Selective Victim Cache Eviction for Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern mobile devices face increased power consumption due to high memory access requests, which limits battery life, as conventional caches often send all capacity evictions to a victim cache, diminishing its utility and increasing power consumption.
Innovation Solution
Implementing a multi-level cache hierarchy with a non-inclusive L3 cache, where the L2 cache uses counters to track access frequencies and traffic levels to selectively drop clean victims instead of writing them back to the L3 cache, reducing unnecessary write-backs during low traffic conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If all capacity evictions are sent to the L3 victim cache, then the L3 cache utility is maximized for data retention, but power consumption increases and bandwidth availability decreases
Solution Approach 1:
The patent implements dynamic victim selection by monitoring traffic conditions and adjusting eviction behavior. The L2 cache transitions between two operational modes: when traffic exceeds a threshold, clean victims are dropped; when traffic is below the threshold, clean victims are sent to L3. This dynamic adaptation resolves the contradiction by optimizing power consumption during low-traffic periods while maintaining L3 utility during high-traffic periods.
Solution Approach 2:
The system changes the parameter of victim cache behavior based on traffic conditions. By monitoring the traffic indicator and adjusting whether to send or drop clean victims, the system optimizes the balance between power consumption and L3 cache effectiveness. This parameter change allows the system to adapt to varying workloads and resolve the fixed contradiction.
2Reliability
If all capacity evictions are sent to the L3 victim cache, then data retention is improved, but bandwidth availability for other agents decreases
Solution Approach 1:
The system dynamically adjusts bandwidth consumption by monitoring traffic conditions. When the traffic indicator exceeds the threshold, the L2 cache drops clean victims locally, preserving bandwidth for other agents. When traffic is below the threshold, clean victims are sent to L3, maintaining data retention utility. This dynamic behavior resolves the contradiction between bandwidth availability and L3 cache effectiveness.
Solution Approach 2:
The victim cache behavior parameter is changed based on traffic conditions. The system monitors the traffic indicator and adjusts whether to send or drop clean victims, optimizing the balance between bandwidth consumption and L3 cache utility. This parameter adaptation resolves the fixed trade-off between bandwidth availability and data retention.
3Adaptability or versatility
If the L2 cache tracks access frequencies using counters, then selective victimization can be implemented, but device complexity increases
Solution Approach 1:
The patent uses simple counters with limited functionality to track traffic conditions. These counters are periodically cleared and restarted, rather than maintaining complex long-term statistics. This approach provides the necessary adaptability for selective victimization while keeping the implementation simple and low-cost, resolving the contradiction between versatility and complexity.
Solution Approach 2:
The system uses simple parameter tracking (counter values) to enable adaptive victim selection. By monitoring basic traffic indicators through counters and adjusting behavior based on threshold comparisons, the system achieves adaptability without requiring complex algorithms or data structures, thus resolving the contradiction between versatility and simplicity.
Data Source
AI summary
Systems, methods, and apparatuses for implementing selective victimization to reduce power and utilized bandwidth in a multi-level cache hierarchy. Each set of an upper-level cache includes a counter that keeps track of the number of times the set was accessed. These counters are periodically decremented by another counter that tracks the total number of accesses to the cache. If a given set counter is below a certain threshold value, clean victims are dropped from this given set instead of being sent to a lower-level cache. Also, a separate counter is used to track the total number of outstanding requests for the cache as a proxy for bus-bandwidth in order to gauge the total amount of traffic in the system. The cache will implement selective victimization whenever there is a large amount of traffic in the system.