Hierarchical throttling for network-on-chip traffic
By using a proxy to calculate the number of incomplete transactions and implementing a dynamic traffic throttling strategy, the performance degradation caused by increased on-chip network traffic was resolved, thus achieving network performance optimization and stability.
Patent Information
- Application Number
- CN202180064640.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-09-22
- Filing Date
- 2021-09-17
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2041-09-17
AI Technical Summary
As on-chip network traffic increases, the communication performance between functional components on the chip decreases, and existing technologies struggle to effectively manage and regulate traffic to avoid overload.
The agent calculates the number of incomplete transactions and compares them with multiple thresholds to implement dynamic traffic throttling strategies, including light throttling, heavy throttling, and stop throttling, dynamically adjusting the traffic generation rate to adapt to changes in network load.
It improves the communication performance of the on-chip network, effectively manages traffic through a hierarchical throttling strategy, avoids network overload and latency, and improves the overall performance of the computing system.
Smart Images

Figure CN116210213B_ABST
Abstract
Description
Background Technology
[0001] Chips such as systems-on-chips (SoCs) use on-chip networks to facilitate communication between functional components. As traffic on on-chip networks increases, the risk of performance degradation also increases. Attached Figure Description
[0002] Figure 1A This is a block diagram of a chip for hierarchical throttling of on-chip network traffic according to some implementation schemes.
[0003] Figure 1B This is a block diagram of a chip for dynamic on-chip network traffic throttling according to some implementation schemes.
[0004] Figure 2 This is a flowchart of an exemplary method for hierarchical throttling of on-chip network traffic according to some implementation schemes.
[0005] Figure 3 This is a flowchart of an exemplary method for hierarchical throttling of on-chip network traffic according to some implementation schemes.
[0006] Figure 4 This is a flowchart of an exemplary method for hierarchical throttling of on-chip network traffic according to some implementation schemes.
[0007] Figure 5 This is a flowchart of an exemplary method for hierarchical throttling of on-chip network traffic according to some implementation schemes.
[0008] Figure 6 This is a flowchart of an exemplary method for dynamic on-chip network traffic throttling according to some implementation schemes.
[0009] Figure 7 This is a flowchart of an exemplary method for dynamic on-chip network traffic throttling according to some implementation schemes.
[0010] Figure 8 This is a flowchart of an exemplary method for dynamic on-chip network traffic throttling according to some implementation schemes.
[0011] Figure 9 This is a flowchart of an exemplary method for dynamic on-chip network traffic throttling according to some implementation schemes. Detailed Implementation
[0012] In some implementations, a method for tiered throttling of on-chip network traffic includes having an on-chip network agent calculate the number of incomplete transactions issued by the agent, along with other elements. Such tiered throttling also includes determining that the number of incomplete transactions meets a threshold. In response to determining that the number of incomplete transactions meets the threshold, tiered throttling can be performed by having the agent implement a traffic throttling strategy.
[0013] In some implementations, the number of incomplete transactions includes the difference between the number of issued transactions for which no response has yet been received from the agent. In some implementations, a threshold is included among multiple thresholds, and a traffic throttling policy is included among multiple throttling policies. In such implementations, implementing a transaction policy includes implementing the policy based on which of a plurality of thresholds the number of incomplete transactions satisfies. In some implementations, tiered throttling for on-chip network traffic includes recalculating the number of incomplete transactions issued by the agent and determining that the number of incomplete transactions has fallen below a threshold. In response to such a determination, the agent terminates the traffic throttling policy. In some implementations, terminating a transaction policy includes implementing another throttling policy. In other implementations, terminating a transaction policy includes issuing transactions without throttling. In some implementations, determining that the number of incomplete transactions has fallen below a threshold includes determining that the number of incomplete transactions has fallen below the threshold by a predefined amount.
[0014] In some implementations, chip execution for tiered throttling of on-chip network traffic includes the step of calculating the number of incomplete transactions issued by the on-chip network's agent. Such tiered throttling also includes determining that the number of incomplete transactions meets a threshold. In response to determining that the number of incomplete transactions meets the threshold, tiered throttling can be performed by implementing a traffic throttling strategy through the agent.
[0015] In some implementations, the number of incomplete transactions includes the difference between the number of issued transactions for which no response has yet been received from the agent. In some implementations, a threshold is included among multiple thresholds, and a traffic throttling strategy is included among multiple throttling strategies. In such implementations, implementing a transaction strategy includes implementing the strategy based on which of a plurality of thresholds the number of incomplete transactions satisfies. In some implementations, these steps include recalculating the number of incomplete transactions issued by the agent and determining that the number of incomplete transactions has fallen below a threshold. In response to such a determination, the agent terminates the traffic throttling strategy. In some implementations, terminating a transaction strategy includes implementing another throttling strategy. In other implementations, terminating a transaction strategy includes issuing transactions without throttling. In some implementations, determining that the number of incomplete transactions has fallen below a threshold includes determining that the number of incomplete transactions has fallen below the threshold by a predefined amount.
[0016] In some implementations, the means for hierarchical throttling of on-chip network traffic includes a chip that performs steps including calculating the number of incomplete transactions issued by an agent of the on-chip network. Such hierarchical throttling also includes determining that the number of incomplete transactions meets a threshold. In response to determining that the number of incomplete transactions meets the threshold, hierarchical throttling can be performed by implementing a traffic throttling strategy by an agent.
[0017] In some implementations, the number of incomplete transactions includes the difference between the number of issued transactions for which no response has yet been received from the agent. In some implementations, a threshold is included among multiple thresholds, and a traffic throttling strategy is included among multiple throttling strategies. In such implementations, implementing a transaction strategy includes implementing the strategy based on which of a plurality of thresholds the number of incomplete transactions satisfies. In some implementations, these steps include recalculating the number of incomplete transactions issued by the agent and determining that the number of incomplete transactions has fallen below a threshold. In response to such a determination, the agent terminates the traffic throttling strategy. In some implementations, terminating a transaction strategy includes implementing another throttling strategy. In other implementations, terminating a transaction strategy includes issuing transactions without throttling. In some implementations, determining that the number of incomplete transactions has fallen below a threshold includes determining that the number of incomplete transactions has fallen below the threshold by a predefined amount.
[0018] Figure 1A This is a block diagram of a non-limiting exemplary chip 102a. Exemplary chip 102a may include various chips, including microprocessors, integrated circuits, or systems-on-a-chip. Exemplary chip 102a may be implemented in various computing devices, including mobile devices, personal computers, peripheral hardware components, gaming devices, set-top boxes, etc. Chip 102a includes a plurality of agents 104a-n. Agents 104a-n are functional units or hardware modules of chip 102a. For example, in some embodiments, agents 104a-n include various modules of a system-on-a-chip architecture.
[0019] To facilitate communication between agents 104a-n, chip 102a also includes a plurality of routing agents 106. Routing agents 106 include hardware modules for exchanging and / or routing traffic and / or messages between agents 104a-n. Therefore, routing agents 106 implement an on-chip network 108. In some embodiments, the on-chip network 108 includes a packet-switched network of routing agents 106.
[0020] Communication between agents 104a-n is decentralized because none of the agents 104a-n directly knows the status (e.g., traffic generation, responses, queues, etc.) of the other agents 104a-n and the routing agent 106, and there is no centralized entity managing the traffic generation and / or response behavior of agents 104a-n. Therefore, agents 104a-n do not directly know whether the on-chip network 108 is overloaded or near a route.
[0021] To improve the performance of traffic via the on-chip network 108, each agent 104a-n uses the number of outstanding transactions issued by that agent 104a-n as an estimate of the state of the on-chip network 108. The number of outstanding transactions for a given agent 104a-n is the number of messages sent via the on-chip network 108 to other agents 104a-n expecting responses (e.g., responses to requests, acknowledgements of receipt) that have not yet been received. When communication via the on-chip network 108 slows down or the routing agent 106 is handling an increased volume of traffic, the time it takes for agents 104a-n to respond to messages will increase. Therefore, assuming the transaction generation rate of that agent 104a-n does not decrease, the number of outstanding transactions for a given agent 104a-n will increase.
[0022] To implement tiered throttling of traffic to the on-chip network 108, agent 104a-n calculates the number of incomplete transactions issued by agent 104a-n. Agent 104a-n then compares this number to a threshold. If the number falls below the threshold, agent 104a-n continues to generate traffic according to the currently implemented traffic throttling strategy (if any). The traffic throttling strategy is a configurable or programmable limit on the amount of traffic generated by agent 104a-n for the on-chip network 108. If the number exceeds the threshold, agent 104a-n implements the traffic throttling strategy. For example, suppose agent 104a-n is generating traffic for the on-chip network 108 independently of any traffic throttling strategy (e.g., without any limits). In response to the number of incomplete transactions of agent 104a-n meeting the threshold, agent 104a-n then implements a traffic throttling strategy that limits the rate at which agent 104a-n provides traffic to the on-chip network 108.
[0023] In some implementations, the threshold is one of a plurality of thresholds, and the traffic throttling strategy is one of a plurality of traffic throttling strategies. For example, in some implementations, each traffic throttling strategy corresponds to one of a plurality of thresholds, such that when the number of incomplete transactions for a given agent 104a-n exceeds a given threshold, the corresponding traffic throttling strategy is implemented. For example, a first threshold corresponds to a first traffic throttling strategy (e.g., a "light throttling strategy"), a second threshold greater than the first threshold corresponds to a second traffic throttling strategy that is more restrictive than the first traffic throttling strategy (e.g., a "heavy throttling strategy"), and a third threshold greater than the second threshold corresponds to a third traffic throttling strategy in which agents 104a-n stop generating traffic (e.g., a "stop throttling strategy").
[0024] Using this example, where the number of pending transactions for agent 104a-n exceeds a first threshold, agent 104a-n will implement a light throttling strategy. If, even with the light throttling strategy implemented, the number of pending transactions for agent 104a-n continues to increase and exceeds a second threshold, then agent 104a-n will implement a heavy throttling strategy. If the number continues to increase and meets a third threshold, then a stop throttling strategy will be implemented.
[0025] Agent 104a-n will continuously (e.g., at predefined intervals) recalculate the number of incomplete transactions for the agent. When the number of incomplete transactions falls below a threshold (e.g., the last threshold met or exceeded), agent 104a-n terminates the currently implemented traffic throttling strategy. In some implementations, this includes removing any implemented traffic throttling strategy and sending traffic to the on-chip network 108 without restriction. In other implementations, this includes implementing another (e.g., less restrictive) traffic throttling strategy. Continuing the example above, assuming agent 104a-n is implementing a heavy throttling strategy, where the number of incomplete transactions falls below a second threshold but remains above the first threshold, agent 104a-n will implement a light throttling strategy instead of a heavy throttling strategy.
[0026] If agent 104a-n terminates its implemented throttling strategy when the number of outstanding transactions falls below a threshold, there is a risk that the number of outstanding transactions will increase rapidly, exceeding the threshold again and causing the traffic throttling strategy to be reimplemented. This will cause agent 104a-n to oscillate between implementing and terminating the traffic throttling strategy, as the number of outstanding transactions oscillates between meeting and falling below the threshold. To prevent this, in some implementations, agent 104a-n terminates its implemented throttling strategy in response to the number of outstanding transactions falling below a predefined amount below the corresponding threshold. Continuing the example above, suppose the first threshold is fifty outstanding transactions. Further suppose the number of outstanding transactions has exceeded the first threshold and the light throttling strategy is appropriate. Instead of removing the light throttling strategy when the number of outstanding transactions falls below fifty, agent 104a-n removes the light throttling strategy when the number of outstanding transactions falls below the threshold by ten transactions (e.g., forty outstanding transactions).
[0027] Figure 1B This is a block diagram of a non-limiting exemplary chip 102b. Exemplary chip 102b is similar to... Figure 1A The chip 102a, because the exemplary chip 102b includes a plurality of agents 104a-n communicatively coupled via an on-chip network 108 of a plurality of routing agents 106. In some embodiments, agent 104a performs actions such as regarding Figure 1ASimilar tiered traffic throttling is described. The difference between chip 102b and chip 102a is that chip 102b includes multiple detector modules 110. Detector modules 110 are hardware components included in or coupled to agents 104a-n and / or routing agents 106. Although Figure 1B Each agent 104a-n and routing agent 106 is shown as having a corresponding detector module 110; however, it should be understood that in some embodiments, the detector module 110 is installed on or coupled to a subset of agents 104a-n and / or routing agents 106. For example, in some embodiments, the detector module 110 is included only with routing agent 106.
[0028] Detector modules 110 monitor their corresponding components (e.g., agents 104a-n and / or routing agent 106) to determine whether predefined conditions are met. In some embodiments, predefined conditions include queue occupancy meeting a threshold. For example, routing agent 106 maintains queues of messages and / or packets to be routed. As on-chip network 108 traffic increases to a rate greater than the rate at which routing agent 106 processes messages, the queues will increase. Therefore, detector modules 110 of routing agent 106 determine whether the queue occupancy of unrouted messages meets a threshold. As another example, predefined conditions include on-chip network 108 traffic being associated with a specific service class. For example, detector modules 110 determine whether traffic generated or routed by their corresponding components belongs to a specific service class. In some embodiments, each component (e.g., agents 104a-n and / or routing agent 106) includes multiple detector modules 110. Each detector module 110 for a given component monitors different predefined conditions.
[0029] In response to the fulfillment of predefined conditions, detector module 110 (e.g., detector module 110 that determines which predefined conditions are met) sends a signal to mediator module 112. Each detector module 110 is communicatively coupled to mediator module 112. For example, each detector module 110 has a direct signal path to mediator module 112 outside of routing agent 106. Therefore, the signal to mediator module 112 does not need to be routed via routing agent 106 and may experience delays or slowdowns in the on-chip network 108.
[0030] In response to receiving the signal, the mediator module 112 sends an instruction to the plurality of agents 104a-n to implement a traffic throttling strategy. In some embodiments, the mediator module 112 sends the instruction to each of the agents 104a-n by asserting a signal on a direct (e.g., unrouted) signal path. In other embodiments, the mediator module 112 sends the instruction as a message sent via routing agent 106. In some embodiments, the instruction is sent to a subset of the agents 104a-n that generate traffic for the on-chip network 108 (e.g., those agents 104a-n that issued the transaction, excluding any agents 104a-n that only responded to other issued transactions). In some embodiments, the specific traffic throttling strategy to be implemented is indicated in the signal or message sent to the agents 104a-n. In other embodiments, the traffic throttling strategy is predefined or default.
[0031] By sending an instruction to implement a traffic throttling strategy, the mediator module 112 reduces traffic to the on-chip network 108 when the queue occupancy rate meets a threshold (indicating that the routing agent 106 is reaching its capacity). Furthermore, the implemented traffic throttling strategy improves the performance of the on-chip network 108 and enhances the overall quality of service when traffic of a specific service class is detected via the on-chip network 108.
[0032] In some implementations, agents 104a-n that receive instructions to implement a traffic throttling policy determine whether to override the traffic throttling policy. Determining to override the traffic throttling policy includes implementing a different traffic throttling policy or not implementing a traffic throttling policy. For example, agents 104a-n that generate traffic associated with a specific service class will override the traffic throttling policy by implementing a less restrictive traffic throttling policy or not implementing a traffic throttling policy at all. As another example, agents 104a-n that implement a traffic throttling policy in response to a threshold number of incomplete transactions continuing to implement their current traffic throttling policy, or implementing a more restrictive traffic throttling policy than indicated by mediator module 112.
[0033] In some implementations, detector module 110 determines that a predefined condition is not met (e.g., no longer met). Detector module 110 then indicates to mediator module 112 that the predefined condition is not met. For example, where detector module 110 indicates that the predefined condition is met by asserting a signal on the communication path to mediator module 112, indicating that the predefined condition is not met includes a deassertion signal. In other implementations, indicating that the predefined condition is not met includes sending another signal to mediator module 112 indicating that the predefined condition is not met. Mediator module 112 then causes agents 104a-n (e.g., agents 104a-n that have received an instruction to implement a traffic throttling policy) to terminate the traffic throttling policy. For example, in some implementations, mediator module 112 sends another signal to agents 104a-n to terminate the traffic throttling policy. In other implementations, mediator module 112 deassertes a signal used to indicate that a traffic throttling policy should be implemented.
[0034] although Figure 1B The functions performed by mediator module 112 are described, but it should be understood that in some implementations, the functions of mediator module 112 may alternatively be performed by detector module 110 of routing agent 106, which communicates directly with detector modules 110 of agents 104a-n.
[0035] To further explain, Figure 2 A flowchart illustrating an exemplary method for hierarchical throttling of on-chip network traffic is provided. Figure 2 The method in chip 200, such as Figure 1A Chip 102a and / or Figure 1B It is implemented in the 102b chip. Figure 2 The method includes calculating 202 the number of incomplete transactions issued by agents 104a-n of the on-chip network 108 (e.g., communicatively coupled to the on-chip network 108). The number of incomplete transactions for a given agent 104a-n is determined via the on-chip network 108 to its...
[0036] The number of messages sent by agent 104a-n that are expected to receive responses (e.g., responses to requests, acknowledgements of receipt) that have not yet been received. When communication via on-chip network 108 slows down or routing agent 106 is handling increased traffic volumes, the time it takes for agent 104a-n to respond to messages will increase. Therefore, assuming the transaction generation rate of agent 104a-n does not decrease, the number of incomplete transactions for a given agent 104a-n will increase.
[0037] Figure 2 The method also includes determining the number of 204 incomplete transactions that meets a threshold. Figure 2The method also includes implementing a traffic throttling policy 206 by agent 104a-n in response to the number of incomplete transactions meeting the threshold. The traffic throttling policy is a configurable or programmable limit on the traffic generated by agent 104a-n for on-chip network 108. For example, suppose agent 104a-n is generating traffic for on-chip network 108 independently of any traffic throttling policy (e.g., without any limits). In response to the number of incomplete transactions of agent 104a-n meeting the threshold, agent 104a-n will then implement a traffic throttling policy that limits the rate at which agent 104a-n provides traffic to on-chip network 108.
[0038] To further explain, Figure 3 A flowchart illustrating an exemplary method for hierarchical throttling of on-chip network traffic according to an embodiment of this disclosure is provided. Figure 3 Methods and Figure 2 The similarities are, Figure 3 The method includes: calculating 202 the number of incomplete transactions issued by agents 104a-n; determining 204 that the number of incomplete transactions meets a threshold; and implementing 206 a traffic throttling strategy.
[0039] Figure 3 Methods and Figure 2 The difference lies in that the traffic throttling strategy 206 implemented by agents 104a-n in response to the number of incomplete transactions meeting a threshold includes implementing the traffic throttling strategy 302 based on which of a plurality of thresholds the number of incomplete transactions meets. For example, suppose that the threshold is one of a plurality of thresholds, and that the traffic throttling strategy is one of a plurality of traffic throttling strategies. Each traffic throttling strategy corresponds to one of a plurality of thresholds such that when the number of incomplete transactions for a given agent 104a-n exceeds a given threshold, the corresponding traffic throttling strategy is implemented. For example, a first threshold corresponds to a first traffic throttling strategy (e.g., a "light throttling strategy"), a second threshold greater than the first threshold corresponds to a second traffic throttling strategy that is more restrictive than the first traffic throttling strategy (e.g., a "heavy throttling strategy"), and a third threshold greater than the second threshold corresponds to a third traffic throttling strategy in which agents 104a-n stop generating traffic (e.g., a "stop throttling strategy").
[0040] Using this example, where the number of pending transactions for agent 104a-n exceeds a first threshold, agent 104a-n will implement a light throttling strategy. If, even with the light throttling strategy implemented, the number of pending transactions for agent 104a-n continues to increase and exceeds a second threshold, then agent 104a-n will implement a heavy throttling strategy. If the number continues to increase and meets a third threshold, then a stop throttling strategy will be implemented.
[0041] To further explain, Figure 4 A flowchart illustrating an exemplary method for hierarchical throttling of on-chip network traffic according to an embodiment of this disclosure is provided. Figure 4 Methods and Figure 2 The similarities are, Figure 4 The method includes: calculating 202 the number of incomplete transactions issued by agents 104a-n; determining 204 that the number of incomplete transactions meets a threshold; and implementing 206 a traffic throttling strategy.
[0042] Figure 4 Methods and Figure 2 The difference is that, Figure 4 The method also includes (e.g., by agent 104a-n) recalculating the number of incomplete transactions issued by agent 104a-n. For example, agent 104a-n continuously recalculates the number of incomplete transactions at predefined intervals or in response to another event. Figure 4 The method also includes determining that the number of 404 incomplete transactions has decreased below a threshold (e.g., a previously met threshold). In some implementations, determining that the number of 404 incomplete transactions has decreased below the threshold includes determining a predetermined amount of time (e.g., a predefined number of clock cycles) for the number of incomplete transactions to decrease below the threshold. Thus, it is determined that the number of incomplete transactions remains below the threshold for a certain amount of time without being met or exceeding the threshold again. Figure 4 This also includes (e.g., terminating the 406 traffic throttling policy in response to the number of incomplete transactions falling below a threshold). In some embodiments, terminating the 406 traffic throttling policy includes removing any implemented traffic throttling policy and sending traffic to the on-chip network 108 without restriction. In other embodiments, terminating the 406 traffic throttling policy includes implementing another (e.g., less restrictive) traffic throttling policy. For example, if the number of incomplete transactions falls below a threshold but still exceeds another threshold, implementing another traffic throttling policy corresponding to that other threshold.
[0043] To further explain, Figure 5 A flowchart illustrating an exemplary method for hierarchical throttling of on-chip network traffic according to an embodiment of this disclosure is provided. Figure 5 Methods and Figure 4 The similarities are, Figure 5 The method includes: calculating the number of incomplete transactions 202 issued by agents 104a-n; determining that the number of incomplete transactions 204 meets the threshold; implementing the traffic throttling strategy 206; recalculating the number of incomplete transactions 402; determining that the number of incomplete transactions 404 has dropped below the threshold; and ending the traffic throttling strategy 406.
[0044] Figure 5and Figure 4 The difference lies in determining that the number of 404 incomplete transactions has fallen below a threshold, which includes determining that the number of 502 incomplete transactions has fallen below a predefined amount. For example, suppose this threshold is fifty incomplete transactions. Further assume that the number of incomplete transactions has already exceeded a first threshold and the traffic throttling policy is appropriate. Instead of removing the traffic throttling policy when the number of incomplete transactions falls below fifty, agent 104a-n removes the traffic throttling policy when the number of incomplete transactions falls below the threshold by ten transactions (e.g., forty incomplete transactions). This prevents oscillations between implementing and ending the traffic throttling policy, as the number of incomplete transactions oscillates between being met and falling below the threshold.
[0045] To further explain, Figure 6 A flowchart illustrating an exemplary method for dynamic on-chip network traffic throttling is provided. Figure 6 The method in chip 600, such as Figure 1A Chip 102a and / or Figure 1B It is implemented in the 102b chip. Figure 6 The method includes having a detector module 110 of the on-chip network 108 (e.g., corresponding to a component communicatively coupled to the on-chip network 108) determine that 602 meets predefined conditions. In some embodiments, the predefined conditions include queue occupancy meeting a threshold. For example, routing agent 106 maintains queues of messages and / or packets to be routed. As on-chip network 108 traffic increases to a rate greater than the rate at which routing agent 106 processes messages, the queues will increase. Therefore, detector module 110 of routing agent 106 determines whether the queue occupancy of unrouted messages meets a threshold. As another example, the predefined conditions include on-chip network 108 traffic being associated with a specific service class. For example, detector module 110 determines whether traffic generated or routed by its corresponding component belongs to a specific service class. In some embodiments, each component (e.g., agents 104a-n and / or routing agent 106) includes multiple detector modules 110. Each detector module 110 of a given component monitors different predefined conditions.
[0046] Figure 6 The method also includes sending a signal 604 606 from detector module 110 to mediator module 112 of on-chip network 108. Each detector module 110 is communicatively coupled to mediator module 112. For example, each detector module 110 has a direct signaling path to mediator module 112 outside of routing agent 106. Therefore, signal 606 to mediator module 112 does not need to be routed via routing agent 106 and may experience delays or slowdowns in on-chip network 108. Therefore, sending signal 604 606 includes asserting signal 606 on the signaling path to mediator module 112.
[0047] Figure 6 The method also includes sending an instruction 610 by the mediator module 112 to the plurality of agents 104a-n in response to signal 606, indicating the implementation of a traffic throttling strategy. In some embodiments, the mediator 112 sends the instruction to each of the plurality of agents 104a-n by asserting a signal on a direct (e.g., unrouted) signal path. In other embodiments, the mediator module 112 sends the instruction 610 as a message sent via the routing agent 106. In some embodiments, the instruction 610 is sent to a subset of the agents 104a-n that generate traffic for the on-chip network 108 (e.g., those agents 104a-n that issued the transaction, excluding any agents 104a-n that only responded to other issued transactions). In some embodiments, the specific traffic throttling strategy to be implemented is indicated in the signal or message sent to the agents 104a-n. In other embodiments, the traffic throttling strategy is predefined or default.
[0048] To further explain, Figure 7 A flowchart illustrating an exemplary method for dynamic on-chip network traffic throttling according to an embodiment of the present disclosure is provided. Figure 7 Methods and Figure 6 The similarities are, Figure 7 The method includes: determining that 602 meets a predefined condition; sending a signal 604 606 to the mediator module 112; and in response to the signal 606, sending an instruction 610 608 to a plurality of agents 104a-n to implement a flow throttling strategy.
[0049] Figure 7 Methods and Figure 6 The difference is that, Figure 7 The method also includes the detector module 110 determining that 702 does not meet (e.g., no longer meets) a predefined condition. Figure 7 The method also includes the detector module 110 instructing the mediator module 112 that a predefined condition is not met. For example, if the detector module 110 indicates that a predefined condition is met by asserting a signal on the communication path to the mediator module 112, indicating that the predefined condition is not met includes releasing the assertion signal. In other embodiments, indicating that the predefined condition is not met includes sending another signal to the mediator module 112 indicating that the predefined condition is not met.
[0050] Figure 7The method also includes the mediator module 112 causing 706 agents 104a-n (e.g., agents 104a-n that have received an instruction to implement a traffic throttling policy) to terminate the traffic throttling policy. For example, in some embodiments, the mediator module 112 sends another signal to the agents 104a-n to terminate the traffic throttling policy. In other embodiments, the mediator module 112 removes the assertion used to indicate that a traffic throttling policy should be implemented.
[0051] To further explain, Figure 8 A flowchart illustrating an exemplary method for dynamic on-chip network traffic throttling according to an embodiment of the present disclosure is provided. Figure 8 Methods and Figure 7 The similarities are, Figure 8 The method includes: determining that 602 meets a predefined condition; sending a signal 604 606 to the mediator module 112; in response to the signal 606, sending an instruction 610 608 to a plurality of agents 104a-n to implement a traffic throttling strategy; determining that 702 does not meet the predefined condition; indicating that 704 does not meet the predefined condition; and causing the plurality of agents 104a-n to terminate the traffic throttling strategy.
[0052] Figure 8 and Figure 7 The difference lies in that the detector module 110 determines 702 that the predefined condition is not met, including determining 804 that the predefined condition is not met within a predefined time period. This predefined time period is programmable or configurable. Therefore, the mediator module 112 is notified of the predefined condition not being met until the predefined time period has elapsed. This reduces the likelihood of the predefined condition oscillating between occurrence and non-occurrence, thereby reducing the oscillations caused by the mediator module 112 between implementing and not implementing the flow throttling strategy in agents 104a-n.
[0053] To further explain, Figure 9 A flowchart illustrating an exemplary method for dynamic on-chip network traffic throttling according to an embodiment of the present disclosure is provided. Figure 9 Methods and Figure 6 The similarities are, Figure 9 The method includes: determining that 602 meets a predefined condition; sending a signal 604 606 to the mediator module 112; and in response to the signal 606, sending an instruction 610 608 to a plurality of agents 104a-n to implement a flow throttling strategy.
[0054] Figure 9 Methods and Figure 6 The difference is that, Figure 9The method also includes having one of the multiple agents 104a-n override the instruction 610 of 902 to implement a traffic throttling policy. In some implementations, overriding the instruction 610 of 902 to implement a traffic throttling policy includes implementing a different traffic throttling policy or not implementing a traffic throttling policy. For example, an agent 104a-n that generates traffic associated with a specific service class will override the traffic throttling policy by implementing a less restrictive traffic throttling policy or not implementing a traffic throttling policy. As another example, an agent 104a-n that implements a traffic throttling policy in response to the number of incomplete transactions meeting a threshold continues to implement its current traffic throttling policy, or implements a more restrictive traffic throttling policy than indicated by the mediator module 112.
[0055] Based on the explanations above, readers will recognize the beneficial effects of tiered throttling for on-chip network traffic, including:
[0056] • Improve computing system performance by allowing distributed agents to use on-chip networks to throttle their traffic.
[0057] • Improve computing system performance by throttling on-chip network traffic based on network performance.
[0058] The exemplary embodiments of this disclosure are described primarily in the context of a full-featured computer system for hierarchical throttling of on-chip network traffic. However, those skilled in the art will recognize that this disclosure can also be embodied in a computer program product disposed on a computer-readable storage medium for use with any suitable data processing system. Such a computer-readable storage medium can be any storage medium for machine-readable information, including magnetic media, optical media, or other suitable media. Examples of such media include disks in hard disk drives or floppy disks, optical disks in optical disk drives, magnetic tapes, and other media conceivable to those skilled in the art. Those skilled in the art will readily recognize that any computer system with appropriate programming means will be able to perform the steps of the methods of this disclosure as embodied in a computer program product. Those skilled in the art will also recognize that while some exemplary embodiments described in this specification are directed to software installed and executed on computer hardware, alternative embodiments implemented as firmware or hardware are also within the scope of this disclosure.
[0059] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium (or medium) having computer-readable program instructions thereon for causing a processor to perform aspects of this disclosure.
[0060] Computer-readable storage media can be tangible devices capable of retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be (e.g., but not limited to) electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A less complete list of more specific examples of computer-readable storage media includes: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanically encoded devices (such as punched cards or raised structures in slots on which instructions are recorded), and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as instantaneous signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0061] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a corresponding computing / processing device, or downloaded via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network) to an external computer or external storage device. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the corresponding computing / processing device.
[0062] Computer-readable program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk and C++, and conventional procedural programming languages such as the "C" programming language or similar programming languages. The computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as a standalone software package, partially on a user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet through an Internet service provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, a field-programmable gate array (FPGA), or a programmable logic array (PLA) may execute the computer-readable program instructions, specifically by using state information from the computer-readable program instructions to personalize the electronic circuitry to perform aspects of this disclosure.
[0063] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0064] These computer-readable program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to obtain a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, form components for implementing the functions / actions specified in one or more blocks of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can instruct a computer, programmable data processing apparatus, and / or other device to function in a particular manner, such that the computer-readable storage medium storing the instructions includes an article of writing containing instructions that implement aspects of the functions / actions specified in one or more blocks of a flowchart and / or block diagram.
[0065] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to obtain a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus or other device, perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0066] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible specific implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions comprising one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions indicated in the blocks may not occur in the order shown in the figures. For example, depending on the functions involved, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may sometimes be executed in reverse order. It will also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.
[0067] As will be understood from the foregoing description, modifications and changes can be made to various embodiments of this disclosure. The description in this specification is for illustrative purposes only and should not be construed as limiting. The scope of this disclosure is defined only by the language of the appended claims.
Claims
1. A method for hierarchical throttling of on-chip network traffic, the method comprising: The number of incomplete transactions is calculated by the agent of the on-chip network, wherein the number of incomplete transactions includes the difference between the number of transactions sent and the number of transactions sent that have not yet received a response; The agent determines that the number of incomplete transactions meets a threshold. as well as In response to the determination, the agent implements a traffic throttling strategy.
2. The method of claim 1, wherein the threshold is included among a plurality of thresholds, the traffic throttling strategy is included among a plurality of throttling strategies, and implementing the traffic throttling strategy includes implementing the traffic throttling strategy based on which of the plurality of thresholds the number of incomplete transactions satisfies.
3. The method according to claim 1, further comprising: Recalculate the number of incomplete transactions; Determine that the number of incomplete transactions has decreased below the threshold; as well as End the aforementioned flow throttling strategy.
4. The method of claim 3, wherein ending the flow throttling strategy includes implementing another throttling strategy.
5. The method of claim 3, wherein ending the traffic throttling strategy comprises issuing a transaction without throttling.
6. The method of claim 3, wherein determining that the number of incomplete transactions has fallen below the threshold includes determining that the number of incomplete transactions has fallen below the threshold by a predefined amount.
7. The method of claim 3, wherein determining that the number of incomplete transactions has fallen below the threshold comprises determining that the number of incomplete transactions has fallen below the threshold within a predefined duration.
8. A chip having tiered throttling for on-chip network traffic, the chip including an on-chip network and one or more agents, the one or more agents being configured to: Calculate the number of incomplete transactions, wherein the number of incomplete transactions includes the difference between the number of transactions sent and the number of transactions sent that have not yet received a response; Determine that the number of incomplete transactions meets a threshold; and In response to the determination, a flow throttling strategy is implemented.
9. The chip of claim 8, wherein the threshold is included in a plurality of thresholds, the traffic throttling strategy is included in a plurality of throttling strategies, and the traffic throttling strategy is implemented based on which of the plurality of thresholds the number of incomplete transactions satisfies.
10. The chip of claim 8, wherein the agent is further configured to: Recalculate the number of incomplete transactions; Determine that the number of incomplete transactions has decreased below the threshold; and End the aforementioned flow throttling strategy.
11. The chip of claim 10, wherein ending the flow throttling strategy includes implementing another throttling strategy.
12. The chip of claim 10, wherein ending the traffic throttling strategy includes issuing a transaction without throttling.
Citation Information
Patent Citations
Mechanism for throttling untrusted interconnect agents
US20180039777A1
