Master-Slave Server Cluster Interconnect Architecture
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing integrated circuit inspection systems face challenges with high nonrecurring engineering costs, lower reliability, and inflexibility due to customized hardware and high or insufficient bandwidth interconnects, leading to excessive costs or inadequate performance.
Innovation Solution
A computing system with a master server having a high-speed data input and a low-speed data input, along with buffer memory, and multiple slave servers connected via a low-speed network, allowing for efficient data partitioning and processing across clusters, optimizing bandwidth and cost.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If high bandwidth interconnects are used in all processors, then data transfer speed is improved, but system cost increases significantly
Solution Approach 1:
The system divides processors into master processors and slave processors, segmenting the high bandwidth interconnect usage to only where it is most needed (master processors and data transfer paths), while slave processors use standard bandwidth connections, thus reducing overall system cost while maintaining high data transfer capability where required
Solution Approach 2:
High bandwidth interconnects are applied locally and selectively to master processors and critical data transfer paths rather than uniformly to all processors, optimizing the balance between performance and cost by providing high speed connectivity only where it delivers maximum value
2Reliability
If customized hardware is used in inspection systems, then specific performance requirements are met, but nonrecurring engineering costs and development time increase
Solution Approach 1:
The system employs standard, commercially available hardware components that can serve multiple inspection applications and be reused across different projects, eliminating the need for expensive customized hardware while maintaining inspection accuracy through software-based solutions and standardized interfaces
3Productivity
If high speed data interconnect is used throughout the system, then bandwidth requirements are met, but the cost becomes excessive
Solution Approach 1:
The data interconnect system is segmented into high speed links for master processors requiring high bandwidth and standard speed links for slave processors, allowing the system to meet overall bandwidth requirements through intelligent data routing while avoiding the excessive cost of providing high speed connectivity to every processor
Solution Approach 2:
Master processors act as intermediaries between the high speed interconnect and slave processors, receiving data through high bandwidth channels and redistributing to slaves through standard channels, thus enabling the system to achieve high overall productivity without requiring every component to have high speed connectivity
Data Source
AI summary
A computing system having at least one cluster. Each cluster has only one master server. The master server has at least one general processing unit, a relatively high speed data input adapted to send and receive data, a relatively low speed data input adapted to send and receive data, and a buffer memory adapted to buffer data between the relatively high speed data input and the relatively low speed data input. Each cluster also has at least one slave server, which has at least one general processing unit, and a relatively low speed data input adapted to send and receive data, the relatively low speed data input having data communication with the relatively low speed data input of the master server. In this manner, the relatively high speed data input, which tends to be very expensive, is present only in the master server, and is used for the high speed transfer of large amounts of data. However, the master server sends the data out to the slave servers on lower speed connections. Thus, there is a blend of both high speed and low speed data communication within the system, and the data is economically processed in clusters within the system.