Network-on-Chip Rate Limiting for Bounded Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Network on Chip (NoC) systems face challenges in providing a service guarantee due to indeterminate and large latencies, which are incompatible with real-time applications, especially in mesh topology networks with simple routing methods like wormhole routing.
Innovation Solution
Implementing a rate limiting mechanism with a budget of data units allocated to each communication over a reference time interval, suspending communications when their quota is reached, and resuming them at the next interval, while also applying additional rate limitations at intermediate nodes to reduce maximum latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If simple routing methods like wormhole routing are used in NoC, then device complexity is reduced and silicon area is saved, but latency becomes indeterminate and large, making service guarantee incompatible with real-time applications
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing optimal paths in routing tables before communication occurs. The NoC interface unit queries the routing table to obtain predetermined paths, eliminating the need for complex real-time routing decisions while ensuring deterministic latency bounds are met.
Solution Approach 2:
The patent introduces an intermediary routing table that mediates between simple wormhole routing and complex service guarantee requirements. The routing table acts as a pre-computed guide that directs packets along paths guaranteed to meet latency constraints, without requiring complex real-time negotiations or virtual channel management.
2Reliability
If complex service guarantee mechanisms are implemented in NoC routers, then service quality is improved, but device complexity and silicon area requirements increase significantly
Solution Approach 1:
The patent segments the service guarantee function into two parts: complex path calculation is performed offline and stored in routing tables, while the NoC interface unit only performs simple table lookups and path verification. This segmentation moves complexity from the runtime router to the initialization phase, keeping runtime operations simple.
Solution Approach 2:
Service guarantee paths are pre-calculated and stored in routing tables before runtime operations. The NoC interface unit simply queries these predetermined paths, avoiding the need for complex real-time service guarantee mechanisms in the router itself.
3Loss of time
If larger FIFO sizes are used to reduce latency, then latency is improved, but device complexity and silicon area increase
Solution Approach 1:
The patent pre-calculates optimal paths and transmission timing in the routing table, allowing smaller FIFO sizes to suffice. By knowing the predetermined path and timing constraints in advance, the system can use minimal buffering while still guaranteeing latency bounds, eliminating the need for large FIFOs.
Data Source
AI summary
The method involves allocating fixed paths to communications to be established on a mesh network, and identifying the communications to take up a mesh segment (N12). Respective throughput quotas (4, 8) are allocated to the identified communications such that the sum of the quotas is less than or equal to a maximum throughput quota (16) of the segment. Throughput of each communication is measured at input of the network, and the communication is suspended when the quota is reached. A budget of data units is allocated to the mesh segment over a reference time interval. An independent claim is also included for a mesh network comprising source nodes to initiate communications taking fixed paths to destination nodes.