JP6197817B2 - Relay device, relay method, and relay system - Google Patents

Description

  The present invention relates to a relay device, a relay method, and a relay system, and more particularly, to a relay device, a relay method, and a relay system that relay packets between a plurality of nodes.

  In recent years, with the advancement of semiconductor technology, multi-core semiconductor devices (LSI: Large Scale Integration, etc.) have progressed, and network-on-chip (NoC) is attracting attention in order to cope with the increasing bus communication between cores. Has been. In NoC, routers that relay packets are arranged between buses connected to each Core, and arbitration of packets relayed by the routers is performed.

  A round robin method is adopted as a general NoC bus arbitration method. As a related arbitration method, for example, the method of Patent Document 1 is known. In Patent Document 1, in order to improve transmission performance, information indicating a deadline time is given to a packet, and the router determines a priority order based on the information. This deadline time is, for example, the time when a packet should arrive by the destination. Patent Documents 2 and 3 are known as other related techniques. Patent Document 2 detects the temperature of its own router.

JP 2014-197789 A JP 2006-173938 A JP 2013-005145 A

  As described above, in a related technique such as Patent Document 1, deadline time information, which is a time at which a packet should arrive at a destination, is given to a packet by a request transmission source, and the deadline time is assigned to a transmission order. Is used to determine the priority.

  In recent years, traffic has increased with the increase in the number of multi-cores. Therefore, higher performance is required for NoC circuits. In the related technology, in order to guarantee required performance such as delay and throughput, a packet whose deadline is set earlier is transmitted preferentially. In this case, the related technology gives priority only to the performance, so that there is a problem that the operating rate of some Cores increases, the temperature rises, and the failure rate of the NoC device may increase.

  In view of such problems, an object of the present invention is to provide a relay device, a relay method, and a relay system that can reduce the failure rate.

  The relay apparatus according to the present invention supports a plurality of input buffers corresponding to a plurality of input ports connected to a plurality of nodes via a bus, and a plurality of output ports connected to the plurality of nodes via a bus. A plurality of output buffers, a switch for transferring the packets stored in the plurality of input buffers to the plurality of output buffers according to a destination of the packet, and the transfer based on the temperatures of the plurality of nodes And an arbitration control unit that arbitrates a plurality of packets that are transmitted.

  The relay method according to the present invention supports a plurality of input buffers corresponding to a plurality of input ports connected to a plurality of nodes via a bus, and a plurality of output ports connected to the plurality of nodes via a bus. A plurality of output buffers, wherein the packets stored in the plurality of input buffers are transferred to the plurality of output buffers according to destinations of the packets, Based on the temperature of the node, the plurality of packets to be transferred are arbitrated.

  The relay system according to the present invention is a relay system including a plurality of nodes and a relay device that relays a packet between the plurality of nodes. The relay device is connected to the plurality of nodes via a bus. A plurality of input buffers corresponding to a plurality of connected input ports, a plurality of output buffers corresponding to a plurality of output ports connected to the plurality of nodes via a bus, and stored in the plurality of input buffers A switch that transfers a packet to the plurality of output buffers according to a destination of the packet; and an arbitration control unit that arbitrates the plurality of packets to be transferred based on temperatures of the plurality of nodes. It is.

  ADVANTAGE OF THE INVENTION According to this invention, the relay apparatus, the relay method, and relay system which can reduce a failure rate can be provided.

It is a block diagram which shows the outline | summary of the relay apparatus which concerns on embodiment. 3 is a configuration diagram showing a configuration example of NoC according to Embodiment 1. FIG. It is a figure which shows the temperature information collection image of NoC which concerns on Embodiment 1. FIG. 3 is a diagram illustrating a NoC packet transfer image according to Embodiment 1. FIG. 2 is a configuration diagram illustrating a configuration example of a router according to Embodiment 1. FIG. 3 is a flowchart illustrating an operation example according to the first embodiment. 3 is a flowchart illustrating an operation example according to the first embodiment. FIG. 6 is a configuration diagram illustrating a configuration example of a router according to a second embodiment. 6 is a flowchart illustrating an operation example according to the second embodiment. 6 is a flowchart illustrating an operation example according to the second embodiment. FIG. 10 is a configuration diagram illustrating a configuration example of a router according to a third embodiment. 10 is a flowchart illustrating an operation example according to the third embodiment. 10 is a flowchart illustrating an operation example according to the third embodiment. FIG. 10 is a configuration diagram illustrating a configuration example of a router according to a fourth embodiment. 10 is a flowchart illustrating an operation example according to the fourth embodiment. 10 is a flowchart illustrating an operation example according to the fourth embodiment. It is a block diagram which shows the structural example of the router of a reference example.

(Study leading to the embodiment)
Prior to the description of the embodiment, a reference example based on Patent Document 1 will be described.

  FIG. 17 shows a configuration of a router of a reference example that relays a packet between Cores in NoC. As illustrated in FIG. 17, the router 900 of the reference example includes input buffer control units 1-1 to 1-5 corresponding to input ports from adjacent cores, an arbitration control unit 3 that performs packet arbitration, a crossbar 4, and an output. Registers 5-1 to 5-5 are provided.

  The input buffer control units 1-1 to 1-5 corresponding to each port store packets input from each port (Core). The arbitration control unit 3 performs arbitration using destination Core information and deadline time information held by each input buffer packet itself. In order to guarantee the performance (delay and throughput) obtained by determining the priority of the arbitration control unit 3, a packet with a short (early) deadline time is preferentially transmitted. The output registers 5-1 to 5-5 store the packet selected by the arbitration control unit 3, and issue (output) the stored packet to each output port.

  The deadline time can be selected from the generation time of the request in Core and the packet transmission time of the router. When the request generation time is selected, the original deadline time is the request generation time + allowable delay time. For example, CoreC1 (port P1) issues a request for multiple packets at time t0, CoreC4 (port P4) issues a request for one packet at time t1 (t0 <t1), and the allowable delay time of each Core is common. When the value is d, the deadline times of the plurality of packets at the port P1 are all t0 + d, and the deadline times of the packets at the port P4 are t1 + d.

  Since the relationship of the deadline for determining the priority order is t0 + d <t1 + d, the plurality of packets at the port P1 are given priority, and after all the plurality of packets at the port P1 are output to the output port, the packet at the port P4 is output. . As described above, processing is given priority to a specific input port, processing of other ports is delayed, and an average delay from input to the router to output may be increased.

  In order to prevent this, deadline information is given as packet transmission time + allowable delay time. For example, assume that CoreC1 (port P1) requests a plurality of packets at time t0, and CoreC4 (port P4) issues a request for one packet at time t1 (t0 <t1). It is assumed that it takes Δt to reach the router from each Core. In this case, the transmission times of the packets of the ports P1 and P4 are time t0 + n × Δt (1 ≦ n) and time t1 + Δt, respectively. That is, the time of the first packet of the port P1 is (t0 + Δt), the time of the second packet is (t0 + 2 × Δt), and the time of one packet of the port P4 is t1 + Δt.

  When the relationship of the time of the first packet of the port P1 <the time of the first packet of the port P4 <the time of the second packet of the port P1 holds, if the mediation is performed in ascending order of the deadline time, the first packet of the port P1, The packet is transmitted to the output port in the order of one packet of P4 and the second and subsequent packets of port P1. In this way, the average delay from the input to the router to the output can be reduced, and an improvement in transmission performance can be expected.

  From the above, in the reference example such as Patent Document 1, since arbitration between packets is performed in order to comply with the deadline time, it is possible to improve the transmission performance. However, in the reference example, in order to give priority to the performance, the operation rate of each Core is increased, and there is a possibility that the LSI temperature becomes so high that the influence on the failure rate cannot be ignored. Therefore, in the embodiment, it is possible to provide a NoC circuit that maintains the LSI temperature to such an extent that the failure rate does not increase while ensuring the transmission performance.

(Outline of the embodiment)
FIG. 1 is a configuration diagram illustrating an overview of a relay device according to an embodiment. As illustrated in FIG. 1, the relay device 10 according to the embodiment is a NoC router or the like, and includes a plurality of input buffers 11, a plurality of output buffers 12, a switch 13, and an arbitration control unit 14.

  The plurality of input buffers 11 correspond to a plurality of input ports 11a connected to a plurality of nodes 20 via a bus. The plurality of output buffers 12 correspond to a plurality of output ports 12a connected to a plurality of nodes 20 via a bus. The switch 13 transfers the packets stored in the plurality of input buffers 11 to the plurality of output buffers 12 according to the destinations of the packets. The arbitration control unit 14 arbitrates a plurality of transferred packets based on the temperatures of the plurality of nodes 20.

  As described above, in the embodiment, arbitration is controlled based on the temperature of a plurality of nodes such as Core, so that packets can be transferred with a priority according to the temperature of the node. Thereby, the temperature rise of a node can be suppressed and the failure rate of the device mounting the node can be reduced.

(Embodiment 1)
The first embodiment will be described below with reference to the drawings. In this embodiment, a temperature sensor is mounted for each Core, and the router gives priority to outputting a packet coming from a core having a low temperature over a packet coming from a core having a high temperature based on the temperature information. The feature is that the failure rate of the LSI is reduced by delaying the processing of the high temperature core, lowering the operation rate, and lowering the temperature of the LSI.

<Configuration of the embodiment>
First, the configuration of the embodiment will be described with reference to FIGS. FIG. 2 shows the configuration of NoC according to the present embodiment.

  As shown in FIG. 2, the NoC (LSI) 100 according to the present embodiment includes 16 CoreC0 to C15 (also referred to as CoreC or Core at the end) that perform arithmetic processing, and the temperatures at which the temperatures of the CoreC0 to C15 are observed. Sensors S0 to S15 (also referred to as temperature sensors S or simply temperature sensors) and routers R0 to R15 (also referred to as routers R or simply routers) that relay packets arranged between the Cores C0 to C15 are provided. The relay device is not limited to the router, and may be another relay device that relays packets between nodes.

  For example, the NoC 100 forms a network with a two-dimensional mesh topology, but may form a network with other topologies. In the example of FIG. 2, the routers R0 to R15 are arranged in a matrix of 4 rows × 4 columns, and are connected to adjacent routers in the row direction (horizontal direction) and column direction (vertical direction) via buses. Yes. The Cores C0 to C15 are arranged in the vicinity of the routers R0 to R15, respectively, and are connected to the routers R0 to R15 via a bus. The temperature sensors S0 to S15 are arranged in (or in the vicinity of) CoreC0 to C15, respectively, and are connected to the routers R0 to R15. The temperature sensors S0 to S15 and the routers R0 to R15 may be connected via a bus between the CoreC0 to C15 and the routers R0 to R15, or may be connected via other wiring.

  As shown in FIG. 3, the router R (for example, R5) according to the present embodiment acquires the temperatures of all the CoreC0 to C15 from the temperature sensors S0 to S15. For example, as shown in FIG. 4, the router R5 receives the packet PKT1 from the CoreC0 via the router R0, receives the packet PKT2 from the CoreC2 via the router R2, and receives the received PKT1 and PKT2 as the router. When trying to output to R9, packets PKT1 and PKT2 compete at the output destination of router R5. In the present embodiment, the competing packets are arbitrated based on the acquired Core temperature as shown in FIG. For example, the packets are selected and output in the priority order according to the temperatures of the CoreC0 and C2 that are the issue sources of the packets PKT1 and PKT2.

  FIG. 5 shows, for example, the configuration of the router R5 among the routers R in the NoC 100 of FIG. The routers R0 to R15 have the same configuration.

  As shown in FIG. 5, the router R5 according to the present embodiment includes an input buffer control unit 1-1 to 1-5 corresponding to an input port from an adjacent CoreC, and temperature information management for managing temperature information of each CoreC. Unit 2, an arbitration control unit 3 that performs packet arbitration, a crossbar 4, and output registers 5-1 to 5-5.

  The input buffer control units (input buffers) 1-1 to 1-5 correspond to the input ports P1, P4, P6, and P5, and store packets input from the respective ports (CoreC). The input buffer control unit 1-1 has one output destination register 1-1-2 to 1-1-5 corresponding to the four output ports P1, P4, P6, and P9, and four entries shared by the four ports. It has a possible shared buffer 1-1-1. The input buffer control units 1-2 to 1-5 have the same configuration as the input buffer control unit 1-1. The shared buffer 1-1-1 can have more entries than the above in consideration of performance, LSI size, power, and the like.

  The crossbar (switching unit) 4 is a crossbar switch that switches between the input buffer control units 1-1 to 1-5 and the output registers 5-1 to 5-5 according to the destination of the packet. The crossbar 4 follows the arbitration of the arbitration control unit 3 with respect to the packets of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5. Output to -5.

  The output registers (output buffers) 5-1 to 5-5 correspond to the output ports P1, P4, P6, and P5, store the packets selected by the arbitration of the arbitration control unit 3, and store the stored packets. Issue (output) to each output port. For example, the input ports P1, P4, P6, and P5 and the output ports P1, P4, P6, and P5 are the same port, but may be different ports. (Input and output) Port P1 is connected to router R1 (CoreC1), port P4 is connected to router R4 (CoreC4), port P6 is connected to router R6 (CoreC6), and port P5 is connected to CoreC5. Connecting.

  The temperature information management unit (temperature acquisition unit or temperature determination unit) 2 collects temperature information of each CoreC0 to C15, monitors the presence / absence of Core exceeding a certain threshold temperature, and information on Core exceeding the threshold To the arbitration control unit 3. The temperature threshold can be set variably.

  The arbitration control unit (arbitration unit) 3 includes the Core information obtained from the temperature information management unit 2 and the output destination registers 1-1-2 to 1-1 of the input buffer control units 1-1 to 1-5. Arbitration is performed using the issuer core information stored in the packet 5 and the in-process TAT information. For example, when the arbitration control unit 3 selects a packet to be prioritized by arbitration, the crossbar 4 outputs the selected packet to the output register. In-process TAT (Turn Around Time) information indicates the latency after the packet is issued from the transmission source. In-process TAT information is always counted and updated while being stored in the input buffer control units 1-1 to 1-5. That is, the input buffer control units 1-1 to 1-5 include a counter (not shown) that counts in-process TAT information.

<Operation of the embodiment>
A specific operation will be described with reference to the block diagram of FIG. 5 and FIG. 6 and FIG.

  As shown in FIG. 6, the router R according to the present embodiment stores packets from each port (Core) in the input buffer control units 1-1 to 1-5 corresponding to each port (S101).

  In the input buffer control unit 1-1 to 1-5 of the router R itself, when the output destination registers 1-1-2 to 1-1-5 corresponding to the output ports are free, the corresponding output destination register 1- The packet is stored in 1-2 to 1-1-5. When a packet is stored in the corresponding output destination register 1-1-2 to 1-1-5, the packet is stored in the shared buffer 1-1-1. When the packets stored in the corresponding output destination registers 1-1-2 to 1-1-5 are output to the output registers 5-1 to 5-5 and become free, they are stored in the shared buffer 1-1-1. The stored packet is stored in the output destination registers 1-1-2 to 1-1-5.

  The temperature information management unit 2 collects temperature information of each Core C0 to C15 (S102), and determines whether or not there is a Core whose temperature exceeds a certain threshold (S103). The temperature information management unit 2 periodically repeats temperature information collection and comparison with a threshold value to monitor the presence / absence of the Core exceeding the threshold value, and passes the Core information exceeding the threshold value to the arbitration control unit 3.

  The arbitration control unit 3 performs arbitration based on the issuer Core exceeding the threshold and the in-process TAT information (S104). That is, the arbitration control unit 3 receives the Core information obtained from the temperature information management unit 2 and exceeds the threshold value, and the output destination registers 1-1-2 to 1-1 of the input buffer control units 1-1 to 1-5. Mediation is performed using the issuer Core information and in-process TAT information given to the packet stored in 5. In the present embodiment, the arbitration control unit 3 prioritizes transfer of a packet having a lower temperature of the issuer node among a plurality of packets, and particularly prioritizes transfer of a packet whose temperature of the issuer node is lower than a threshold value. Further, priority is given to transfer of a packet having a larger latency among a plurality of packets.

  Specifically, in S104, arbitration is performed as in S111 to S118 shown in FIG. As illustrated in FIG. 7, the arbitration control unit 3 performs arbitration after S112 when a plurality of packets output to the same output port (output registers 5-1 to 5-5) compete (S111).

  The arbitration control unit 3 determines whether the issuer core of some of the plurality of packets exceeds the threshold (S112). That is, the temperature of the issuer Core sets a threshold value in the output destination register (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When a packet that exceeds the threshold and the packet of the issuer Core that does not exceed the threshold value is stored and a conflict occurs, the arbitration control unit 3 sets the threshold value of the issuer Core higher than the packet that the issuer Core exceeds the threshold value. A packet that has not exceeded is preferentially issued (output to the output registers 5-1 to 5-5) (S113).

  Further, the arbitration control unit 3 performs arbitration based on the in-process TAT information for the packet whose issuer core temperature exceeds the threshold (S114). In other words, a packet in which the temperature of the issuer core exceeds a threshold value is preferentially issued when the in-process TAT information reaches a certain value so that a timeout is not detected at the issuer core after losing arbitration.

  On the other hand, the arbitration control unit 3 determines whether or not all issuer cores of a plurality of packets exceed a threshold value (S115). That is, the temperature of the issue source Core is stored in all the output destination registers (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When a packet exceeding the threshold is stored, the arbitration control unit 3 issues a packet with a large in-process TAT information, that is, a packet that takes time to process, giving priority to the in-process TAT information. In this case, priority is determined by round robin (S116).

  On the other hand, the arbitration control unit 3 determines whether or not all issuer cores of a plurality of packets are equal to or less than a threshold value (S117). That is, the temperature of the issue source Core is stored in all the output destination registers (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When packets that do not exceed the threshold value are stored, the arbitration control unit 3 issues a packet with a large value of in-process TAT information with priority, and when the in-process TAT information is the same, priority is given in round robin. (S118).

  In general, a packet from a certain core is a read and the core receives the reply, and then starts the next processing, or performs an operation of confirming whether the packet has been written and completed by a subsequent read. However, if the priority of the packet from the Core exceeding the threshold is dropped, the completion of the operation is delayed, it takes time until the next processing is performed, the operation rate is lowered, and the Core temperature is also lowered. In this way, by lowering the priority of processing of a core packet having a high temperature, the temperature in the LSI is lowered, and the failure rate of the LSI can be reduced.

  Note that, as described above, the arbitration control unit 3 not only drops the priority of the packet whose issuer core temperature exceeds the threshold, but also issues the packet whose issuer core temperature exceeds the threshold for a certain period of time. It is possible to lower the temperature of the issuer core more quickly by not performing or issuing an issue interval. Further, the information used for the arbitration given to the packet is not limited to the in-process TAT information of the present plan, but may be information that can give priority between the packets such as deadline time information of related technology.

<Effect of Embodiment>
According to the NoC arbitration method according to the present embodiment, the core information of the issuer is added to the packet, the temperature sensor is added to each core, and a circuit that determines the priority of the packet based on the temperature information of each core is added to the router, By lowering the priority of core packets having a high temperature, it is possible to lower the core operation rate and lower the temperature. In particular, the temperature rise of the packet issuer Core can be suppressed. Therefore, according to the present embodiment, it is possible to suppress the heat generation of the LSI, reduce the failure rate of the LSI, that is, improve the system availability.

(Embodiment 2)
In the second embodiment, the basic configuration is the same as that in FIGS. 2 and 5 of the first embodiment, but the arbitration method of the arbitration control unit 3 is devised in order to lower the temperature of the LSI.

  FIG. 8 shows a configuration of a router R (for example, R5) according to the present embodiment. 8 differs from FIG. 5 of the first embodiment in that the arbitration control unit 3 obtains destination Core information instead of packet issuer Core information, and performs arbitration based on the temperature state of the destination Core. is there.

  The operation of this embodiment will be described with reference to FIGS. As shown in FIG. 9, as in the first embodiment, the router R stores packets from each port in the input buffer control units 1-1 to 1-5 corresponding to each port (S201), and the temperature The information management unit 2 collects temperature information of each Core (S202), and determines whether there is a Core that exceeds a certain threshold (S203).

  Further, the arbitration control unit 3 according to the present embodiment performs arbitration based on the destination Core exceeding the threshold and the in-process TAT information (S204). That is, the arbitration control unit 3 receives the Core information obtained from the temperature information management unit 2 and exceeds the threshold value, and the output destination registers 1-1-2 to 1-1 of the input buffer control units 1-1 to 1-5. Mediation is performed using the destination Core information and in-process TAT information given to the packet stored in 5. In the present embodiment, the arbitration control unit 3 prioritizes transfer of a packet having a lower destination node temperature among a plurality of packets, particularly prioritizes transfer of a packet having a destination node temperature lower than a threshold, Priority is given to transfer of a packet having a larger latency among the plurality of packets.

  Specifically, in S204, arbitration is performed as in S211 to S218 shown in FIG. As shown in FIG. 10, the arbitration control unit 3 performs arbitration after S212 when a plurality of packets output to the same output port (output registers 5-1 to 5-5) compete (S211).

  The arbitration control unit 3 determines whether the destination core of some of the plurality of packets exceeds the threshold (S212). That is, the temperature of the destination Core exceeds the threshold in the output destination register (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When the contention and the packet of the destination Core not exceeding the threshold value are stored and a conflict occurs, the arbitration control unit 3 determines that the packet whose destination Core does not exceed the threshold value than the packet whose destination Core exceeds the threshold value. Is issued preferentially (S213).

  In addition, the arbitration control unit 3 prioritizes when the in-process TAT information reaches a certain value so that a packet whose destination Core temperature exceeds the threshold continues to lose the arbitration and does not detect a timeout at the issuer Core. Is issued (S214).

  On the other hand, the arbitration control unit 3 determines whether or not all destination Cores of a plurality of packets exceed a threshold value (S215). That is, the temperature of the destination Core is a threshold value in all of the output destination registers (any of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When packets exceeding the threshold are stored, the arbitration control unit 3 preferentially issues a packet with a large in-process TAT information, that is, a packet that takes a long time to process, and when the in-process TAT information is the same. Determines the priority in round robin (S216).

  On the other hand, the arbitration control unit 3 determines whether or not all destination Cores of the plurality of packets are equal to or less than a threshold value (S217). That is, the temperature of the destination Core is a threshold value in all of the output destination registers (any of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. If packets that do not exceed are stored, the arbitration control unit 3 gives priority to issuing a packet with a large value of in-process TAT information, and when the in-process TAT information is the same, prioritizes the round-robin priority. Decide (S218).

  As described above, according to the present embodiment, by delaying access to the destination core that exceeds the threshold, lowering the operation rate of the destination core, and lowering the temperature in the LSI, the LSI failure rate as in the first embodiment. Can be reduced. In particular, an increase in the temperature of the packet destination core can be suppressed.

(Embodiment 3)
In the third embodiment, the basic configuration is the same as in FIGS. 2 and 5 of the first embodiment, but the arbitration method of the arbitration control unit 3 is devised to lower the temperature of the LSI.

  FIG. 11 shows a configuration of a router R (for example, R5) according to the present embodiment. In FIG. 11, the difference from FIG. 5 of the first embodiment is that the arbitration control unit 3 acquires the issuer core information and the destination core information of the packet, and arbitrates based on the temperature states of both the issuer core and the destination core. It is a point to do.

  The operation of this embodiment will be described with reference to FIGS. As shown in FIG. 12, as in the first embodiment, the router R stores packets from each port in the input buffer control units 1-1 to 1-5 corresponding to each port (S301), and the temperature The information management unit 2 collects temperature information of each Core (S302), and determines whether there is a Core whose temperature exceeds a certain threshold (S303).

  Furthermore, the mediation control unit 3 according to the present embodiment performs mediation based on the issuer core and destination Core that exceed the threshold and in-process TAT information (S304). That is, the arbitration control unit 3 receives the Core information obtained from the temperature information management unit 2 and exceeds the threshold value, and the output destination registers 1-1-2 to 1-1 of the input buffer control units 1-1 to 1-5. Mediation is performed by using the issuer core and destination core information and in-process TAT information given to the packet stored in 5. In the present embodiment, the arbitration control unit 3 gives priority to the transfer of a packet having a lower temperature of the source node and the destination node among the plurality of packets, and in particular, the temperature of the source node and the destination node is lower than the threshold value. Prioritize the transfer of packets, then prioritize the transfer of packets whose destination node or issuer node temperature is lower than the threshold, and prioritize the transfer of packets with higher latency among the plurality of packets.

  Specifically, in S304, arbitration is performed as in S311 to S318 shown in FIG. As illustrated in FIG. 13, the arbitration control unit 3 performs arbitration after S312 when a plurality of packets output to the same output port (output registers 5-1 to 5-5) compete (S311).

  The arbitration control unit 3 determines whether the issuer core and / or the destination core of some of the plurality of packets exceed a threshold (S312). That is, the issuer core and / or the destination core are stored in the output destination register (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. In the case where there is a conflict between the packet of which the temperature of the source exceeds the threshold and the packet of the issuer Core and / or the destination Core not exceeding the threshold, the arbitration control unit 3 determines that the issuer Core and / or the destination Core A packet whose issuer Core and / or destination Core do not exceed the threshold is issued with priority over the packet whose threshold exceeds the threshold (S313).

Specifically, arbitration is performed as follows between the packets exceeding the threshold and the packets below the threshold.
Packets in which the temperature of both the destination core and the issuer core exceeds the threshold <a packet in which the temperature of the destination core exceeds the threshold <a packet in which the temperature of the issuer Core exceeds the threshold <a packet in which the temperature of the destination Core is less than the threshold < Packets whose issuer Core temperature is below the threshold

  In addition, the arbitration control unit 3 sets the in-process TAT information to a certain value so that a packet in which the temperature of the issuer core or the destination core exceeds the threshold continues to lose the arbitration and does not detect a timeout at the issuer core. If it is, it is issued preferentially (S314).

  On the other hand, the arbitration control unit 3 determines whether or not all the issuer cores and the destination cores of the plurality of packets exceed the threshold (S315). In other words, all of the output destination registers (any of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port are included in the issuer core and the destination core. If a packet whose temperature exceeds the threshold value is stored, the arbitration control unit 3 gives priority to issuing a packet with a large in-process TAT information, that is, a packet that takes a long time to process, and further processes the in-process TAT. When the information is the same, priority is determined by round robin (S316).

  On the other hand, the arbitration control unit 3 determines whether all the issuer cores and the destination cores of a plurality of packets are equal to or less than a threshold value (S317). In other words, all of the output destination registers (any of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port are included in the issuer core and the destination core. If the packet whose temperature does not exceed the threshold value is stored, the arbitration control unit 3 issues a packet with a large value of the in-process TAT information preferentially, and when the in-process TAT information is the same, round-robin is performed. The priority order is determined (S318).

  As described above, according to the present embodiment, by delaying the access related to both the destination core and the issuer core exceeding the threshold, the operation rate of the destination core and the issuer core can be further efficiently reduced, By reducing the temperature of the LSI, it is possible to reduce the LSI failure rate.

(Embodiment 4)
In the fourth embodiment, the basic configuration is the same as in FIGS. 2 and 5 of the first embodiment, but the temperature information management unit 2 and the arbitration control unit 3 are devised to lower the temperature of the LSI. .

  FIG. 14 shows a configuration of a router R (for example, R5) according to the present embodiment. In FIG. 14, the difference from FIG. 5 of the first embodiment is that the temperature information management unit 2 obtains temperature information from each Core, and does not only determine whether the temperature of the Core exceeds a certain threshold value. The threshold setting is subdivided, for example, threshold 1 = 40 ° C., threshold 2 = 60 ° C., threshold 3 = 80 ° C., and the state of each Core is passed to the arbitration control unit 3, and the arbitration control unit 3 performs priority control based on the information. Note that the threshold is not limited to three, and an arbitrary number of thresholds may be set. Moreover, not only a threshold value but temperature and a priority may be linked | related and priority control may be performed by the priority corresponding to temperature.

  The operation of this embodiment will be described with reference to FIGS. As shown in FIG. 15, as in the first embodiment, the router R stores packets from each port in the input buffer control units 1-1 to 1-5 corresponding to each port (S401). The temperature information management unit 2 collects temperature information of each Core (S402), and determines whether there is a Core whose temperature exceeds the threshold 1, threshold 2, and threshold 3 (S403).

  Further, the arbitration control unit 3 according to the present embodiment performs arbitration based on the issuer Core that exceeds the threshold 1, threshold 2, and threshold 3, and the in-process TAT information (S404). That is, the arbitration control unit 3 receives the information on the cores exceeding the threshold 1, threshold 2, and threshold 3 obtained from the temperature information management unit 2, and the output destination registers 1-1 of the input buffer control units 1-1 to 1-5. Arbitration is performed using issuer Core information and in-process TAT information given to the packets stored in -2-1-1-5. In the present embodiment, the arbitration control unit 3 gives priority to transfer of a packet having a lower temperature of the issuer node among the plurality of packets, and in particular, according to a comparison result between the temperature of the issuer node and the plurality of threshold values. The packet is arbitrated, and the transfer of a packet having a larger latency among the plurality of packets is prioritized.

  Specifically, in S404, arbitration is performed as in S411 to S418 shown in FIG. As illustrated in FIG. 16, the arbitration control unit 3 performs arbitration after S412 when a plurality of packets output to the same output port (output registers 5-1 to 5-5) compete (S411).

  The arbitration control unit 3 determines whether the issuer core of some of the plurality of packets exceeds the threshold 1, the threshold 2, or the threshold 3 (S412). In other words, the temperature of the issue source Core is the threshold value 1 in the output destination register (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When a packet that exceeds the threshold 2 or 3 and the packet of the issuer Core that does not exceed the threshold 1, threshold 2, or threshold 3 is stored and a conflict occurs, the arbitration control unit 3 Packets whose issuer Core does not exceed threshold 1, threshold 2, or threshold 3 are issued with priority over packets whose Core exceeds threshold 1, threshold 2, or threshold 3 (S413).

Specifically, arbitration is performed in the following manner for the priorities of packets around threshold 1, threshold 2, and threshold 3.
Packets with issuer core temperature exceeding threshold 3 <packets with issuer core temperature between thresholds 2 and 3 <packets with issuer core temperature between thresholds 1 and 2 <issuer core temperature Packets whose threshold is lower than 1

  In addition, the arbitration control unit 3 performs in-process TAT information so that a packet whose temperature of the issuer core exceeds the threshold value 1, the threshold value 2, or the threshold value 3 continues to lose the arbitration and does not detect a timeout at the issuer core. When the value reaches a certain value, it is issued preferentially (S414).

  On the other hand, the arbitration control unit 3 determines whether or not all issuer cores of a plurality of packets exceed all thresholds (S415). That is, the temperature of the issue source Core is stored in all the output destination registers (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When packets exceeding all the threshold values are stored, the arbitration control unit 3 issues a packet with large in-process TAT information, that is, a packet that takes time to process, giving priority to issuing the in-process TAT information. At the same time, priority is determined by round robin (S416).

  On the other hand, the arbitration control unit 3 determines whether or not all issuer cores of the plurality of packets are equal to or less than all threshold values (S417). That is, the temperature of the issue source Core is stored in all the output destination registers (any one of the output destination registers 1-1-2 to 1-1-5 of the input buffer control units 1-1 to 1-5) to the same output port. When packets that do not exceed all the threshold values are stored, the arbitration control unit 3 issues a packet with a large value in the in-process TAT information with priority, and further gives priority in round robin when the in-process TAT information is the same. The order is determined (S418).

  As described above, according to the present embodiment, by subdividing the threshold value, it is possible to efficiently lower the temperature from the high temperature Core. Note that the present embodiment can be applied to the second and third embodiments, and a plurality of threshold values can be used.

  Note that the present invention is not limited to the above-described embodiment, and can be changed as appropriate without departing from the spirit of the present invention.

  Each configuration in the above-described embodiment is configured by hardware and / or software, and may be configured by one piece of hardware or software, or may be configured by a plurality of pieces of hardware or software. Each function (each process) of the information processing apparatus may be realized by a computer having a CPU, a memory, and the like. For example, a program for performing the arbitration method in the embodiment may be stored in the storage device, and each function may be realized by executing the program stored in the storage device by the CPU.

  This program can be stored using various types of non-transitory computer readable media and supplied to a computer. Non-transitory computer readable media include various types of tangible storage media. Examples of non-transitory computer-readable media include magnetic recording media (for example, flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (for example, magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / W and semiconductor memory (for example, mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (random access memory)) are included. The program may also be supplied to the computer by various types of transitory computer readable media. Examples of transitory computer readable media include electrical signals, optical signals, and electromagnetic waves. The temporary computer-readable medium can supply the program to the computer via a wired communication path such as an electric wire and an optical fiber, or a wireless communication path.

1-1 to 1-5 Input buffer control unit 1-1-1 Shared buffer 1-1-2 to 1-1-5 Output destination register 2 Temperature information management unit 3 Arbitration control unit 4 Crossbars 5-1 to 5-5 Output register 10 Relay device 11 Input buffer 11a Input port 12 Output buffer 12a Output port 13 Switch 14 Arbitration controller 20 Nodes C0 to C15 Core
P1, P4, P5, P6, P9 ports (input port / output port)
R0 to R15 Router S0 to S15 Temperature sensor

Claims (10)

A plurality of input buffers corresponding to a plurality of input ports connected to a plurality of nodes via a bus;
A plurality of output buffers corresponding to a plurality of output ports connected to the plurality of nodes via a bus;
A switch for transferring packets stored in the plurality of input buffers to the plurality of output buffers according to a destination of the packet;
Based on the temperature of the plurality of nodes, an arbitration control unit that arbitrates the transferred packets,
A relay device comprising: The arbitration control unit prioritizes transfer of a packet having a lower temperature of an issuer node among the plurality of packets.
The relay device according to claim 1. The arbitration control unit prioritizes transfer of a packet in which the temperature of the issuing node is lower than a threshold value.
The relay device according to claim 2. The arbitration control unit performs arbitration according to a comparison result between the temperature of the issuing node and a plurality of threshold values.
The relay device according to claim 3. The arbitration control unit prioritizes transfer of a packet having a lower destination node temperature among the plurality of packets.
The relay apparatus as described in any one of Claims 1 thru | or 4. The arbitration control unit prioritizes transfer of a packet in which the temperature of the destination node is lower than a threshold;
The relay device according to claim 5. The arbitration control unit performs arbitration according to a comparison result between the temperature of the destination node and a plurality of threshold values.
The relay device according to claim 6. The arbitration control unit prioritizes transfer of a packet having a higher latency among the plurality of packets.
The relay device according to any one of claims 1 to 7. A plurality of input buffers corresponding to a plurality of input ports connected to a plurality of nodes via a bus; and a plurality of output buffers corresponding to a plurality of output ports connected to the plurality of nodes via a bus; A relay method in a relay device comprising:
A packet stored in the plurality of input buffers is transferred to the plurality of output buffers according to a destination of the packet;
Arbitrating the forwarded packets based on the temperature of the nodes;
Relay method. A relay system comprising a plurality of nodes and a relay device that relays packets between the plurality of nodes,
The relay device is
A plurality of input buffers corresponding to a plurality of input ports connected to the plurality of nodes via a bus;
A plurality of output buffers corresponding to a plurality of output ports connected to the plurality of nodes via a bus;
A switch for transferring packets stored in the plurality of input buffers to the plurality of output buffers according to a destination of the packet;
Based on the temperature of the plurality of nodes, an arbitration control unit that arbitrates the transferred packets,
A relay system comprising:

Images