Preemptive Decentralized Cache Line Tracking for Fair Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cache line access mechanisms in computing environments with multiple processor chips and shared caches face scalability issues, leading to unfair access and prolonged wait times for requestors, impacting performance and response times.
Innovation Solution
Implement a decentralized, hierarchical hot cache line fairness mechanism using requested and serviced vectors that track cache line requests and access status, preemptively building a structure to ensure fair access for all requestors, reducing the need for centralized storage space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a centralized mechanism is used to track cache line requests, then complete tracking information can be maintained, but scalability issues arise and large physical space is required
Solution Approach 1:
The patent divides the centralized tracking mechanism into decentralized components distributed across multiple caches. Each cache maintains local request tracking information independently, eliminating the need for a single centralized tracker. This segmentation allows the system to scale horizontally by adding more caches without proportionally increasing central tracking complexity.
Solution Approach 2:
The patent introduces hierarchical levels of tracking (local cache level, drawer level, and system level) to manage request information. By organizing tracking across multiple dimensions rather than a single centralized point, the system achieves both complete tracking and improved scalability through distributed responsibility.
2Ease of operation
If a centralized structure is used to track all cache line requests, then fair access can be ensured, but large centralized physical space is required
Solution Approach 1:
The patent distributes the fairness tracking functionality across multiple independent caches rather than concentrating it in a single centralized structure. Each cache maintains its own request tracking state, reducing the physical space required at any single location while collectively ensuring fair access through the distributed network of caches.
Solution Approach 2:
The patent creates replicated copies of tracking functionality at each cache level rather than maintaining a single master copy. This allows fairness to be enforced locally at each cache while reducing the need for large centralized storage, as each cache only needs to maintain its own local state.
3Loss of time
If proactive prediction of coherence operations is implemented, then access time can be reduced, but system complexity increases
Solution Approach 1:
The patent implements predictors that proactively anticipate future coherence operations based on historical access patterns. By performing coherence operations in advance before they are actually needed, the system reduces latency. The predictors are integrated into the existing cache controller logic, minimizing additional complexity while achieving proactive optimization.
Solution Approach 2:
The patent uses feedback from historical coherence operation data to train and refine predictors. The system monitors actual access patterns and uses this information to improve prediction accuracy over time, allowing proactive operations to be targeted precisely when beneficial, thereby reducing unnecessary complexity.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments are for preemptive tracking of remote requests for decentralized hot cache line fairness tracking. Authority is requested for a cache line in conjunction with querying for outstanding requests for the cache line. One or more responses are received regarding the outstanding requests for the cache line. In response to receiving the one or more responses regarding the outstanding requests and in advance of receiving the authority for the cache line, the outstanding requests are preemptively tracked in a requested structure associated with the cache line.