Cross-Node Memory Read Routing for CPU-Bypass Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Computer clusters face inefficiencies due to overloaded and underutilized computing nodes, leading to suboptimal processing of burst services.
Innovation Solution
Implement cross-computing-node resource sharing by establishing memory and computing resource pools, allowing nodes to access and utilize resources from other nodes directly, bypassing the central processing unit and operating system for enhanced efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If cross-computing-node data transmission is implemented through traditional CPU and operating system pathways, then data can be transmitted between computing nodes, but transmission efficiency is reduced due to CPU and OS overhead
Solution Approach 1:
The patent extracts the data transmission function from the traditional CPU-OS pathway and implements it directly through device-to-device communication. The first device obtains cross-node read instructions and directly transmits data to the second device without CPU intervention, eliminating the overhead of CPU scheduling and operating system memory management while maintaining data transmission capability between computing nodes
Solution Approach 2:
The patent introduces a correspondence table as an intermediary mechanism that maps virtual addresses to physical addresses and computing node identifiers. This correspondence table enables direct device-to-device transmission by providing the necessary routing information without requiring CPU or operating system involvement in the actual data transmission process
2Productivity
If memory resources are isolated to individual computing nodes, then each node has dedicated memory access, but resource utilization is suboptimal when some nodes are overloaded and others have excess resources
Solution Approach 1:
The patent implements a memory resource pool that aggregates memory resources from multiple computing nodes into a shared pool. The correspondence table enables any computing node to access memory resources from any other node by translating virtual addresses to the appropriate physical address and node identifier, making the memory system universal and adaptable to varying workload demands across the cluster
Solution Approach 2:
The patent merges memory resources from multiple independent computing nodes into a unified memory resource pool. By combining the memory capacities of individual nodes and establishing a shared address space through the correspondence table, the system creates a pooled resource that can be dynamically allocated to any node based on demand, improving overall cluster resource utilization
Data Source
AI summary
A computer system includes a first computing node and a second computing node. The first computing node includes a first device and a first memory. The second computing node includes a second device and a second memory. The first memory includes a first memory space. The second memory includes a second memory space. The first device obtains a cross-node read instruction. The cross-node read instruction includes a virtual address of the second memory space and a size of first data. The first device determines an identifier (ID) of the second computing node based on the virtual address of the second memory space and a first correspondence, to obtain a first network transmission packet, and sends the packet to the second device. The second device receives the packet, reads the first data from the second memory space, and sends the first data to the first device.