Unified Cache Reliability via Dynamic Error Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data processing systems face a conflict between performance and reliability due to the inherent tradeoff in cache configuration, where increased sophistication in error management techniques improves reliability but degrades performance, and vice versa, especially in unified caches with varying error rates across different types of cached information.
Innovation Solution
The solution involves configuring an n-way set associative unified cache by allocating cache ways differently for data and instruction-type information, using error detection codes (EDC) and error correction codes (ECC) based on the type of information stored, allowing for a tradeoff between performance and reliability by selecting cache way assignments that balance error rates and performance ratings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If error detection codes (EDC) and error correction codes (ECC) are applied to improve cache reliability, then reliability is improved, but performance degrades
Solution Approach 1:
The patent applies different error management techniques (EDC or ECC) to different cache ways based on their specific error rates. Instead of applying the same error management to all cache ways, the system selectively applies EDC to ways with lower error rates and ECC to ways with higher error rates, optimizing the balance between reliability improvement and performance impact for each local region of the cache.
Solution Approach 2:
The patent dynamically changes the error management parameters by selecting between EDC and ECC based on measured error rates. The system monitors cache error characteristics and adjusts the error management strategy accordingly, changing from simpler EDC to more robust ECC when higher reliability is needed, thereby adapting the reliability-performance tradeoff to actual operating conditions.
2Reliability
If sophisticated error management techniques are used, then reliability is improved, but performance is degraded
Solution Approach 1:
The patent applies error management partially rather than universally. Instead of applying sophisticated ECC to all cache ways, it applies error management only where needed based on measured error rates. This partial application avoids the performance penalty of excessive error management while still providing reliable protection where errors are detected.
3Productivity
If cache configuration is optimized for performance, then performance is improved, but reliability is degraded
Solution Approach 1:
The patent makes the cache configuration dynamic by measuring actual error rates during operation and adjusting error management strategies accordingly. Rather than using a fixed configuration, the system adapts its error management approach based on real-time error characteristics, allowing optimal performance-reliability balance that responds to changing operational conditions.
Data Source
AI summary
A method of configuring a cache includes identifying a plurality of cache configurations of a configurable cache for a processor-executable application unit. Each configuration has an associated error rate. A selected configuration is selected based at least in part on the associated error rate. The configurable cache is configured in accordance with the selected configuration for execution of the application-unit.