Recency Score Filter for Memory Page Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory systems face challenges in efficiently tracking memory accesses at a granularity that is practical, as the resources and overhead required to do so for billions of pages overwhelm the system, leading to increased latency and reduced capacity.
Innovation Solution
The implementation of multiple filters, including a recency score filter and a bloom filter, allows for tracking memory accesses at different granularities, reducing the overhead and resources needed by only updating the bloom filter for hot regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If memory accesses are tracked at fine granularity for all pages, then measurement precision is improved, but device complexity and resource overhead increase significantly
Solution Approach 1:
The patent divides memory into regions and uses multiple filters (recency score filter and bloom filter) at different granularities to track accesses. The recency score filter operates at a coarser granularity to identify hot regions, while the bloom filter operates at a finer granularity only for those hot regions, thus achieving fine-grained tracking where needed without system-wide overhead.
Solution Approach 2:
The patent applies different tracking mechanisms to different memory regions based on their access patterns. Hot regions (identified by the recency score filter) receive fine-grained tracking via the bloom filter, while cold regions use coarser tracking, optimizing resource usage according to local access characteristics.
2Loss of energy
If recency score filters are implemented, then power consumption and hardware complexity are reduced, but false identification rates may increase
Solution Approach 1:
The recency score filter serves as an intermediary between coarse memory regions and the fine-grained bloom filter. It pre-processes access patterns to identify hot regions, reducing the burden on the bloom filter and enabling the system to use lower-power coarse tracking for most regions while maintaining accurate fine-grained tracking for hot regions.
Solution Approach 2:
The system applies fine-grained bloom filter tracking only partially - specifically only to hot regions identified by the recency score filter, rather than to all memory regions. This partial application reduces overall power consumption and hardware complexity while maintaining reliability for the critical hot regions.
Data Source
AI summary
A memory buffer device and memory module for accurate hot and cold page detection is disclosed. The memory buffer device is coupled to a device memory including a plurality of regions. The memory buffer device identifies frequently accessed regions of the plurality of regions by counting accesses to the plurality of regions at a first granularity. The frequently accessed regions are associated with counters that satisfy a threshold criterion and are further tracked using a filter at a second granularity. The second granularity is smaller than the first granularity.