Multi-Level Cache Partitioning by Data Hotness for Lower Overhead

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The challenge of balancing high cache hit rates with reduced management overheads becomes critical as cache capacities increase, particularly in storage systems with limited resources, due to the high costs associated with cache technologies.

Innovation Solution

Implementing a cache management method that segregates data into at least two levels of storage, with each level having distinct hotness value intervals, and performing swap-out operations at the granularity of storage areas to ensure data with similar hotness values are stored together, thereby avoiding mixing of cold and hot data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If cache capacity is increased to meet large-scale high-performance data storage requirements, then the storage capability is improved, but the management overheads increase and costs rise

Engineering Contradiction:
Improvecache capacityVSAvoidmanagement overheads
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The first-level storage is divided into multiple storage areas, each dedicated to storing data within a specific hotness value interval. This segmentation allows the system to manage different hotness levels independently, reducing the complexity of managing a monolithic cache structure while maintaining high cache capacity for large-scale data storage.

Inventive Principle:
Principle #1Segmentation

2Adaptability or versatility

If data is swapped out at the granularity of individual data items, then the flexibility of cache management is improved, but the management overheads increase significantly

Engineering Contradiction:
Improvecache management flexibilityVSAvoidmanagement overheads
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The cache management unit is set to the storage area level rather than individual data items. Each storage area contains multiple data items with similar hotness values, allowing the system to manage cache at a coarser granularity. This reduces the number of management operations required while maintaining adaptability through hotness-based storage area selection.

Inventive Principle:
Principle #1Segmentation

3Quantity of substance

If storage areas contain mixed hot and cold data, then the utilization of cache space is improved, but the cache hit rate decreases

Engineering Contradiction:
Improvecache space utilizationVSAvoidcache hit rate
Core Design Contradiction:
Quantity of substanceVSReliability

Solution Approach 1:

Each storage area is assigned a specific hotness value interval, creating local quality differentiation within the cache structure. Data items are placed in storage areas matching their hotness characteristics, ensuring that hot data resides in designated hot storage areas and cold data in cold storage areas. This local quality approach maintains high cache space utilization while preserving high cache hit rates for frequently accessed data.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP4290354B1Cache management method and storage device
Publication Date: 2025.12.31 HUAWEI TECH CO LTD
  • EP4290354B1 patent drawingFigure 1~2
  • EP4290354B1 patent drawingFigure 3
  • EP4290354B1 patent drawingFigure 4

AI summary

This application provides a cache management method and a storage device. The storage device includes at least two levels of storages. The at least two levels of storages include a first-level storage and a second-level storage, and a speed of processing data by the first-level storage is higher than that of the second-level storage. The first-level storage includes at least a first storage area and a second storage area, hotness values of all pieces of data stored in the first storage area are within a first hotness value interval, hotness values of all pieces of data stored in the second storage area are within a second hotness value interval, and the data in the first-level storage is swapped out at a granularity of one storage area. In the foregoing manner, data with same or similar hotness values may be stored in a same storage area, so that a case in which the data that is swapped out is mixed with cold data and hot data can be avoided, a hit rate can be improved, and management overheads can be reduced by performing cache management at a granularity of a storage area.