Computing system and data transmission system

Abstract

A computing system includes a router configured to transmit data to a node; the transfer station is used for transmitting data to the node through the router; the first path is arranged between adjacent routers; and the second path is arranged between adjacent transfer stations. The computing system operates in a first mode that transmits data from a first target device to a second target device via a first path and a second path.

Description

Background Art

[0001] A conventional network on chip (NoC) system may include multiple routers, so that the NoC system can receive input data and transmit the input data to nodes through the routers. However, when data is input to the router or when data is output by the router, transmission delay may be caused. Therefore, if the transmission of data passes through many routers, a large data transmission delay may be caused. Summary of the invention

[0002] One object of the present application is to provide a computing system capable of reducing transmission delay.

[0003] Another object of the present application is to provide a data transmission system capable of reducing transmission delay.

[0004] The embodiment of the present invention discloses a computing system, comprising: a plurality of routers for transmitting data to nodes; a plurality of transfer stations for transmitting data to nodes or routers; a plurality of first paths arranged between adjacent routers; a plurality of second paths arranged between adjacent transfer stations. The computing system operates in a first mode, and transmits data from a first target device to a second target device via the first path and the second path during a first time interval. The first target device is one of a router or a node, and the second target device is one of a router or a node.

[0005] Another embodiment of the present invention discloses a data transmission system, comprising: a plurality of first transmission devices for transmitting data to a destination device; a plurality of second transmission devices for transmitting data to a destination device or a first transmission device; a plurality of first paths arranged between adjacent first transmission devices; a plurality of second paths arranged between adjacent second transmission devices. The data transmission system selectively operates in a first mode, and transmits data from a first destination device to a second destination device via the first path and the second path during a second time interval. The first destination device is one of the first transmission devices or one of the destination devices, and the second destination device is one of the first transmission devices or one of the destination devices.

[0006] In view of the above-described embodiments, transmission delay can be reduced because data can be transmitted through shortcuts formed by transfer stations rather than being limited to transmission through paths formed only by routers.

[0007] These and other objects of the present application will no doubt become apparent to those of ordinary skill in the art after reading the following detailed description of the various drawings and the preferred embodiments illustrated in the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1A schematic diagram showing a portion of a computing system according to an embodiment of the present application.

[0009] Figure 2 Show Figure 1 A schematic diagram of a more complete structure of a computing system is shown.

[0010] Figure 3 A schematic diagram showing the paths of the transfer station or router corresponding to different directions according to an embodiment of the present application.

[0011] Figure 4 , Figure 5 , Figure 6 and Figure 7 Schematic diagram showing details of a transfer station or router according to various embodiments of the present application.

[0012] Figure 8 A block diagram of a supercomputer system according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0013] In the following description, the terms "first", "second", and "third" in the following description are only used for the purpose of distinguishing different elements, and do not mean the order of the elements. For example, a first device and a second device only mean that these devices may have the same structure but are different devices. In addition, the term "data" in the following embodiments may refer to any signal or any information that can be used in an electronic device.

[0014] In addition, the following embodiments are all described by taking the NoC system as an example. However, the concepts disclosed in the following embodiments can be applied to any other computing system. Figure 1 A schematic diagram showing a portion of a computing system according to an embodiment of the present application. Figure 2 Show Figure 1 A more complete schematic diagram of the computing system is shown in the reference Figure 1 At the same time reference Figure 2 , in order to more clearly understand the concept of this application.

[0015] like Figure 1 As shown, the computing system 100 includes a plurality of routers (four routers RO_1, RO_2, RO_3 and RO_4 are symbolically identified for ease of description). The routers are configured to transmit data to nodes (four nodes NO_1, NO_2, NO_3 and NO_4 are symbolically identified for ease of description). The node can be any electronic device that can receive data, such as a CPU or a GPU.

[0016] Please note that Figure 1In the embodiment of FIG. 1 , each router corresponds to a node. For example, router RO_1 is used to output data to node NO_1, and router RO_2 is used to output data to node NO_2. However, a router may correspond to multiple nodes. Moreover, a router may output data to any connected node according to the destination information. Moreover, in FIG. Figure 1 In an embodiment, a first path is provided between adjacent routers. For example, a first path is provided between router RO_1 and router RO_2, and a first path is provided between router RO_1 and router RO_3. Note that adjacent routers mean that no other routers are required when transmitting data between two adjacent routers. For example, routers RO_1 and RO_2 are adjacent routers, routers RO_3 and RO_4 are adjacent routers, but routers RO_1 and RO_3 are not adjacent routers. In one embodiment, the transmission time of all first paths is the same. In another embodiment, the transmission time of different first paths is different. For example, the transmission time of the first path between routers RO_1 and RO_2 is different from the transmission time of the first path between routers RO_3 and RO_4.

[0017] Figure 2 Show Figure 1 A schematic diagram of a more complete structure of the computing system shown. Note that for ease of illustration, Figure 1 Some symbols in Figure 2 Not shown, and Figure 1 Some of the components in Figure 2 In addition to Figure 1 In addition to the components shown, Figure 2 The embodiment also includes multiple transfer stations. Four transfer stations TR_1 to TR_4 are used as an example for description. Transfer stations TR_1 to TR_4 are configured to transmit data to nodes through routers. For example, transfer station TR_1 can transmit data to node NO_5 or router RO_5.

[0018] A second path is provided between adjacent transfer stations. For example, a second path is provided between transfer station TR_1 and transfer station TR_2, and a second path is provided between transfer station TR_2 and transfer station TR_3. Note that adjacent transfer stations mean that no other transfer stations are required when transmitting data between two adjacent transfer stations. Figure 2 In one embodiment, the transfer stations TR_1 to TR_4 are adjacent transfer stations. In one embodiment, the transmission time of all second paths is the same. In another embodiment, the transmission time of different second paths is different. For example, the transmission time of the second path between transfer stations TR_1 and TR_2 is different from the transmission time of the second path between transfer stations TR_3 and TR_4.

[0019] According to the locations of the first target device and the second target device, the computing system 100 can adaptively select the first path (second mode), the second path (third mode), and a combination with a shorter time interval (first mode) to send data from the first target device to the second target device. The first device and the second device can be one of a node or a router.

[0020] In one embodiment, the computing system operates in a first mode, which transmits data from a first target device to a second target device via a first path and a second path during a first time interval. The second path can replace multiple first paths. For example, the second path between transfer stations TR_1 and TR_2 can replace multiple first paths of a router between transfer stations TR_1 and TR_2. Therefore, when the second path is used instead of the first path, fewer routers are required.

[0021] Additionally, the computing system may operate in a second mode that transmits data from the first target device to the second target device via the first path without using the second path during a second time interval that is shorter than the first time interval.

[0022] The first target device and the second target device may be routers or nodes. Figure 2 In the first mode, the computing system 100 can transmit data from the router RO_5 to the router RO_6 through the first route 101 during the first time interval. For the first route 101, the data must be transmitted from the router RO_5 to the router RO_6 via two routers and two transfer stations TR_2 and TR_3.

[0023] Alternatively, computing system 100 may transmit data from router RO_5 to router RO_6 during a second time interval in a second mode through second route 102 that uses the first path but not the second path. For second route 102, data must be transmitted from router RO_5 to router RO_6 through eight routers.

[0024] In this embodiment, since the number of intermediate stations (routers or transfer stations) in the first route 101 is less than that in the second route 102, the first time interval of the first route 101 is shorter than the second time interval of the second route 102. Therefore, the computing system will select the first route 101 to transmit data from the router RO_5 to the router RO_6.

[0025] However, for other embodiments, if the first route has more intermediate stations than the second route, the first time interval of the first route is greater than the second time interval of the second route. In this case, the second route is selected to transmit data.

[0026] In one embodiment, the computing system 100 may also selectively operate in a third mode, which transmits data from the first target device to the second target device via the second path, but not via the first path. For example, since the node NO_5 and the node NO_6 are adjacent to the transfer station TR_1 and the transfer station TR_3, respectively, the computing system 100 transmits data from the node NO_5 to the node NO_6 only via the second path between the transfer station TR_1 and the transfer station TR_3, but not via any of the first paths.

[0027] In one embodiment, the router is divided into a plurality of system groups SG_1…SG_4 according to its location. The location may refer to the physical location or access address of the router. It should be understood that the number of system groups may be any other number and is not limited to 4. In one embodiment, the computing system 100 operates in one of the first mode, the second mode, and the third mode according to the system group. For example, if the first target device and the second target device are both in the system group SG_1, the computing system 100 operates in the second mode because no suitable second path can be used. For another example, if the first target device is located in the system group SG_1 and the second target device is located in the system group SG_4, the data will first be transmitted to the transfer station TR_1 and then transmitted to the transfer station TR_3 via the second path. Afterwards, the data is transmitted from the transfer station TR_3 to the second target device.

[0028] Figure 3 Schematic diagram of the paths of the transfer station or router corresponding to different directions in different embodiments of the present invention. Please note that the path structure of the transfer station TS and the router RO is not limited to Figure 3 The structure shown. Figure 3 In the embodiment of the present invention, the transfer station TS and the router RO are shown as a 3D diagram. However, it is not limited to that the transfer station TS and the router RO are arranged in different planes. In one embodiment, the transfer station TS and the router RO can be arranged in the same plane.

[0029] like Figure 3As shown, the first paths P_12, P_13, P_14 and P_11 can have different directions: east, south, west and north. Please note that the directions of "east, south, west and north" mentioned here are only used to indicate different directions and do not represent geographical directions. Accordingly, the first paths P_12, P_13, P_14 and P_11 are connected to / from routers in the east, south, west and north (hereinafter referred to as E router, S router, W router and N router) respectively. Following the same rule, the second paths P_22, P_23, P_24 and P_21 are connected to / from transfer stations in the east, south, west and north (hereinafter referred to as E TS, S TS, W TS and N TS) respectively. In addition, paths P_31 and P_32 are provided between the transfer station TS and the router RO so that data can be transmitted between the transfer station TS and the router RO. In addition, the router RO can also include paths connected to / from nodes 0-3.

[0030] Figure 4 , Figure 5 , Figure 6 and Figure 7 Schematic diagram showing details of a transfer station or router according to different embodiments of the present application. Figure 4 In the embodiment of the invention, the transfer station TS includes a device 401, and the device 401 includes at least one crossbar switch and at least one buffer. The crossbar switch is arranged between the transfer station TS and at least one router, and is configured to change its switch state to transmit data from the router to the transfer station N TS, E TS, W TS or S TS. The buffer is arranged between the transfer station TS and the router, and is configured to buffer data from the router to the transfer station N TS, E TS, W TS or S TS.

[0031] The arbitrator 403 is disposed between the transfer station N TS, E TS, W TS or S TS and the router, and is used to distribute data from the transfer station N TS, E TS, W TS or S TS to the router. For example, the arbitrator 403 is a time-division multiplexing arbitrator. Figure 4 In the embodiment, the second path between the transfer station TS and N TS, E TS, W TS or S TS is a 4n-bit trunk bus, and the path between the transfer station TS and the router is also a 4n-bit trunk bus.

[0032] Different types of paths can have different structures or bandwidths. Figure 5In an embodiment of the present invention, the computing system transmits data from one of the nodes to another of the nodes via at least one physical channel (PC). For example, data from node N0 to node N1 is transmitted to router RO via physical channel N0 PC, and then transmitted from router RO to node N1 via physical channel N1 PC. A physical channel refers to a real physical channel, such as a transmission line or a bus.

[0033] In addition, Figure 5 In an embodiment, the computing system transmits data from one of the routers to another router in the routers via at least one virtual channel. For example, data from router E to router W is transmitted to router RO via virtual channel E VC, and then transmitted from router RO to router W via virtual channel W VC. A virtual channel (VC) is a communication track between two devices, providing the bandwidth required for a virtual connection. In an embodiment, the bandwidth of a virtual channel is less than the bandwidth of a physical channel, so more virtual channels can be combined to form a single physical channel.

[0034] Figure 6 A schematic diagram showing how a computing system according to an embodiment of the present application transmits data. In such an embodiment, the transfer station TS includes a plurality of output buffers, such as a W buffer, an N buffer, an E buffer, and an S buffer. In addition, the router RO includes multiplexers 601, 603, an arbiter 605, and a router buffer. Data is transmitted in a packet format, such as Figure 6 Packets PA_1 and PA_2 in .

[0035] The symbol in the packet indicates the destination to which the packet should be transmitted. For example, suppose packet PA_1 is to be transmitted to Figure 2 The system group SG_1 in the Figure 2 In the system group SG_3. Figure 6 In the embodiment of FIG. 6 , the arbiter 605 is a TDM arbiter, so the router buffer stores packets from the E, W, N, S routers, and then the multiplexer 601 selectively outputs these packets at times T0, T2, T4 and T6. Figure 6 In the embodiment of , a node buffer is also included. In this case, the node buffer stores packets from nodes 0, 1, 2, and 3, and then the multiplexer 603 selectively outputs these packets at times T1, T3, T5, and T7. Figure 6 In the embodiment, the order of time T1-T7 is T1, T2, T3, T4, T5, T6, T7.

[0036] The data output by the arbiter 605 is stored in the output buffer corresponding to the destination of the data. For example, the destination of packet PA_1 is system group SG1, and it should be transmitted by transfer station E TS, so at time T0, packet PA_1 is stored in the output buffer E Buffer corresponding to transfer station E TS. For another example, the destination of packet PA_2 is system group SG3, and it should be transmitted by transfer station N TS, so at time T1, packet PA_2 is stored in the output buffer N Buffer corresponding to transfer station N TS. Figure 6 The paths in can have different bandwidths. For example, the path between routers (e.g., the first path) can be an n-bit local bus, and the path of multiplexers 601, 603 and arbiter 605 can be a 4n-bit backbone bus. In addition, the path between transfer stations (e.g., the second path) can be a 4n-bit backbone bus. Accordingly, in Figure 6 In an embodiment of the present invention, a bandwidth of the second path between adjacent transfer stations is greater than a bandwidth of the first path between adjacent routers.

[0037] exist Figure 6 In an embodiment, data is transferred from a router or node to a transfer station. Figure 7 In the embodiment, the data transmission direction is Figure 6 In other words, data is transferred from a transfer station to a router or node. Figure 7 In the embodiment of the present invention, the router RO further includes multiplexers 601, 603 and an arbiter 605, and the transfer station TS includes a TDM arbiter 701. The TDM arbiter 701 selectively outputs data from the transfer stations WTS, N TS, N TS, E TS and S TS to the arbiter 605. The arbiter 605 selectively outputs data to the multiplexer 601 or 603. The multiplexer 601 selectively outputs data to the E router, W router, N router or S router. The multiplexer 603 selectively outputs data to the node 0, node 1, node 2 or node 3.

[0038] In the above embodiment, the computing system is a NoC system. Therefore, the router, the transfer station, the first path, and the second path are provided in a single chip. However, the computing system disclosed in the present application can be applied to a multi-chip system, which can be a supercomputer system.

[0039] Figure 8 FIG. 8 is a block diagram of a supercomputer system 800 according to an embodiment of the present application. Figure 8 As shown, multiple transfer stations TS are arranged in different chips ch1-ch15. In one embodiment, chips ch1-ch15 are set in at least two computers. The transfer stations TS can be connected by different cables, such as Figure 8 Cables Ca_1…Ca_4 in. Figure 8 The operation is similar to the above embodiment. Data can be transmitted between different transfer stations, and then transmitted to routers in different chips or to nodes connected to the transfer station TS.

[0040] In view of the above embodiments, the computing system provided by the present application can be summarized as a computing system including multiple routers, multiple transfer stations, multiple first paths and multiple second paths. The router is configured to transmit data to a node. The transfer station is configured to transmit data to a node or a router. A first path is provided between adjacent routers. The second path is provided between adjacent transfer stations. The computing system selectively operates in a first mode, which transmits data from a first target device to a second target device via the first path and the second path during a first time interval. The first target device is one of a router or a node, and the second target device is one of a router or a node.

[0041] In the above-described embodiments, a computing system including a router for transmitting data to a node is applied to a single-chip system (NoC system) or a multi-chip system. However, the concept disclosed in the present invention can be implemented in any data transmission system instead of a computing system including a router for transmitting data to a node. In this case, the router is replaced by a "first transmission device", and the transfer station is replaced by a "second transmission device". In addition, the node is replaced by a "destination device".

[0042] In view of the above-described embodiments, transmission delay can be reduced because data can be transmitted through a shortcut formed by a transfer station, rather than being limited to being transmitted through a path formed only by a router.

[0043] Those skilled in the art will readily observe that various modifications and changes can be made to the apparatus and method while retaining the teachings of the present invention.Therefore, the above disclosure should be interpreted as being limited only by the scope and limits of the appended claims.

Claims

1. A computing system comprising: Multiple routers for transmitting data to multiple nodes; a plurality of transfer stations, configured to transfer data to the plurality of nodes via the plurality of routers; A plurality of first paths are arranged between adjacent routers; A plurality of second paths are arranged between adjacent transfer stations; wherein the computing system operates in a first mode to transmit the data from the first target device to the second target device via the plurality of first paths and the plurality of second paths during a first time interval; The first target device is one of the multiple routers or the multiple nodes, and the second target device is one of the multiple routers or the multiple nodes.

2. The computing system of claim 1, wherein: The computing system is further operated in a second mode, wherein the computing system transmits the data from the first target device to the second target device via the plurality of first paths and not via the plurality of second paths during a second time interval; And the first time interval is greater than the second time interval.

3. The computing system of claim 1, wherein: The computing system is further selectively operable in a third mode in which the computing system transmits the data from the first target device to the second target device via the plurality of second paths, but not via the plurality of first paths.

4. The computing system of claim 1, wherein: The plurality of bandwidths of the plurality of second paths is greater than the plurality of bandwidths of the plurality of first paths.

5. The computing system of claim 1, wherein: The plurality of routers are divided into a plurality of system groups according to a plurality of locations of the plurality of routers or the plurality of nodes, wherein the computing system operates in the first mode or the second mode according to the system group where the first target device and the second target device are located.

6. The computing system of claim 1, further comprising: at least one cross-point switch, disposed between the plurality of transfer stations and the plurality of routers, for transmitting the data from the plurality of routers to the plurality of transfer stations; as well as At least one buffer is disposed between the plurality of transfer stations and the plurality of routers, and is used for buffering the data from the plurality of routers to the plurality of transfer stations.

7. The computing system of claim 1, wherein: The computing system further comprises: At least one arbitrator is disposed between the plurality of transfer stations and the plurality of routers, and is used to distribute the data from the plurality of transfer stations to the plurality of routers.

8. The computing system of claim 1, wherein: The computing system transmits the data from one of the nodes to another node via at least one physical channel, and transmits the data from one of the routers to another router via at least one virtual channel.

9. The computing system of claim 1, wherein: The computing system is a network-on-a-chip system, wherein the plurality of routers, the plurality of transfer stations, the plurality of first paths, and the plurality of second paths are arranged in a single chip.

10. The computing system of claim 1, wherein: The multiple transfer stations are arranged in multiple different chips.

11. The computing system of claim 10, wherein: The computing system is a supercomputer system, and the plurality of chips are arranged in at least two computers.

12. A data transmission system, comprising: A plurality of first transmission devices, used to transmit data to a plurality of destination devices; A plurality of second transmission devices, used to transmit the data to the plurality of destination devices through the plurality of first transmission devices; A plurality of first paths are arranged between adjacent first transmission devices; A plurality of second paths, arranged between adjacent second transmission devices; wherein the data transmission system operates in a first mode, which transmits the data from a first target device to a second target device via the plurality of first paths and the plurality of second paths during a first time interval; The first target device is one of the multiple first transmission devices or one of the multiple destination devices, and the second target device is one of the multiple first transmission devices or one of the multiple destination devices.

13. The data transmission system according to claim 12, characterized in that: The data transmission system operates in a second mode, and the data transmission system also transmits the data from the first target device to the second target device via the multiple first paths during a second time interval, but not via the multiple second paths; and the first time interval is greater than the second time interval.

14. The data transmission system according to claim 12, characterized in that: The data transmission system is further selectively operable in a third mode, wherein the data transmission system transmits the data from the first target device to the second target device via the plurality of second paths, but not via the plurality of first paths.

15. The data transmission system according to claim 12, characterized in that: The plurality of bandwidths of the plurality of second paths is greater than the plurality of bandwidths of the plurality of first paths.

16. The data transmission system according to claim 12, characterized in that: According to the multiple positions of the multiple first transmission devices, the multiple first transmission devices are divided into multiple system groups, wherein the data transmission system operates in the first mode or the second mode according to the system groups where the first target device and the second target device are located.

17. The data transmission system according to claim 12, characterized in that: The data transmission system further comprises: at least one cross-point switch, provided between the plurality of second transmission devices and the plurality of first transmission devices, for transmitting the data from the plurality of first transmission devices to the plurality of second transmission devices; and At least one buffer is arranged between the plurality of second transmission devices and the plurality of first transmission devices, and is used for buffering the data from the plurality of first transmission devices to the plurality of second transmission devices.

18. The data transmission system according to claim 12, characterized in that: The data transmission system further comprises: At least one arbitrator is disposed between the plurality of second transmission devices and the plurality of first transmission devices, and is used to distribute the data from the plurality of second transmission devices to the plurality of first transmission devices.

19. The data transmission system according to claim 12, characterized in that: The data transmission system transmits the data from one destination device among the multiple destination devices to another destination device via at least one physical channel, and transmits the data from at least one transmission device among the multiple second transmission devices to another second transmission device via at least one virtual channel.

20. The data transmission system according to claim 12, characterized in that: The data transmission system is a network-on-a-chip system, wherein the plurality of routers, the plurality of transfer stations, the plurality of first paths and the plurality of second paths are arranged in a single chip.

21. The data transmission system according to claim 12, characterized in that: The plurality of second transmission devices are arranged in a plurality of different chips.

22. The data transmission system according to claim 21, characterized in that The data transmission system is a supercomputer system, and the plurality of chips are arranged in at least two computers.

Images