Cloud Control Device Memory Access Method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed Shared Memory (DSM) technologies face high operation overheads and implementation costs due to software layer or hardware-dependent cache consistency management, limiting flexible resource scheduling in dynamic load changes within data centers.
Innovation Solution
A memory access method that uses a cloud control device to manage requests between nodes, employing mapping relationships to efficiently route data access requests, reducing the need for direct node-to-node communication and optimizing resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If DSM is implemented by OS or user programming library at software layer, then shared memory functionality is achieved, but operation overheads and implementation costs are relatively high
Solution Approach 1:
The patent introduces a network adapter as an intermediary hardware component that maintains cache consistency, eliminating the need for software-based DSM implementation. The network adapter acts as a mediator between nodes, handling address translation and cache coherence protocols hardware-level, thereby reducing operation overheads and implementation costs while maintaining shared memory functionality.
Solution Approach 2:
The patent replaces the software-based DSM mechanism with a hardware-based implementation using network adapters. This substitution moves the DSM functionality from the software layer to the hardware layer, utilizing physical components (network adapters with integrated address translation and cache consistency mechanisms) to replace software processes, thereby reducing operational overhead.
2Ease of operation
If DSM performs unified addressing on memory, then a single image is provided for programmers, but data consistency maintenance results in inability to favorably support flexible resource scheduling
Solution Approach 1:
The patent segments the address translation and memory management functions by implementing them in hardware at the network adapter level rather than through centralized software control. This segmentation allows each node to independently manage its memory resources while maintaining unified addressing through hardware-based address translation, enabling both programmer simplicity and flexible resource scheduling.
Solution Approach 2:
The patent introduces dynamic resource allocation capabilities by implementing hardware-based address translation that can adapt to changing workload requirements. The network adapter's address translation mechanism can dynamically adjust mapping relationships based on real-time resource availability and demand, enabling flexible resource scheduling while maintaining the unified addressing interface for programmers.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Embodiments of the present invention provide a memory access method, device, and system, which can allocate, to a node in a system, a memory resource on which an exclusive access operation can be performed, reduce maintenance overheads of the system, and implement flexible scheduling of memory resources. The device includes a cloud control device on the side of a requesting node and a cloud control device on the side of a contributing node. The cloud control device on the side of the requesting node generates a request packet for accessing to-be-accessed data stored in a memory resource provided by the contributing node to the requesting node for use, receives a response packet, and provides a response message to the requesting node. The cloud control device on the side of the contributing node receives the request packet, provides a request message to the contributing node, generates the response packet for the contributing node, and sends the response packet to the requesting node.