Heterogeneous Server Resource Pooling for Scalable Collaborative Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The integration of multivariate heterogeneous computing power faces challenges due to significant differences in instruction sets between kernels, leading to poor scalability and inefficiencies in collaborative computing.
Innovation Solution
A server system architecture that includes a multivariate computing resource pool, data storage resource pool, and a control module connected via a cache coherence bus, enabling computing power scheduling and dynamic resource allocation among general and heterogeneous computing units, with hardware decoupling and modular design for flexible resource pooling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If heterogeneous computing units with different instruction sets are integrated, then computing power versatility is improved, but collaborative computing efficiency deteriorates due to instruction set differences
Solution Approach 1:
The patent introduces a unified operating system kernel as an intermediary layer between heterogeneous computing units with different instruction sets. This kernel provides standardized interfaces and translation mechanisms that enable different ISA kernels (x86, ARM, RISC-V, etc.) to communicate and collaborate efficiently, resolving the instruction set compatibility issues while maintaining computing power versatility
2Adaptability or versatility
If more heterogeneous computing units are integrated, then computing power integration is improved, but system complexity increases making collaboration difficult
Solution Approach 1:
The patent implements a universal operating system kernel that can manage multiple types of heterogeneous computing units simultaneously. This kernel provides unified resource management, scheduling, and coordination mechanisms that work across different computing unit types, reducing system complexity despite integrating diverse hardware components
3Device complexity
If unified operating system is used for all computing units, then system simplicity is improved, but computing power scalability deteriorates
Solution Approach 1:
The patent segments the operating system into a unified kernel layer and multiple ISA-specific kernel layers. The unified kernel handles common resource management and coordination, while ISA-specific kernels provide instruction set architecture adaptations. This segmentation allows the system to maintain simplicity at the unified management level while achieving scalability through modular ISA-specific implementations
Data Source
Figure 1~2
Figure 3~4
Figure 5
AI summary
Provided in the embodiments of the present disclosure are a server system, a method for scheduling a resource of a server system, and a chip and a die. The server system includes: a multivariate computing resource pool, a data storage resource pool, a Switching Module (SM), and a control module. The multivariate computing resource pool includes a general computing resource pool and a heterogeneous computing resource pool; the general computing resource pool includes a set of general computing units; the heterogeneous computing resource pool includes a set of heterogeneous computing units; the multivariate computing resource pool is connected to the data storage resource pool via the SM by means of a cache coherence bus inside the server system; the data storage resource pool includes one or more data storage resources that are configured to be shared by at least two multivariate computing resources among a plurality of multivariate computing resources in the multivariate computing resource pool; and the control module is configured to perform computing power scheduling on one or more multivariate computing resources among the plurality of multivariate computing resources in the multivariate computing resource pool, and dynamically allocate one or more data storage resources among a plurality of data storage resources in the data storage resource pool to the one or more multivariate computing resources among the plurality of multivariate computing resources in the multivariate computing resource pool.