Multi-Tier Access Tracking for Accurate Hot-Cold Page Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing access tracking mechanisms in multi-tier memory systems face limitations in accuracy due to limited data structure sizes, which restrict their ability to track memory accesses accurately as the number of unique pages increases over time.
Innovation Solution
Implementing multiple access trackers, including a first tracker that increments access counts and sorts page IDs based on access counts, and a second tracker that uses a hash function to manage access counts and transfer data between trackers, allowing for more efficient determination of hot and cold pages.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single access tracker with limited data structure size is used, then device complexity is reduced, but measurement precision of access counts deteriorates as the number of unique pages increases
Solution Approach 1:
The access tracking system is divided into multiple independent access trackers, each responsible for tracking a subset of pages. This segmentation allows the system to maintain high measurement precision for access counts across a large number of unique pages while keeping each individual tracker's data structure size manageable and device complexity controlled.
Solution Approach 2:
The patent introduces a hierarchical dimension to access tracking by organizing trackers in levels (e.g., L1 trackers for frequently accessed pages, L2 trackers for less frequently accessed pages). This dimensional organization allows the system to track access counts for a vast number of pages without requiring a single large data structure, thus maintaining measurement precision while controlling device complexity.
2Measurement precision
If multiple access trackers are implemented to improve access count accuracy, then measurement precision improves, but device complexity increases
Solution Approach 1:
Each access tracker is designed with specialized local characteristics optimized for its specific function. For example, L1 trackers are optimized for tracking frequently accessed pages with high access count precision, while L2 trackers handle less frequently accessed pages. This local optimization allows the system to achieve high overall measurement precision without uniformly increasing the complexity of all trackers.
Solution Approach 2:
The system dynamically manages the distribution of page tracking across multiple access trackers based on access patterns. Pages can be migrated between trackers as their access frequency changes, allowing the system to adapt to varying workloads and maintain measurement precision without requiring a static complex structure for all possible scenarios.
3Ease of operation
If access tracker data structure size is limited to reduce device complexity, then ease of operation is improved, but loss of information occurs when access counts overflow
Solution Approach 1:
The patent introduces intermediary structures and mechanisms between access trackers and the main memory system. These intermediaries include overflow handling protocols and data transfer mechanisms between L1 and L2 trackers, which prevent information loss by systematically managing cases where access counts would otherwise overflow the limited data structure size of individual trackers.
Data Source
AI summary
An access tracker configured to receive a request to access a page, determine whether a page identification (ID) associated with the page is in the access tracker, increment an access count of the page in response to determining the page ID is in the access tracker, sort a number of page IDs based on an access count of each page ID, and determine whether a different page is hot or cold in response to sorting the number of page IDs.