blocking service are used, Dijkstra algorithm and port awareness are used for deployment;

[0059] When the inter-group service is blocked, the final deployment time is calculated first, then all wavelength selective switches outside the groups where the source server s and the target server d are located are scanned to obtain all ToRs connected by the wavelength selective switches, signal reception is performed on the ToRs first, and then the transmitter is called for retransmission;

[0060] Traverse the ToR, calculate the shortest route r1 of the source server s to the intermediate ToR and the shortest route r2 of the intermediate ToR to the target server d, and record the final deployment time under the new route as r1 and r2;

[0061] Select the route with the earliest final deployment time for business deployment, wherein the ToR is the top of the rack.

[0062] In S3, for each business sequence, the default business start scheduling deployment time is the business arrival time slot, and it is scanned whether there are corresponding transceivers, wavelengths and WSS port configurations in the all-optical data center network in the next T continuous TSs for business deployment;

[0063] If there is, the current business deployment is successful by establishing an optical channel between the corresponding WSS using the wavelength; otherwise, it will be judged again at the next time slot until the business deployment is successful.

[0064] The optical-electric-optical algorithm in S3 specifically includes:

[0065] S31, obtain the shortest route r of the source server s and the target server d according to the Dijkstra algorithm, and set flag=0, wherein the flag is used to mark whether the tf constraint is considered in the algorithm process;

[0066] S32, the business scheduling start time is initially defaulted as the business arrival time; when judging the time slot t, it is judged whether the transceivers are all available, if not, set t=t+1, and record the first available time tf of the business transceiver;

[0067] S33, scan the previously established business to obtain the final wavelength configuration set Wu of the ports related to the route passed by the current business, and remove the final wavelength configuration set Wu from the available wavelength set to update the available wavelength set;

[0068] For inter-group business, when flag=0, the available wavelength set is updated again according to the tf constraint; wherein the tf constraint is: when t-tf>T / W, the available wavelength set is empty, and the wavelength is selected according to the least use principle, wherein W is the maximum available wavelength of a single link; when no available wavelength is found, update t=t+1, flag=1;

[0069] S34, scan the configuration information of the WSS ports passed by the shortest route r, and determine whether reconfiguration is needed according to whether the wavelengths are configured between the ports;

[0070] If reconfiguration is needed, find the earliest available time slot tθ of the WSS port after reconfiguration, set t=tθ, and return to step S32; if reconfiguration is not needed, record the information under this route, and directly deploy the business;

[0071] S35, for the service requirement of flag = 1 and between groups, scanning to obtain all WSSs not in the groups where the source server s and the target server d are located, traversing all intermediate level WSSs connected ToR;

[0072] According to the Dijkstra algorithm, the route r1 of the source server s to the intermediate level ToR, the route r2 of the intermediate level ToR to d are calculated, and it is checked whether the scheduling time data corresponding to the continuous r1 and r2 are the same; if not, then t = t + 1 is updated, r is updated as r1 and r2 respectively, and is brought into step S32, and the scheduling time under different routes after the introduction of optoelectronic optical conversion in the intermediate level is recorded; if there is, then the intermediate level traversal ends this time;

[0073] The service start scheduling time slot t under different intermediate levels introducing optoelectronic optical is compared, and the route and wavelength corresponding to the minimum value are selected for service deployment: when the minimum value is the same as the initial record, it is indicated that the service does not need to be optoelectronic optical, and the shortest route of the source server s and the target server d is directly used for the establishment of the optical channel.

[0074] S4, after all the service sequences are processed, the performance indicators of all successfully deployed services are obtained to evaluate the algorithm performance.

[0075] The performance indicators include one or more of the following: total time slot, service average waiting delay and WSS average configuration times.

[0076] The application discloses a service deployment system of a Dragonfly all-optical data center network, which comprises a network construction module, a service sequence generation module and a service sequence deployment module.

[0077] The network construction module is used for constructing the all-optical data center network.

[0078] The service sequence generation module is used for obtaining service requirements and generating service sequences to be deployed based on the all-optical data center network.

[0079] The service sequence deployment module deploys the service sequence using an optoelectronic optical algorithm, including: when the intra-group service and the inter-group non-blocking service, the Dijkstra algorithm and port sensing are used for deployment; when the inter-group service is blocked, the final deployment time is calculated first, then all ToRs connected by the wavelength selection switches outside the groups where the source server s and the target server d are located are obtained, the signal is received on the ToR first, and then the transmitter is called for retransmission; the shortest route r1 of the source server s to the intermediate level ToR and the shortest route r2 of the intermediate level ToR to the target server d are calculated, and the final deployment time under the new route taking r1 and r2 as the new route is recorded; the route with the earliest final deployment time is selected for service deployment, wherein the ToR is the rack top.

[0080] Further, the application also includes: an evaluation module is used for obtaining performance indexes of all successfully deployed services to evaluate algorithm performance after all service sequences are completely processed.

[0081] The application discloses an optical channel of an all-optical data center network, which is obtained based on the service deployment method.

[0082] In the following, the technical solutions in the application are further described and explained in combination with specific embodiments.

[0083] The application provides the following technical solutions: a service deployment method of a Dragonfly all-optical data center network based on a reconfigurable WSS, comprising the following steps:

[0084] (1) constructing an all-optical data center network based on a WSS and adopting a Dragonfly topology

[0085] The application adopts a WSS (Wavelength Selective Switch) as an optical switch module and forms a Dragonfly topology by the WSS. The all-optical data center network is composed of servers and all-optical switches, and each switch is an N×N WSS composed of multiple 1×N WSSs. In each N×N WSS, p ports are directly connected to ToRs (Top of Rack), a-1 ports are connected to other WSSs and form a group with a WSS, and the remaining h ports are connected to WSSs in other groups. The all-optical data center network can be recorded as Dragonfly (p, a, h). In addition, it is assumed that each port is connected to only one ToR, and multiple tunable optical transceivers can be configured on each ToR. The servers with high-performance computing requirements can establish optical path connections of specific wavelengths between different WSS ports through optical transceivers on the top of the rack, and the wavelength reconfiguration between WSS ports can be determined by an algorithm.

[0086] The application uses (s, d, T, r, t) to represent a high-performance computing service, wherein s, d, T and t respectively represent a source server, a target server, a time required for service transmission and a time at which service scheduling deployment is started, and a time slot (TS) is used as a time measurement unit, and all data is randomly generated. r represents an initial route of the service, which is generally empty. Then, a series of service sequences are generated according to the total service amount. For each service requirement, a path r between the servers s and d needs to be found, and the wavelengths available in T consecutive TSs along the path.

[0087] (3) deploying high-performance computing services

[0088] For each service in the high performance service sequence, the algorithm sets the default service start deployment time as the service arrival time slot, and scans the all-optical data center network to determine whether there are corresponding transceivers, wavelengths and WSS port configurations for service deployment in the next T consecutive TSs. If there are, the service can use the wavelength, establish an optical channel between the corresponding WSSs, and the current service deployment is successful; otherwise, the algorithm will determine again in the next time slot until the service deployment is successful.

[0089] The present application is based on optical switching and takes into account the advantages of electrical switching, and proposes a port-aware optical-electrical-optical algorithm, which can efficiently deploy services.

[0090] 1) The flowchart of the port-aware optical-electrical-optical algorithm is shown in Figure 2 The main steps are as follows:

[0091] ① Select the route. Obtain the shortest route r of the source server s and the target server d according to the Dijkstra algorithm, and set flag = 0.

[0092] ② Update the transceiver time. The service deployment start time is initially set as the service arrival time, and when the time slot t is determined, it is judged whether all the transceivers are available. If not, set t = t + 1. Record the first available time tf of the service transceivers (only once).

[0093] ③ Assign the wavelength. First, perform port awareness, i.e. scan the previously established services to obtain the final wavelength configuration Wu of the ports related to the route of the current service. In order to avoid port reconfiguration, remove Wu from the available wavelength set of the current service. For inter-group services, when flag = 0, update the wavelength set again according to the tf constraint. The tf constraint is: when t-tf>T / W (W is the maximum available wavelength of a single link), the available wavelength set is empty. Finally, select the wavelength according to the least use principle. If no available wavelength is found, update t = t + 1 and flag = 1.

[0094] ④ Assign the WSS port. Scan the configuration information of the WSS ports passed by the route r, and determine whether reconfiguration is needed according to whether the wavelengths have been configured between the ports. If reconfiguration is needed, find the earliest available time slot tθ of the WSS ports after reconfiguration, set t = tθ, and return to step ②; if reconfiguration is not needed, record the information under this route and directly deploy the service.

[0095] ⑤Photo-optical operation. For flag = 1 and inter-group service requirement, first need to scan all WSSs not in the group where s and d are located, and traverse all intermediate-level ToRs connected by WSSs. According to Dijkstra algorithm, calculate the route r1 from s to the intermediate-level ToR, the route r2 from the intermediate-level ToR to d, and check whether the scheduling time corresponding to the continuous r1 and r2 is the same. If not, update t = t + 1, and update r as r1 and r2 respectively, and bring into step

[0096] ②, record the scheduling time under different routes after adding photo-optical conversion in the intermediate level; if there is, the intermediate level traversal ends this time. Finally, compare the service start scheduling time slots t under different intermediate levels with photo-optical operation, select the minimum value corresponding to the route and wavelength for service deployment. When the minimum value is the same as the initial record, it indicates that the service does not need photo-optical operation, and the shortest route of the source server s and the target server d is directly used for optical channel establishment.

[0097] (4) Evaluate the performance of the algorithm

[0098] After all high-performance service sequences are processed, the total time slot required for successful deployment of all services, the average service waiting delay and the average WSS configuration times can be obtained, which are used as performance indicators of the algorithm.

[0099] Further, the application provides a service deployment method of a Dragonfly all-optical data center network based on a reconfigurable WSS, comprising the following steps:

[0100] (1) Construct an all-optical data center network with a Dragonfly topology using WSS as the switching system. In this embodiment, the all-optical data center network based on WSS with Dragonfly topology is composed of servers and switches. Figure 3 An example of such an all-optical data center network with Dragonfly (2, 2, 1) is shown, in which the total number of ports, the number of groups, the total number of WSSs and the total number of racks are 8, 3, 6 and 12 respectively. The index number of the WSS is 1-6, and the index number of the rack is 1-12. Each rack is equipped with 2 transceivers, supporting 4 wavelengths, and the corresponding index is 1-4. A plurality of servers are hung inside the rack. The specific structure of WSS6 is also shown in the figure, which uses a 4*4 WSS for connection, and each pair of ports of the WSS is connected to the input and output ports of the corresponding device.

[0101] (2) Generate the service sequence of high performance computing. The total number of service requests in the network is fixed, varying from 1000 to 10000, with an interval of 1000, and the time slots required for service transmission are randomly generated in the range of [1, 10] TSs. The service arrival time slot is initialized to 0. In the service configuration process, wavelength conversion is not allowed on a single route, but after the introduction of intermediate O / E / O conversion, the wavelengths before and after conversion can be inconsistent.

[0102] (3) Service high performance computing services. When the service is served, the application adopts the O / E / O algorithm based on port awareness to establish the optical channel.

[0103] (4) Return the performance index after the service deployment is completed.

[0104] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) having computer-usable program code embodied in the medium.

[0105] The application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the embodiments of the application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device that implements the functions specified in the flowchart and / or block diagram. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The means for performing the functions specified in one or more flows and / or blocks.

[0106] These computer program instructions can also be stored in a computer-readable memory that can guide the computer or other programmable data processing devices to work in a specific way, so that the instructions stored in the computer-readable memory produce a product including instruction means, which implements the functions specified in the flowchart and / or block diagram. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The means for performing the functions specified in one or more flows and / or blocks.

[0107] These computer program instructions can also be loaded into a computer or other programmable data processing devices, so that a series of operational steps are performed on the computer or other programmable data processing devices to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing devices provide steps for implementing the function specified in the flowchart Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0108] Obviously, the above embodiments are only examples for clearly illustrating the present application, and are not intended to limit the embodiments. Based on the above description, other different forms of changes or variations can be made by those skilled in the art. Here, all the embodiments are not required to be exhausted, and the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A service deployment method for a Dragonfly all-optical data center network, characterized in that, Includes the following steps: S1. Construct an all-optical data center network; S2. Obtain business requirements and generate the required service sequence based on the all-optical data center network; S3. Deploy service sequences using photoelectric algorithms, including: When there are no blocking services within a group or between groups, the Dijkstra algorithm and port awareness are used for deployment. When inter-group services are blocked, first calculate the final deployment time, then scan all wavelength selection switches outside the group where the source server s and target server d are located to obtain all ToRs connected to the wavelength selection switches. First receive the signal on the ToR, and then call the transmitter to retransmit. Traverse the ToRs, calculate the shortest route r1 from the source server s to the intermediate ToR and the shortest route r2 from the intermediate ToR to the target server d, and record the final deployment time under the new routes r1 and r2; Select the route with the earliest final deployment time for service deployment, where ToR is the top of the rack.

2. The service deployment method for the Dragonfly all-optical data center network according to claim 1, characterized in that, S1 includes: The all-optical data center network includes servers and all-optical switches. Each switch consists of N × N WSSs composed of N 1 × N WSSs, where WSS represents wavelength selective switch and 1 × N WSS represents a WSS containing one inlet and N outlets. In each N×N WSS, there are p ports directly connected to the server, a-1 ports connected to other WSSs, forming a group with a WSSs, and the remaining h ports connected to WSSs in other different groups, where p+a-1+h=N; Each WSS port connects to only one server, and each server can be configured with multiple tunable optical transceivers.

3. The service deployment method for the Dragonfly all-optical data center network according to claim 2, characterized in that, S2 includes: Obtain a set of high-performance computing service requirements, each of which is represented as (s,d,T,r,t); Where s represents the source server, d represents the target server, T represents the time required for service transmission, t represents the time when the service begins scheduling and deployment, time slots are used as the unit of time measurement, all data is generated randomly, and r represents the initial route of the service.

4. The service deployment method for the Dragonfly all-optical data center network according to claim 3, characterized in that, S3 includes: S31. Obtain the shortest route r between source server s and destination server d according to Dijkstra's algorithm, and set flag=0, where flag is used to mark whether the tf constraint is considered in the algorithm process; S32. The service scheduling start time is initially set to the service arrival time by default. It is determined whether the transceivers are available at time slot t. If they are not available, t is set to t+1, and the first available time tf of the service transceiver is recorded. S33. Scan the previously established services to obtain the final wavelength configuration set Wu of the ports related to the route traversed by the current service. Remove the final wavelength configuration set Wu from the available wavelength set of the current service to update the available wavelength set. For inter-group services, when flag=0, the available wavelength set is updated again according to the tf constraint; where the tf constraint is: when t-tf>T / W, the available wavelength set is empty, and the wavelength is selected according to the principle of least use, where W is the maximum available wavelength of a single link; when no available wavelength is found, t=t+1 is updated, and flag=1. S34. Scan the configuration information of the WSS ports traversed by the shortest route r, and determine whether reconfiguration is needed based on whether wavelength has been configured between the ports. If reconfiguration is required, calculate the earliest available timeslot tθ of the WSS port after reconfiguration, set t=tθ, and return to step S32; if reconfiguration is not required, record the information under this route and deploy the service directly. S35. For service requirements with flag=1 and between groups, scan to get all WSSs that are not in the groups of source server s and target server d, and traverse the ToRs connected to all intermediate WSSs. According to Dijkstra's algorithm, calculate route r1 from source server s to intermediate level ToR and route r2 from intermediate level ToR to d. Check if there are consecutive scheduling time data corresponding to r1 and r2 that are the same. If not, update t=t+1, update r to r1 and r2 respectively, and substitute into step S32 to record the scheduling time under different routes after adding photoelectric conversion in the intermediate level. If they exist, the intermediate level traversal ends. Compare the service start scheduling time slot t under different intermediate levels of photoelectric and optical introduction, and select the route and wavelength corresponding to the minimum value for service deployment: when the minimum value is the same as the initial record, it indicates that the service does not need to perform photoelectric and optical operations, and the optical channel is established directly using the shortest route between the source server s and the target server d.

5. The service deployment method for the Dragonfly all-optical data center network according to claim 3, characterized in that, In S3, for each service sequence, the default service start scheduling and deployment time is the service arrival time slot, and it scans whether the all-optical data center network has the corresponding transceiver, wavelength and WSS port configuration to deploy the service in the next T consecutive TSs. If it exists, the service will establish an optical channel between the corresponding WSSs using the wavelength, and the current service deployment will be successful; otherwise, it will be checked again in the next time slot, and will continue until the service is successfully deployed before exiting.

6. The service deployment method for the Dragonfly all-optical data center network according to claim 1, characterized in that, Following S3, the following also includes: S4. After all business sequences have been processed, obtain the performance metrics of all successfully deployed businesses to evaluate the algorithm performance.

7. The service deployment method for the Dragonfly all-optical data center network according to claim 6, characterized in that, The performance metrics include one or more of the following: total time slots, average service latency, and average number of WSS configurations.

8. A service deployment system for a Dragonfly all-optical data center network, characterized in that, include: Network construction module, which is used to construct an all-optical data center network; A service sequence generation module is used to obtain service requirements and generate service sequences to be deployed based on the all-optical data center network. The service sequence deployment module uses photoelectric and optical algorithms to deploy service sequences, including: when intra-group services and inter-group services are not blocked, deployment is performed using the Dijkstra algorithm and port awareness; when inter-group services are blocked, the final deployment time is first calculated, then all wavelength selection switches outside the groups where the source server s and target server d are located are scanned to obtain all ToRs connected to the wavelength selection switches, the signal is first received on the ToR, and then the transmitter is called to retransmit; the ToRs are traversed, the shortest route r1 from the source server s to the intermediate ToR and the shortest route r2 from the intermediate ToR to the target server d are calculated, and the final deployment time under the new routes r1 and r2 is recorded; the route with the earliest final deployment time is selected for service deployment, where the ToR is the top of the rack.

9. The service deployment system for the Dragonfly all-optical data center network according to claim 8, characterized in that, Also includes: The evaluation module is used to obtain the performance metrics of all successfully deployed services after all business sequences have been processed in order to evaluate the algorithm performance.

Images