Bandwidth weighting mechanism based network-on-chip (NoC) configuration

Abstract

The present disclosure relates to a bandwidth weighting mechanism based NoC configuration/constructions for packet routing. In an aspect, the present disclosure relates to a method for packet routing in a circuit architecture, wherein the method includes the steps of managing, at a router of the circuit architecture, one or more catch-up bits, each of the one or more catch-up bits indicating that the router has reset a round of round-robin based packet routing without allowing an agent corresponding to the each of the one or more catch-up bits to complete its respective round; and allowing, by the router, the agent to continue its respective round in catch-up state such that upon completion of the respective round, the agent is switched to normal state.

Claims

1. A circuit architecture, comprising: one or more routers configured as arbitration points for one or more agents, the one or more routers configured to implement a weighted round-robin arbitration scheme with an early termination circuitry, the early termination circuitry configured to receive an early termination message from an input source and determine an end of a round from the early termination message.

2. The circuit architecture of claim 1, wherein the circuit architecture is implemented as a distributed non-tree network where completion of a round at one of the arbitration points is indicated to the other ones of the arbitration points.

3. The circuit architecture of claim 1, wherein the one or more agents are configured to manage weights of the weighted round-robin arbitration scheme of the one or more routers, and are configured to communicate when the round ends to the arbitration points, wherein the circuit architecture is programmable to configure the weights for each of the one or more agents.

4. The circuit architecture of claim 3, wherein the one or more routers are configured to rely on the early termination signals of the sources to determine when a round has ended, and do not manage weights of the one or more sources.

5. The circuit architecture of claim 1, wherein each of the one or more routers are configured to: track which round each of the one or more agents are operating; for a completion of a round: determine ones of the one or more agents move to a next round and other ones the one or more agents that remain in the round; for the other ones of the one or more agents that remain in the round: allow the other ones of the one or more agents to complete requests; transition the other ones of the one or more agents to the next round; and allocate resources to the other ones of the one more agents for the next round to include remaining resources allocated from the round.

6. The circuit architecture of claim 5, wherein the one or more routers are configured to selectively allow the other ones of the one or more agents to complete requests before processing of requests from the ones of the one or more agents transitioned to the next round.

7. A Network on Chip (NoC) configured with a distributed arbitration system, comprising: one or more routers configured as arbitration points for one or more agents, the one or more routers configured to implement a weighted round-robin arbitration scheme with an early termination circuitry, the early termination circuitry configured to receive an early termination message from a source agent from the one or more agents and determine an end of a round from the early termination message; wherein each of the one or more routers are configured to transmit signals to other ones of the one or more routers indicative of round completion of the each of the one or more routers.

8. The NoC of claim 7, wherein the NoC is implemented as a distributed non-tree network where completion of a round at one of the arbitration points is fanned out to the other ones of the arbitration points.

9. The NoC of claim 7, wherein the one or more agents are configured to manage weights of the weighted round-robin arbitration scheme of the one or more routers, and are configured to communicate when the round ends to the arbitration points, wherein the circuit architecture is programmable to configure the weights for each of the one or more agents.

10. The NoC of claim 9, wherein the one or more routers are configured to rely on the early termination signals of the sources to determine when a round has ended, and do not manage weights of the one or more sources.

11. The NoC of claim 7, wherein each of the one or more routers are configured to: track which round each of the one or more agents are operating; for a completion of a round: determine ones of the one or more agents move to a next round and other ones the one or more agents that remain in the round; for the other ones of the one or more agents that remain in the round: allow the other ones of the one or more agents to complete requests; transition the other ones of the one or more agents to the next round; and allocate resources to the other ones of the one more agents for the next round to include remaining resources allocated from the round.

12. The NoC of claim 11, wherein the one or more routers are configured to selectively allow the other ones of the one or more agents to complete requests before processing of requests from the ones of the one or more agents transitioned to the next round.

Description

BRIEF DESCRIPTION OF DRAWINGS

(1) FIGS. 1A, 1B, 1C, and 1D illustrate examples of Bidirectional ring, 2D Mesh, 2D Taurus, and 3D Mesh NoC Topologies.

(2) FIG. 2A illustrates an example of XY routing in a related art two dimensional mesh.

(3) FIG. 2B illustrates three different routes between a source and destination nodes.

(4) FIG. 3A illustrates an example of a related art two layer NoC interconnect.

(5) FIG. 3B illustrates the related art bridge logic between host and multiple NoC layers.

(6) FIG. 4 illustrates an exemplary representation of a router that forms part of the present invention in accordance with example implementations.

(7) FIG. 5A illustrates an existing (prior art) weighted round robin implementation for packet routing.

(8) FIG. 5B illustrates an exemplary proposed weighted round robin implementation for packet routing.

(9) FIGS. 6A and 6B illustrate how weighted round-robin with early termination based packet routing is performed in accordance with example implementations.

(10) FIGS. 7A and 7B illustrate working of how weighted round-robin with early termination based packet routing and catch-up state based agent operation is performed in accordance in accordance with an example implementation.

(11) FIG. 8 illustrates how arbitration in various functional states is performed for NoC agent request processing in accordance with an example implementation.

(12) FIG. 9 illustrates an exemplary flow diagram using which example embodiments may be implemented.

(13) FIG. 10 illustrates an example computer system on which example embodiments may be implemented.

DETAILED DESCRIPTION

(14) The following detailed description provides further details of the figures and example implementations of the present application. Reference numerals and descriptions of redundant elements between figures are omitted for clarity. Terms used throughout the description are provided as examples and are not intended to be limiting. For example, the use of the term “automatic” may involve fully automatic or semi-automatic implementations involving user or administrator control over certain aspects of the implementation, depending on the desired implementation of one of ordinary skill in the art practicing implementations of the present application.

(15) Network-on-Chip (NoC) has emerged as a paradigm to interconnect a large number of components on the chip. NoC is a global shared communication infrastructure made up of several routing nodes interconnected with each other using point-to-point physical links. In example implementations, a NoC interconnect is generated from a specification by utilizing design tools. The specification can include constraints such as bandwidth/Quality of Service (QoS)/latency attributes that is to be met by the NoC, and can be in various software formats depending on the design tools utilized. Once the NoC is generated through the use of design tools on the specification to meet the specification requirements, the physical architecture can be implemented either by manufacturing a chip layout to facilitate the NoC or by generation of a register transfer level (RTL) for execution on a chip to emulate the generated NoC, depending on the desired implementation. Specifications may be in common power format (CPF), Unified Power Format (UPF), or others according to the desired specification. Specifications can be in the form of traffic specifications indicating the traffic, bandwidth requirements, latency requirements, interconnections, etc. depending on the desired implementation. Specifications can also be in the form of power specifications to define power domains, voltage domains, clock domains, and so on, depending on the desired implementation.

(16) Methods and example implementations described herein are generally directed to quality-of-service (QOS) enhancement pertaining to packet routing in Network-on-Chip (NoC) architectures, and more specifically, to implementation of a bandwidth weighting mechanism based NoC configuration/constructions for packet routing.

(17) Aspects of the present disclosure relate to methods, systems, and computer readable mediums for packet routing in a circuit architecture, wherein, in an aspect, the proposed method includes the steps of managing, at a router of the circuit architecture, one or more catch-up bits, each of the one or more catch-up bits indicating that the router has reset a round of round-robin based packet routing without allowing an agent corresponding to said each of the one or more catch-up bits to complete its respective round; and allowing, by the router, the agent to continue its respective round in catch-up state such that upon completion of said respective round, the agent is switched to normal state.

(18) In an aspect, when a respective router is ending a round, a first set of agents that are in end-of-round state return to said normal state, and a second set of agent that are in normal state are switched to catch-up state. In another aspect, arbiter of the respective router allocates highest priority to agents that form part of the catch-up state. In yet another aspect, each output port of the respective router can be configured as a 2-bit state machine. In another aspect, when the router transmits an end-of round signal back to agents in the catch-up state, the agents extend one or more requests to obtain permissions for catch-up round and subsequent new round.

(19) In an aspect, the present disclosure further relates to a non-transitory computer readable medium storing instructions for executing a process for packet routing in a circuit architecture, the instructions comprising: managing, at a router of the circuit architecture, one or more catch-up bits, each of the one or more catch-up bits indicating that the router has reset a round of round-robin based packet routing without allowing an agent corresponding to the each of the one or more catch-up bits to complete its respective round; and allowing, by the router, the agent to continue its respective round in catch-up state such that upon completion of the respective round, the agent is switched to normal state.

(20) In another aspect, the present disclosure relates to a system operatively coupled with a circuit architecture comprising: a catch-up bits management module configured to manage, at a router of the circuit architecture, one or more catch-up bits, said each of the one or more catch-up bits indicating that said router has reset a round of round-robin based packet routing without allowing an agent corresponding to said each of the one or more catch-up bits to complete its respective round; and a catch-up state based agent round execution module configured to allow, by said router, said agent to continue its respective round in catch-up state such that upon completion of said respective round, said agent is switched to normal state.

(21) In an aspect, the present disclosure relates to a weighted round-robin technique where each router (R) contains an arbiter that tracks the rounds, and wherein the weight of each of agent is provided through early termination signaling from the agents. The routers and the source agents can thereby operate independently with respect to when their round is terminated, and the routers can manage the state of each source agent based on the early termination signaling received from the source agents. Sources utilize the early termination signaling to indicate to the routers when their round has completed. In an example implementation, the router is configured to move to the next round when all of the received requests are for a new round. In this manner, the router can still receive requests from a prior round from input sources that are still operating in the prior round, and then moves to the next round once all of the requests received are for the next round.

(22) In another aspect, the present disclosure further relates to allocation of credits to one or more agents of the NoC such that one credit is used while making a request and no requests are possible after all the credits are consumed. A reset of the credits is done for an agent when an agent completes its round. Upon such completion, the credits (also referred to as credit counters) allocated to that agent are reset. Aspects of the present invention enable dynamic/programmable allocation of credits to one or more agents. Also, instead of routers keeping track of the credits, sources can be configured to keep track of the credits such that when the sources indicate the completion of the round, the allocation of credits to the source agents can be reset. Through such implementations, the length of a round can be determined for each source. The round is measured in the number of packets or requests, and a counter is used to check when all packets have been sent for a round and when the next round starts. Depending on the desired implementation, the counter does not have to be implemented in the agent, but can be implemented in the interface logic or bridge in the network to determine the rounds for the source on its behalf.

(23) Example implementations of the present disclosure also change the manner in which information packets are propagated, wherein instead of managing packet flow across the complete NoC network, flow can be managed/controlled point-to-point. For instance, each router can have its own concept of a round, and can therefore determine when the round is completed, and intimate such round completion signal to the next router. For instance, with reference to the system 400 of FIG. 4, in case Router R1 has two input ports 1 and 2 and two output ports 3 and 4, when rounds of the input ports 1 and 2 are completed (detected by the output (O/P) ports 3 and 4), say by receipt of a packet through input port (I/P) 1, the router R1 can detect that its round is completed, and accordingly intimate the next downstream router (say R2) while sending the packet through any (or both) of the output ports. However, the router does not need to wait for input ports 1 and 2 to be completed to switch rounds; depending on the desired implementation the router can be configured to switch rounds even if the input ports have not completed the rounds yet. The next/second round will then start for R1 when the second packet is received. Therefore, in the present invention, each of the ports (input and output) of a router are well aware, using the above-mechanism, when a round has ended, even if they were not actively involved in processing the packet processing that form part of the round.

(24) Using the above example implementation, routers of the present invention can maintain a table indicating, for each source, whether the source is in the current round, ready for the next round, or has fallen behind by at least a round. As also mentioned above, after completion of each round, when the source finishes its round, the credits allocated to the sources are reset to the default value (say 8 bits).

(25) Example implementations of the present disclosure further relate to weighted round-robin (WRR) with early termination, wherein the rounds are managed by the router, and the sources track weights for a particular round. The sources manage when a round is to end, and although the source may not know when a particular round ends within the router, the source can send an early termination signal to indicate to the router that the sources has finished its round. The router can use the signal to move to the next round if all current requests are ready to proceed to the next round based on determining the number of requests available for an existing round. In an instance, in case there are input ports to a router, each port being connected to an agent/node, wherein the 4 inputs ports respectively are allocated 8, 4, 2, and 1 credits such that in case the second input port/agent indicates that its 2 requests have ended through the use of 2 credits, it can send an early termination signal to the router to indicate that it has finished its respective round, and is ready for the next round whenever all the other inputs ports are.

(26) Aspects of the present disclosure further incorporate a catch-up bit that enables an agent/node to indicate that it had not finished its previous round, and that the arbiter responsible for the agent has moved on, allowing the agent to work in the catch-up state and return back to normal state when its respective round is over, which allows the agent to use the credits from the pending round. In an instance, if there are 4 agents, one allocated 8 credits, one 4 credits, one 2 credits, and one 1 credit, and after completion of the 2 credits of the third agent, if the arbiter changes the round, at which time if the first agent has only used its 3 credits, the first agent can work in the catch-up state till its balance/pending/remaining 5 credits are used and then return back to the normal state.

(27) FIG. 5A illustrates existing (prior art) weighted round robin implementations for packet routing. FIG. 5A illustrates a scenario at 500 where weighted round robin is utilized (e.g., arbitration points are set at ½-½, ⅔-⅓, ¾-¼, etc.) and the resulting bandwidth distribution is used. In the representation 530, when some of the sources become idle, the distribution of bandwidth may become uneven/unfair as illustrated with some agents using a larger share of bandwidth than expected, which can also result in the original ratios of the weighted round robin not being preserved.

(28) FIG. 5B illustrates exemplary proposed weighted round robin implementations 530 for packet routing. In contrast to FIG. 5A, in the weighted round robin with early termination of the present invention, sources can send a signal to indicate an early termination of a round so that the router can be informed that the round ends for a particular source. In circumstances when some sources become idle, the router can thereby end a round for the associated sources without having to fill all of the bandwidth allocation available, and can simply proceed to the next round for the sources. For instance, as shown in representation 570, it can be seen that the round at the destination node ends after 60 credits, thereby preserving the initial ratios while also being work conserving.

(29) In an example implementation, the proposed system can use packet processing based on programmable weights being assigned to nodes. Early completion allows a traffic flow to complete before the credits have expired. A simple extension of this is to allow an agent to get less bandwidth by signaling an early completion sooner. Therefore, if node M0 wanted less traffic, it can indicate early completion to R1 at a count of 10. If M0 indicates an early completion to R1 after 10, it will receive half the bandwidth of the other agents/nodes. It would be appreciated that it may not matter what the weights in the WRR credit are if each port is indicating early completion before the WRR runs out of credits for that port, and therefore if the allocated credits very large, say 1000, an early round completion will always control bandwidth. In such a case, WRR arbiters can be conceived as having infinity credits per port, wherein early termination can transition the count from infinity to zero until the WRR arbiter resets its round. Therefore, instead of implementing a credit counter, we can have a single state bit that indicates whether the request port can issue requests, or is out of credits. Or equivalently, did the input port receive a completion as shown in FIG. 7A as the output VC arbiter knows which port sent the completion. If the VC arbiter tracks which input ports receive a round completion indicator, it can select only among the requests that still have credits (those that haven't sent the completions). Eventually, the arbiter will finish its own round when it either sees that all credits are empty, or when all current requests have no credits. When this occurs, it will send a completion of its own to the next arbiter.

(30) FIG. 6A illustrates a single input port trying to access 2 output ports, wherein a single input port exists and has traffic to two different output ports. The request FIFO has a number of requests, including one that is marked with a completion flag (shown in hashed lines). The requests are sent on the output ports in order, as shown. When the early termination reaches the head of the queue, the first task is for each arbiter to drop this input ports credits to zero. Since there are two output ports, there are two arbiters, each with a separate set of “credit counters”. Both will see the completion flag and update their state. This is shown in the “Comp Rcvd” rows of the cycle chart. In that same cycle, the Output Port 1 arbiter will see that the only available request has no credits available, and will instantly reset the round, granting the request and marking it with a completion flag. While the original request had a completion flag, this should not be construed as propagating that flag. In the next cycle, a request is made for Output Port 0. It is to be appreciated that the completion flag has already removed all credits for this requester in the Output Port 0 arbiter, and this is shown in the FIG. 6A with the “Comp Rcvd” indicator. In this cycle, once again the only request available has no credits. Therefore, the arbiter will reset the round, send a request to that port, and mark it with a completion flag. In this case, the original packet did not have the completion flag set. It is sent with a completion flag because that flag is generated by this arbiter. This means that traffic going down divergent paths will send early terminations in all of the required directions, allowing weights to be preserved even when paths diverge.

(31) FIG. 6B illustrates another example showing 2 input ports trying to access 2 output ports, wherein both queues attempt to access Output Port 0 first, but Request FIFO 1 wins the grant, post which they alternate. However, Request FIFO 0 hits a completion flag quickly, and updates the output port arbiters with the appropriate indicator. For the next 4 cycles, Request FIFO 1 is able to send requests on Output Port 0. Eventually, it hits its own completion flag, and both output port arbiters see that no credits are available. However, since there are active requests for Output Port 0, that state machine is immediately reset and a request is selected. Using round-robin, Request FIFO 0 will win. The packet that gets sent will be marked with completion flag. In the next cycle, Output Port 1 will reset its own round because it has active requests with no available credits. This request also comes from Request FIFO 0, and will be marked with a completion flag even though it wasn't originally marked that way. Meanwhile, Request FIFO 1 will be able to send to Output Port 0. Since the round was reset in the prior cycle, no completion flag is sent. Note that this request was originally marked with a completion flag. Once again, that value is not propagated downstream. It would be appreciated that more complex scenarios work the same way. Each Output Port has an arbiter, and holds “credit” state for each of the input ports. The state machine resets when the only valid requests have no available credits, and on that cycle, a completion flag is sent.

(32) In an exemplary aspect of the present disclosure, if a mode transition happens while an interface is idle, the proposed system adds an extra state that allows the interrupted stream to potentially catch up. When a mode transition occurs and an interface is idle, a “Catch-up State” bit (as shown in FIG. 7B) can be set for that interface, which directs the interface to drop the first completion flag it receives. By dropping the first completion flag, that input stream has the potential to finish its current round, and then participate fully in the new round. Therefore, if a stream intended to be 20 long, and only sent 16 before the round reset, it can potentially catch up by sending 4, a completion flag that will be dropped, and the 20 more. If idle cycles are introduced on an interface for any reason, the bandwidth on that interface isn't halved. It is fully able to catch up and receive its full share of the bandwidth. To increase the potential for an interface to catch-up to the existing round, the arbitration can select interfaces in the catch-up state before interface that aren't in that state.

(33) In an aspect, the present disclosure further enables bridges to associate a weight such that the associated weight gets propagated with the packet so as to get treated accordingly by one or more routers that process the said packet. That is, bridges may be configured to determine the weights instead of the agents, depending on the desired implementation.

(34) FIG. 8 illustrates an exemplary architecture of the proposed implementation showing priority based arbitration based on state of agent that is operatively coupled through input ports of a router. When the agent is in a catch-up state, for instance, one or more requests (REQ0-REQ3, for instance) can be sent by the respective agent based on the credits remaining with the agent. In an aspect, the arbiter allocates highest priority to agents that form part of this state so as to allow them to complete their remaining credits and/or round and return back to the normal state. In the normal state, the agents arbitrate evenly with the other agents. When there are no normal state-based requests being sent i.e. no agents forming part of the normal state, end of round requests can be selected to send to the router, post which the normal round-robin process is followed.

(35) FIG. 9 illustrates an exemplary flow diagram 900 using which example embodiments may be implemented. At step 902, the method includes the step of managing, at a router of the circuit architecture, one or more catch-up bits, each of the one or more catch-up bits indicating that the router has reset a round of round-robin based packet routing without allowing an agent corresponding to said each of the one or more catch-up bits to complete its respective round; and at step 904, the method includes the step of allowing, by the router, the agent to continue its respective round in catch-up state such that upon completion of the respective round, the agent is switched to normal state.

(36) FIG. 10 illustrates an example computer system on which example embodiments may be implemented, specifically, for generation of a NoC/SoC with bridges, and routers configured with arbiters to execute the weighted round robin with early termination schemes as described herein. This example system is merely illustrative, and other modules or functional partitioning may therefore be substituted as would be understood by those skilled in the art. Further, this system may be modified by adding, deleting, or modifying modules and operations without departing from the scope of the inventive concept.

(37) In an aspect, computer system 1000 includes a server 1002 that may involve an I/O unit 1012, storage 1016, and a processor 1004 operable to execute one or more units as known to one skilled in the art. The term “computer-readable medium” as used herein refers to any medium that participates in providing instructions to processor 1004 for execution, which may come in the form of computer-readable storage mediums, such as, but not limited to optical disks, magnetic disks, read-only memories, random access memories, solid state devices and drives, or any other types of tangible media suitable for storing electronic information, or computer-readable signal mediums, which can include transitory media such as carrier waves. The I/O unit processes input from user interfaces 1018 and operator interfaces 1020 which may utilize input devices such as a keyboard, mouse, touch device, or verbal command.

(38) The server 1002 may also be connected to an external storage 1022, which can contain removable storage such as a portable hard drive, optical media (CD or DVD), disk media or any other medium from which a computer can read executable code. The server may also be connected an output device 1024, such as a display to output data and other information to a user, as well as request additional information from a user. The connections from the server 902 to the user interface 1018, the operator interface 1024, the external storage 1016, and the output device 1024 may via wireless protocols, such as the 802.11 standards, Bluetooth® or cellular protocols, or via physical transmission media, such as cables or fiber optics. The output device 1024 may therefore further act as an input device for interacting with a user.

(39) The processor 1004 can include a catch-up bits management module 1026 configured to manage, at a router of the circuit architecture, one or more catch-up bits, each of the one or more catch-up bits indicating that the router has reset a round of round-robin based packet routing without allowing an agent corresponding to said each of the one or more catch-up bits to complete its respective round. The processor 1004 can further include a catch-up state based agent round execution module 1028 configured to allow, by the router, the agent to continue its respective round in catch-up state such that upon completion of the respective round, the agent is switched to normal state. Such modules are utilized to configure and generate the NoC with the functionality as described above.

(40) Unless specifically stated otherwise, as apparent from the discussion, it is appreciated that throughout the description, discussions utilizing terms such as “processing,” “computing,” “calculating,” “determining,” “displaying,” or the like, can include the actions and processes of a computer system or other information processing device that manipulates and transforms data represented as physical (electronic) quantities within the computer system's registers and memories into other data similarly represented as physical quantities within the computer system's memories or registers or other information storage, transmission or display devices.

(41) Example implementations may also relate to an apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes, or it may include one or more general-purpose computers selectively activated or reconfigured by one or more computer programs. Such computer programs may be stored in a computer readable medium, such as a computer-readable storage medium or a computer-readable signal medium. A computer-readable storage medium may involve tangible mediums such as, but not limited to optical disks, magnetic disks, read-only memories, random access memories, solid state devices and drives, or any other types of tangible or non-transitory media suitable for storing electronic information. A computer readable signal medium may include mediums such as carrier waves. The algorithms and displays presented herein are not inherently related to any particular computer or other apparatus. Computer programs can involve pure software implementations that involve instructions that perform the operations of the desired implementation.

(42) Various general-purpose systems may be used with programs and modules in accordance with the examples herein, or it may prove convenient to construct a more specialized apparatus to perform desired method steps. In addition, the example implementations are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the example implementations as described herein. The instructions of the programming language(s) may be executed by one or more processing devices, e.g., central processing units (CPUs), processors, or controllers.

(43) As is known in the art, the operations described above can be performed by hardware, software, or some combination of software and hardware. Various aspects of the example implementations may be implemented using circuits and logic devices (hardware), while other aspects may be implemented using instructions stored on a machine-readable medium (software), which if executed by a processor, would cause the processor to perform a method to carry out implementations of the present disclosure. Further, some example implementations of the present disclosure may be performed solely in hardware, whereas other example implementations may be performed solely in software. Moreover, the various functions described can be performed in a single unit, or can be spread across a number of components in any number of ways. When performed by software, the methods may be executed by a processor, such as a general purpose computer, based on instructions stored on a computer-readable medium. If desired, the instructions can be stored on the medium in a compressed and/or encrypted format.

(44) Moreover, other implementations of the present disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the teachings of the present disclosure. Various aspects and/or components of the described example implementations may be used singly or in any combination. It is intended that the specification and example implementations be considered as examples only, with the true scope and spirit of the present disclosure being indicated by the following claims.