Hybrid Memory Page Management for Latency and Power Trade-offs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for data processing involving large datasets result in inefficiencies due to high overhead and power consumption, particularly when only a small portion of a page is used, as they rely on swap processing between main memory and storage devices with high access latency, leading to reduced performance and increased energy usage.
Innovation Solution
An information processing apparatus that utilizes a combination of high-speed DRAM as main memory and large-capacity, low-power nonvolatile memory (NVM) for data storage, employing a dual-access strategy: moving frequently accessed pages to DRAM via swap processing and directly accessing infrequently used pages in NVM to optimize data transfer and reduce power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If swap processing is performed between main memory and storage device, then data can be accessed that exceeds main memory capacity, but access latency increases and processing efficiency decreases
Solution Approach 1:
The patent segments data access into two categories: frequently accessed pages are loaded into main memory for fast access, while infrequently accessed pages remain in storage device. This segmentation allows the system to maintain small working sets in fast memory while keeping large datasets in slow storage, resolving the contradiction between storage capacity and access latency.
Solution Approach 2:
The patent applies local quality by differentiating access patterns for different pages. Hot pages (frequently accessed) are kept in main memory with high-speed access characteristics, while cold pages (infrequently accessed) remain in storage device with large-capacity characteristics. This localized optimization resolves the contradiction by giving different parts of the data different access locations based on their access frequency.
2Speed
If entire pages are loaded into main memory for access, then fast access is achieved, but power consumption increases due to unnecessary data transfer
Solution Approach 1:
The patent applies partial action by loading only the necessary portion of data into main memory based on access frequency analysis. Instead of loading entire pages or all data, it selectively loads only frequently accessed pages, achieving fast access for critical data while avoiding unnecessary power consumption from transferring and maintaining unnecessary data in main memory.
Solution Approach 2:
The patent changes the parameter of data location based on access frequency. By monitoring access patterns and dynamically adjusting which pages reside in main memory versus storage device, the system optimizes the balance between access speed and power consumption. This parameter change allows the system to adapt to varying workload characteristics and minimize unnecessary data transfers.
Data Source
AI summary
According to an embodiment, an information processing apparatus includes a processing device, a first memory, and a second memory. The processing device executes first processing on first data. The second memory stores the first data and has a data access latency higher than that of the first memory. The first data includes first and second pages, the first page being read/written times not less than a threshold in a certain period shorter than a period for executing the first processing, the second page being read/written times less than the threshold in the certain period. The processing device includes a controller configured to execute first access to move the first page to the first memory and then read/write data from/to the moved first page, and execute second access to directly read/write data from/to the second page of the second memory.