Probabilistic Distance-Based Arbitration for Chip Multiprocessors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In many-core chip multiprocessors, existing arbitration methods fail to provide equal and fair access to shared resources, leading to unfair bias and variance in execution time among processing cores due to their location within the chip, which affects memory and I/O access performance.
Innovation Solution
The implementation of probabilistic arbitration combined with distance-based weights, using nonlinear weights such as fixed, constantly increasing, and variably increasing weights, to determine the order of data packet routing and ensure equal access to shared resources without requiring additional buffers or virtual channels, thereby approximating age-based arbitration for fairness and stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional arbitration methods are used in on-chip networks, then the arbitration logic is simple, but processing cores experience unfair bias and variance in execution time due to their location
Solution Approach 1:
The patent changes the arbitration parameter from simple round-robin or priority-based selection to probabilistic selection with dynamically adjusted weights. The weight parameter is modified based on distance metrics, allowing the system to compensate for location-based unfairness without requiring complex additional hardware structures.
Solution Approach 2:
The arbitration weights are made dynamic rather than static. The system continuously adjusts the weights based on current distance calculations between processing cores and shared resources, enabling the arbitration logic to adapt to changing traffic patterns and core locations without requiring reconfiguration of the hardware architecture.
2Ease of operation
If distance-based probabilistic arbitration is implemented, then fair service is achieved across all cores, but the arbitration logic becomes more complex
Solution Approach 1:
The patent applies local quality by making each arbitration decision locally optimal based on the specific distance metrics of the competing requests. Instead of a global arbitration scheme, each arbitration event is tailored to the local configuration of requesting cores and their distances to the shared resource, providing fairness where it is most needed.
Solution Approach 2:
The distance metric serves as an intermediary that translates physical location information into arbitration weights. This intermediary layer allows the system to achieve fair resource allocation without direct complex interactions between all processing cores, simplifying the overall arbitration mechanism while maintaining fairness.
3Reliability
If sophisticated arbitration techniques are used to provide equal service, then fairness is improved, but per-flow state and queues become impractical in on-chip networks
Solution Approach 1:
The patent extracts the essential fairness requirement from complex per-flow state management schemes and implements it through simple distance-based weight calculations. By removing the need to maintain per-flow state and queues, the system achieves service equality using only the core distance metric and probabilistic selection logic.
Solution Approach 2:
Instead of maintaining expensive per-flow state information that must be preserved across multiple arbitration events, the system uses disposable distance calculations that are computed fresh for each arbitration event. This approach trades the storage and management of persistent state for lightweight, transient calculations that are cheaper and more suitable for on-chip implementation.
Data Source
AI summary
Probabilistic arbitration is combined with distance-based weights to achieve equality of service in interconnection networks, such as those used with chip multiprocessors. This arbitration desirably used incorporates nonlinear weights that are assigned to requests. The nonlinear weights incorporate different arbitration weight metrics, namely fixed weight, constantly increasing weight, and variably increasing weight. Probabilistic arbitration for an on-chip router avoids the need for additional buffers or virtual channels, creating a simple, low-cost mechanism for achieving equality of service. The nonlinearly weighted probabilistic arbitration includes additional benefits such as providing quality-of-service features and fairness in terms of both throughput and latency that approaches the global fairness achieved with age-base arbitration. This provides a more stable network by achieving high sustained throughput beyond saturation. Each router or switch in the network may include an arbiter to apply the weighted probabilistic arbitration.