Word-Level Cache Coherence to Reduce False Sharing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cache coherence schemes in computing systems incur inefficiencies and latency due to false sharing, where multiple cores attempting to modify different words within a cache line necessitate invalidating the entire cache line, leading to unnecessary ownership transfers and performance degradation.
Innovation Solution
Implementing a cache management system that tracks writability and readability at a word granularity, using a modified MESI cache coherence scheme with a 'P' state to indicate partial read/write permissions, and maintaining metadata to differentiate word states, allowing multiple cores to access cache lines with different permission states efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional cache coherence schemes are used to ensure data consistency across multiple cores, then cache line ownership must be transferred exclusively to one core at a time, but this causes false sharing and unnecessary invalidation overhead when multiple cores access different words within the same cache line
Solution Approach 1:
The patent segments the cache line permission management into word-granularity units. Instead of treating the entire cache line as a single ownership unit, each word within the cache line can have independent permission states (readable, writable, or invalid). This segmentation allows different cores to simultaneously access different words with appropriate permissions without requiring full cache line invalidation, thereby resolving the contradiction between maintaining coherence and avoiding false sharing overhead.
2Reliability
If entire cache lines are invalidated when multiple cores access different words, then data consistency is maintained, but performance degrades due to unnecessary ownership transfers and loss of cached data
Solution Approach 1:
The patent applies local quality by assigning different permission states to different words within the same cache line based on local access patterns. Each word can be marked as readable, writable, or invalid independently, allowing the system to maintain data consistency for accessed words while preserving cached data for other words that don't require invalidation. This eliminates the need for full cache line invalidation and reduces ownership transfer latency.
3Productivity
If word-granularity permission tracking is implemented, then false sharing is reduced and simultaneous access is enabled, but metadata complexity and storage requirements increase
Solution Approach 1:
The patent implements partial action by tracking permission states only at the word level when necessary, rather than maintaining full metadata for every possible word configuration. The metadata structure uses compact encoding where each word's permission state is tracked with minimal bits, and the system only actively manages permissions for words that are currently being accessed or modified. This approach enables parallel access efficiency while keeping metadata complexity manageable.
Data Source
AI summary
A system and method for cache management. In some embodiments, a system includes: a lower-level cache; a first upper-level cache; and a second upper-level cache, the lower-level cache, the first upper-level cache, and the second upper-level cache being configured: to store first metadata of a cache line, in the first upper-level cache, the first metadata indicating that the cache line is in a partial state, the partial state indicating that at least two words in the cache line are in different permission states.