Dynamic bypass fault-tolerant router in network-on-chip and routing method thereof

Abstract

The invention discloses a dynamic bypass fault-tolerant router in a network-on-chip. The dynamic bypass fault-tolerant router comprises input and output ports, an input buffer area, a crossbar switch, a routing calculation unit, a self-checking circuit and a bypass channel. Key features comprise a fault information unit, a bypass selector, a single flit buffer area and an arbitration unit. The router detects the self-fault through the self-checking circuit, and the fault information unit stores the information. And if the router fails, the bypass selector transmits the data packet through the bypass channel and bypasses the failed router. If the router is healthy, the router exchanges fault information with an adjacent router, the fault information is used for routing decision of a data packet after integration, and meanwhile, the router can select a routing path with smaller congestion in an effective direction according to the flow distribution condition in a network. According to the method, the area overhead is relatively low, the routing hop count of the data packet is remarkably reduced, the transmission delay of the data packet is reduced while the high reliability is maintained, and the network throughput is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fault tolerance in integrated circuit chip design, and particularly relates to a dynamic bypass fault-tolerant router and a routing method thereof for a network-on-chip in a 2D Mesh. Background Art

[0002] With the continuous development of integrated circuit manufacturing technology, the number of cores that can be accommodated on a single chip can reach hundreds or even thousands. The traditional communication paradigm based on the bus structure can no longer meet the growing communication needs, and the network-on-chip (NoC) has emerged. The NoC has advantages such as low communication latency and high scalability. The network topology has a great impact on the cost and performance of the NoC. Due to its simple structure, easy implementation, high scalability and other characteristics, 2D Mesh has received extensive attention from researchers. In the Mesh-based NoC, each router is connected to four neighbor routers in the east, west, south, and north through links, and the processing unit is connected to the corresponding router through the network interface NI. With the continuous improvement of chip integration, the failure rate of the chip will increase during the manufacturing process. Once a component in the NoC fails, it will affect the normal routing process, resulting in a rapid decline in reliability, and the data packet cannot complete normal data transmission. The faults in the NoC are generally divided into permanent faults and transient faults. Permanent faults refer to faults that cannot be recovered for a long time, so it is very important to handle permanent faults. Such faults are generally formed due to manufacturing defects and component aging. Transient faults refer to faults that can recover by themselves after several clock cycles. Such faults are generally formed due to the environment such as electromagnetic interference.

[0003] Many scholars have provided different types of fault-tolerant solutions for permanent faults in the NoC. In the currently known fault-tolerant solutions, most of them achieve fault tolerance by bypassing the faulty router or link. Doing so often involves a relatively complex routing algorithm, and it is very likely to form hotspots near the fault area, accelerating the formation of congested nodes, and further increasing the data packet transmission delay. If the fault occurs in the processing unit, although the local processing unit cannot communicate, it will not affect the normal communication of other processing units. If the fault occurs in the router and link, it will directly affect the communication of other routers, and the data packet cannot be transmitted through the faulty router or link.

[0004] The fault-tolerant scheme based on detour routing not only has many restrictions on the number and location of faulty routers, but also cannot solve the isolation area problem. When the source router and the destination router are located in the upper and lower subnets that are isolated respectively, the detour-based fault-tolerant scheme cannot bypass the faulty router to reach the other subnet, resulting in serious packet loss. The bypass-based fault-tolerant scheme can regard the faulty router as a wire connecting healthy routers, and the packets can bypass the faulty router through the bypass channel, which can well solve the problem of the isolation area. However, there are many problems in the existing bypass fault-tolerant schemes. The bypass channel is only enabled in the faulty router. When the network is fault-free, the bypass channel is in an idle state, causing waste of hardware resources. At the same time, the influence of the diagonal position and long wires on the clock cycle is not fully considered. Summary of the Invention

[0005] The present invention is to avoid the deficiencies in the above-mentioned prior art, and provides a dynamic bypass fault-tolerant router in a network-on-chip, in order to be able to flexibly open the bypass channel for data transmission according to the address information and fault information of the received packets, significantly reduce the packet routing hop count, and prevent the possible interleaving of packets at different positions, so as to reduce the delay of packets in the network with a lower area overhead and improve the throughput of the network.

[0006] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0007] A dynamic bypass fault-tolerant router in a network-on-chip according to the present invention is applied to a two-dimensional mesh structure of a network-on-chip. The mesh structure is composed of m×n routers, and each router corresponds to a processing unit respectively. Any processing unit sends and receives packets to and from other processing units through the router connected to itself in the network-on-chip. Each router includes: input and output ports in the east, west, south, and north directions, an input buffer, a crossbar switch, a routing calculation unit, a self-checking circuit, and a bypass channel. The feature is that in each router, there are also provided: a fault information unit, four bypass selectors, a single-flit buffer, and an arbitration unit, thus constituting a dynamic bypass fault-tolerant router;

[0008] When the network-on-chip is running, the router at the i-th row and j-th column currently detects the fault conditions of its own input buffer and crossbar switch through its own self-checking circuit, and the fault information unit of itself saves the fault conditions as local fault information;

[0009] If the local fault information indicates that the router at the i-th row and j-th column is a faulty router, the fault information units of the router at the i-th row and j-th column respectively send control signals to the bypass selectors at the input ports in the east, west, south, and north directions of itself. The four-direction bypass selectors receive the control signals and select the bypass channel from the channels and bypass channels inside their own routers to transmit the subsequent arriving data packets to the next-hop router, thereby bypassing the faulty router to transmit the data packets;

[0010] If the local fault information indicates that the router at the i-th row and j-th column is a healthy router, the fault information units of the router at the i-th row and j-th column receive the local fault information of themselves sent by the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column. After integrating with their own local fault information, they respectively send the integrated local fault information to the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column. At the same time, the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column receive the local fault information sent by their adjacent routers, and after integrating with their own local fault information, they respectively send the integrated local fault information to their adjacent routers. Thus, after several clock cycles, the fault information units in the router at the i-th row and j-th column obtain the local fault information of all routers in the i-th row and j-th column for the subsequent channel selection of data packets;

[0011] The routing calculation unit of the router at the i-th row and j-th column judges the relative position relationship between its own router and the destination router;

[0012] If the destination router is a router at the same row and the same column position, the bypass selector of the downstream router of the router at the i-th row and j-th column selects the single-flit buffer in the bypass channel to cache the arriving data packets. If there are data packets stored in the channel inside the router at the i-th row and j-th column, the arbitration unit at the output port arbitrates the data packets in the bypass channel and the internal channel: only after the previous data packet is sent out, the arbitration unit receives the next data packet with the highest priority for transmission, where the priority of the data packet is proportional to the waiting time;

[0013] If the destination router is not a router in the same row and column position, it means that the data packet cannot reach the destination router through the direct routing path. Then, according to the local fault information of all routers in the i-th row and the j-th column, the corresponding steering router is selected using the routing method. Thus, the bypass selector of the selected steering router uses the channels inside its own router to transmit the received data packet, and the data packet is steered at the crossbar of the selected steering router. The routing calculation unit of the selected steering router determines the relative position relationship between its own router and the destination router, and thus selects to continue transmitting the data packet through the internal channel or the bypass channel according to the relative position relationship until the destination router is reached.

[0014] The dynamic bypass fault-tolerant router in the network-on-chip according to the present invention is also characterized in that the routing method includes the following steps:

[0015] Step a: The routing calculation unit in the router at the i-th row and the j-th column determines whether the router at the i-th row and the j-th column and the destination router are in the same row and column position according to its own coordinate information and the coordinate information of the destination router. If so, the data packet is directly forwarded to the destination router through the bypass channel of the downstream router of the router at the i-th row and the j-th column, and the routing ends. Otherwise, calculate the remaining hops in the valid direction and execute step b;

[0016] Step b: If the remaining hops in the valid direction between the router at the i-th row and the j-th column and the destination router are 1, and the next-hop router in the valid direction is a faulty router, it means that the data packet cannot reach the destination router through the shortest path, and execute step c; otherwise, execute step d;

[0017] Step c: The router at the i-th row and the j-th column selects a non-shortest routing path with a lower congestion degree from the valid directions according to the current congestion information, so as to forward the data packet to the destination router and end the routing;

[0018] Step d: The routing calculation unit in the router where the data packet is located obtains two routing paths according to the routing selection order of the valid direction, and calculates the coordinate information of a steering router on each of the two routing paths. Thus, according to the local fault information of the routers in the same row and column stored in the fault information unit of its own router, it is determined whether the steering routers on the two routing paths are all healthy routers; if so, execute step e, otherwise, it means that there is at least one faulty router among the steering routers, and execute step f;

[0019] Step e: The router where the data packet is located forwards the data packet to the steering router in the less congested effective direction through the bypass channel. The bypass selector at the corresponding port of the steering router performs a steering operation on the data packet using the crossbar switch of the steering router according to the address information of the received data packet, and then sends the data packet to the destination router at the same row and column position through the bypass channel, and the routing ends.

[0020] Step f: If both steering routers are faulty routers, it means that the data packet cannot perform a steering operation at the crossbar switch of the steering routers on the two routing paths, and step g is executed. Otherwise, the router where the data packet is located forwards the data packet to the healthy steering router through the bypass channel, and forwards the data packet to the destination router through the bypass channel of the healthy steering router, and then the routing ends.

[0021] Step g: The router where the data packet is located selects the upstream healthy router of the faulty steering router in the less congested effective direction as the new steering router according to the congestion information, and after sending the data packet to the new steering router, step d is executed.

[0022] Compared with the prior art, the effects of the present invention are as follows:

[0023] 1. The present invention breaks the limitation on the use of the bypass in the previous bypass fault tolerance scheme. The use of the bypass is no longer limited to the faulty router. The bypass channel of the healthy router can also be used for data transmission when the bypass use conditions are met. While maintaining high reliability, compared with the traditional fault tolerance scheme, the routing hops of the data packet are significantly reduced.

[0024] 2. The present invention fully considers the traffic distribution in the on-chip network. By using Credit to detect the remaining buffer size in the downstream router in real time, the data packet is sent in the effective direction with a larger remaining buffer, reducing the formation of congested nodes and reducing the data packet transmission delay.

[0025] 3. The present invention effectively prevents the packet interleaving situation that may be brought about by using the bypass. By adding a bypass selector at the input port to distinguish bypass packets and ordinary packets according to the address information to prevent packet interleaving in the buffer, and adding an arbitration unit at the output port to prevent packet interleaving at the output port, thus avoiding the situation that the destination router receives incorrect packets. Description of the Drawings

[0026] Figure 1 is a schematic diagram of the overall framework of the router in the present invention;

[0027] Figure 2 is a structural diagram of the bypass selector in the present invention;

[0028] Figure 3 It is a schematic diagram of the bypass structure in the present invention;

[0029] Figure 4 It is an example diagram of routing in the present invention. Detailed implementation manners

[0030] In this example, a dynamic bypass fault-tolerant router in a network-on-chip is applied to a two-dimensional mesh structure of the network-on-chip. The mesh structure is composed of m×n routers, and each router corresponds to a processing unit respectively. Any processing unit sends and receives data packets to other processing units through the router connected to itself in the network-on-chip. The overall framework diagram of the router is as Figure 1 shown. Each router includes input and output ports, an input buffer, a crossbar switch, a routing calculation unit, a self-checking circuit, and a bypass channel in the four directions of east, west, south, and north. It is characterized in that a fault information unit, four bypass selectors, a single-flit buffer, and an arbitration unit are further arranged in each router, thus constituting a dynamic bypass fault-tolerant router;

[0031] When the network-on-chip is running, the router at the i-th row and j-th column currently detects the fault conditions of its own input buffer and crossbar switch through its own self-checking circuit, and the fault information unit of itself saves the fault conditions as local fault information;

[0032] The structure diagram of the bypass selector is as Figure 2 shown, which includes a comparator and a demultiplexer for path selection. The bypass selector distinguishes bypass data packets and ordinary data packets according to the address information of the input data packet flit and the local fault information. The bypass data packets select the bypass channel, and the ordinary data packets select the channel inside the router. If the local fault information indicates that the router at the i-th row and j-th column is a faulty router, the fault information unit of the router at the i-th row and j-th column respectively sends control signals to the bypass selectors at the input ports in the four directions of east, west, south, and north of itself. The four-direction bypass selectors receive the control signals and select the bypass channel from the channels inside their own routers and the bypass channels to transmit the subsequent arriving data packets to the next-hop router; thus bypassing the faulty router to transmit data packets;

[0033] Figure 3 The following is a schematic diagram of the bypass structure in the present invention. The red nodes represent faulty routers, the white nodes represent healthy routers, and the green dotted lines represent bypass channels. The faulty routers are connected to healthy routers through the bypass channels. The faulty routers can be regarded as a wire connecting healthy routers in the horizontal and vertical directions. Data packets can directly bypass the faulty routers, and straight-through data packets can perform data transmission through the bypass channels of healthy routers.

[0034] If the local fault information indicates that the router at the i-th row and j-th column is a healthy router, the fault information unit of the router at the i-th row and j-th column respectively receives the local fault information of the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column. After integrating with its own local fault information, it respectively sends the integrated local fault information to the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column. At the same time, the routers at the (i + 1)-th row and j-th column, (i - 1)-th row and j-th column, i-th row and (j + 1)-th column, and i-th row and (j - 1)-th column receive the local fault information sent by their adjacent routers, and after integrating with their own local fault information, they respectively send the integrated local fault information to their adjacent routers. Thus, after several clock cycles, the fault information unit in the router at the i-th row and j-th column obtains the local fault information of all routers in the i-th row and j-th column, which is used for the channel selection of subsequent data packets;

[0035] In the present invention, the network-on-chip uses the Double-Y network and the mad-y turn model. The network is divided into two subnets, +X and -X. Each subnet has half of the channels in the Y direction. Packets going east pass through the +X subnet (corresponding to virtual channel 1 in the Y direction), and packets going west pass through the -X subnet (corresponding to virtual channel 2 in the Y direction), ensuring that only the shortest path is used for transmission when the shortest path exists. Since the next-hop address is always selected in the valid direction and 180-degree turns are prohibited, deadlocks and livelocks are effectively prevented.

[0036] The routing calculation unit of the router at the i-th row and j-th column determines the relative position relationship between its own router and the destination router;

[0037] If the destination router is a router at the same row and the same column position, the bypass selector of the downstream router of the router at the i-th row and j-th column selects the single-flit buffer in the bypass channel to cache the arriving data packets. If there are data packets stored in the channels inside the router at the i-th row and j-th column, the arbitration unit at the output port arbitrates the data packets in the bypass channel and the internal channel: only after the previous data packet is sent, the arbitration unit receives the next data packet with the highest priority for transmission, where the priority of the data packet is proportional to the waiting time;

[0038] If the destination router is not a router in the same row and column position, it means that the data packet cannot reach the destination router through the direct routing path. Then, based on the local fault information of all routers in the i-th row and j-th column, a corresponding steering router is selected using the routing method. Thus, the bypass selector of the selected steering router uses the internal channels of its own router to transmit the received data packet, and the data packet is steered at the crossbar switch of the selected steering router. The routing calculation unit of the selected steering router determines the relative position relationship between its own router and the destination router, and then selects to continue transmitting the data packet through the internal channel or the bypass channel according to the relative position relationship until the destination router is reached.

[0039] In this example, the routing method includes the following steps:

[0040] Step a: The routing calculation unit in the router at the i-th row and j-th column determines whether the router at the i-th row and j-th column and the destination router are in the same row and column position according to its own coordinate information and the coordinate information of the destination router. If so, the data packet is directly forwarded to the destination router through the bypass channel of the downstream router of the router at the i-th row and j-th column, and the routing ends. Otherwise, calculate the remaining hops in the effective direction and execute step b;

[0041] Step b: If the remaining hops of the router at the i-th row and j-th column and the destination router in the effective direction are 1, and the next-hop router in the effective direction is a faulty router, it means that the data packet cannot reach the destination router through the shortest path, and execute step c; otherwise, execute step d;

[0042] Step c: The router obtains congestion information by detecting the size of the remaining buffer in the next-hop router in the effective direction in real time through Credit. The router at the i-th row and j-th column selects a non-shortest routing path with a lower congestion degree from the effective directions according to the current congestion information, so as to forward the data packet to the destination router and end the routing;

[0043] Step d: The routing calculation unit in the router where the data packet is located obtains two routing paths according to the routing selection order of the effective direction, and calculates the coordinate information of a steering router on each of the two routing paths. Then, according to the local fault information of the routers in the same row and column stored in the fault information unit of its own router, it determines whether the steering routers on the two routing paths are all healthy routers; if so, execute step e, otherwise, it means that there is at least one faulty router among the steering routers, and execute step f;

[0044] Step e: The router where the data packet is located forwards the data packet to the steering router in the less congested effective direction through the bypass channel. The bypass selector at the corresponding port of the steering router performs a steering operation on the data packet using the crossbar switch of the steering router, and then sends the data packet to the destination router at the same row and column position through the bypass channel, and the routing ends.

[0045] Step f: If both steering routers are faulty routers, it means that the data packet cannot perform a steering operation at the crossbar switch of the steering routers on the two routing paths, and step g is executed. Otherwise, the router where the data packet is located forwards the data packet to the healthy steering router through the bypass channel, and then forwards the data packet to the destination router through the bypass channel of the healthy steering router, and the routing ends.

[0046] Step g: The router where the data packet is located selects the upstream healthy router of the faulty steering router in the less congested effective direction as the new steering router according to the congestion information, sends the data packet to the new steering router, and then executes step d.

[0047] Figure 4 The following shows an example of data packet routing of the present invention. S represents the source router, D represents the destination router, T represents the steering router, the red nodes are faulty routers, and the green nodes represent healthy routers bypassed using the bypass channel. Define the coordinates of the bottom left router as (0, 0). In this example, the coordinates of the steering routers are (0, 3) and (3, 1). According to the number of faulty steering routers, the routing is divided into three cases: no fault case, single steering router fault, and double steering router fault, corresponding to Figure 4 parts (a), (b), and (c) in. The present invention significantly reduces the routing hops of data packets by flexibly enabling the bypass channel of the router to transmit data packets, thereby significantly reducing the data packet transmission delay.

[0048] Figure 4 Part (a) corresponds to the case where the steering router has no fault. Among them, the two shortest paths passing through the steering router can be used to forward the data packet. The data packet at the source router selects the less congested effective direction to forward the data packet to the steering router (0, 3). The steering router (0, 3) and the destination router (3, 3) are located in the same row and column, and can be directly forwarded to the destination router through the bypass channel subsequently.

[0049] Figure 4 Part (b) corresponds to the case of a single steering router fault. Among them, the steering router (0, 3) in one effective direction is faulty, and the data packet selects the effective direction corresponding to the healthy steering router (3, 1) to forward the data packet.

[0050] Figure 4 Part (c) corresponds to the situation where both steering routers fail. When both steering routers fail, the data packet cannot complete the turning operation at the crossbar of the failed steering router. The data packet needs to select a new steering router according to the above routing method. The data packet selects the previous-hop router of the failed steering router (0, 3) in the direction with less congestion degree as the new steering router according to the congestion information, that is, router (0, 2). At this time, continue to judge the number of failed steering routers at the position of router (0, 2), and repeat the above process subsequently until the data packet reaches the destination router.

Claims

1. A dynamic bypass fault-tolerant router in a network on chip is applied to a two-dimensional mesh structure of the network on chip, wherein the mesh structure is composed of m×n routers, and each router corresponds to a processing unit. Any processing unit sends and receives data packets to other processing units through the router connected to itself in the network on chip. Each router comprises: input and output ports in four directions of east, west, south and north, an input buffer, a cross switch, a routing calculation unit, a self-checking circuit, and a bypass channel, characterized in that: Each router is also provided with: a fault information unit, four bypass selectors, a single flit buffer, and an arbitration unit, thereby forming a dynamic bypass fault-tolerant router; When the on-chip network is running, the current router in the i-th row and j-th column detects the fault conditions of its own input buffer and crossbar switch through its own self-checking circuit, and saves the fault conditions as local fault information through its own fault information unit; If the local fault information indicates that the router in the i-th row and j-th column is a faulty router, the fault information unit of the i-th row and j-th column router sends control signals to the bypass selectors at the input ports in the east, west, south, and north directions of the router, respectively. The bypass selectors in the four directions receive the control signals and select the bypass channel from the channels inside the router and the bypass channel to transmit the subsequently arriving data packets to the next-hop router, thereby bypassing the faulty router to transmit the data packets; If the local fault information indicates that the router in the i-th row, j-th column is a healthy router, the fault information unit of the i-th row, j-th column router receives its own local fault information sent by the i+1-th row, j-th column, i-1-th row, j+1-th column and i-th row, j-1-th column routers respectively, and after integrating with its own local fault information, sends the integrated local fault information to the i+1-th row, j-th column, i-1-th row, j+1-th column and i-th row, j-1-th column routers respectively. At the same time, the i+1-th row, j-th column, i-1-th row, j+1-th column and i-th row, j-1-th column routers receive the local fault information sent by their respective adjacent routers, and after integrating with their own local fault information, send the integrated local fault information to their respective adjacent routers respectively. Therefore, after several clock cycles, the fault information unit in the i-th row, j-th column router obtains the local fault information of all routers on the i-th row and j-th column, which is used for channel selection of subsequent data packets. The routing calculation unit of the router in the i-th row and j-th column determines the relative position relationship between the router itself and the destination router; If the destination router is a router in the same row and column, the bypass selector of the downstream router of the i-th row and j-th column router selects the single flit buffer in the bypass channel to cache the arriving data packet. If the internal channel of the i-th row and j-th column router stores data packets, the arbitration unit at the output port arbitrates the data packets in the bypass channel and the internal channel: only when the previous data packet is sent, the arbitration unit receives the next data packet with the highest priority for transmission, where the priority of the data packet is proportional to the waiting time. If the destination router is not a router in the same row and column, indicating that the data packet cannot reach the destination router through the straight routing path, then the corresponding redirection router is selected using the routing method according to the local fault information of all routers in the i-th row and the j-th column, so that the bypass selector of the selected redirection router uses the channel inside its own router to transmit the received data packet, and completes the data packet redirection operation at the cross switch of the selected redirection router, and the routing calculation unit of the selected redirection router determines the relative position relationship between its own router and the destination router, and thus selects to continue to transmit the data packet through the internal channel or the bypass channel according to the relative position relationship until the data packet reaches the destination router.

2. The dynamic bypass fault-tolerant router in the network on chip according to claim 1, characterized in that: The routing method comprises the following steps: Step a, the routing calculation unit in the router in the i-th row and j-th column determines whether the router in the i-th row and j-th column and the destination router are located in the same row and column according to its own coordinate information and the coordinate information of the destination router. If so, the data packet is directly forwarded to the destination router through the bypass channel of the downstream router of the i-th row and j-th column router, and the routing is terminated. Otherwise, the remaining hops in the effective direction are calculated, and step b is executed; Step b: If the number of remaining hops between the router in the i-th row and the j-th column and the destination router in the effective direction is 1, and the next hop router in the effective direction is a faulty router, it means that the data packet cannot reach the destination router through the shortest path, and step c is executed; Otherwise, execute step d; Step c: The router in row i and column j selects a non-shortest routing path with less congestion from the valid directions according to the current congestion information, thereby forwarding the data packet to the destination router and ending the routing; Step d, the routing calculation unit in the router where the data packet is located obtains two routing paths according to the routing selection order of the effective direction, and calculates the coordinate information of a steering router on each of the two routing paths, thereby judging whether the steering routers on the two routing paths are both healthy routers according to the local fault information of the routers in the same row and column stored in the fault information unit of the own router; If yes, execute step e, otherwise, if there is at least one faulty router in the redirection router, execute step f; Step e, the router where the data packet is located forwards the data packet to the steering router in the effective direction with less congestion through the bypass channel, and the bypass selector of the corresponding port of the steering router performs a steering operation on the data packet according to the address information of the received data packet and uses the cross switch of the steering router, and then sends the data packet to the destination router in the same row and column through the bypass channel, and the routing is ended; Step f, if both of the two redirection routers are faulty routers, it means that the data packet cannot be redirected at the cross switch of the redirection routers on the two routing paths, and step g is executed; otherwise, the router where the data packet is located forwards the data packet to the healthy redirection router through the bypass channel, and forwards the data packet to the destination router through the bypass channel of the healthy redirection router, and then the routing is ended; Step g: The router where the data packet is located selects the previous healthy router of the faulty redirection router in the effective direction with less congestion as the new redirection router according to the congestion information, and sends the data packet to the new redirection router, and then executes step d.

Images