Reconfigurable Server Nodes via Midplane Fabric Interconnects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data center operators face challenges in predicting the optimal server configuration between two-socket (2S) and four-socket (4S) servers for application performance, as the best configuration can vary depending on the application and is difficult to predict for existing or future applications.
Innovation Solution
A computing system with reconfigurable server nodes connected via a midplane that enables or disables inter-processor connections, allowing the same hardware to be configured as either two independent 2S servers or a single 4S server, with a management interface for administrators to scale up or out based on performance needs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a data center operator purchases both 2S and 4S servers to handle different application performance needs, then application performance requirements can be met, but hardware asset utilization is reduced and costs increase
Solution Approach 1:
The server configuration is made dynamic through reconfigurable fabric interconnects that can change connectivity topology between server nodes. The system transitions from a static hardware configuration to a dynamic one where the same physical hardware can be reconfigured between 2S and 4S modes, allowing adaptation to different application performance requirements without purchasing separate server types.
Solution Approach 2:
The server nodes are designed with universal interfaces and reconfigurable fabric interconnects that enable the same hardware to perform multiple functions. The system can function as either two independent 2S servers or one consolidated 4S server depending on the application requirements, maximizing hardware asset utilization while meeting diverse performance needs.
2Ease of manufacture
If the server configuration is fixed at deployment, then hardware setup is simple, but the configuration cannot adapt to changing application demands over time
Solution Approach 1:
The fabric interconnects incorporate reconfigurable switching fabric that can dynamically change connectivity patterns. The system maintains simple initial deployment but gains the ability to reconfigure between 2S and 4S modes as applications change, resolving the contradiction between ease of setup and future adaptability.
Solution Approach 2:
The server nodes are pre-equipped with reconfigurable fabric interconnects and universal interfaces during manufacturing, preparing them for future reconfiguration needs. This preliminary setup maintains ease of initial deployment while enabling later adaptation to changing application demands without complex hardware modifications.
3Adaptability or versatility
If 4S servers are used for all applications, then hardware asset utilization is maximized, but applications that perform better on 2S servers cannot achieve optimal performance
Solution Approach 1:
The system dynamically selects between 2S and 4S configurations based on application-specific performance requirements. The reconfigurable fabric interconnects enable the same hardware to be optimized for different workloads, ensuring that applications receive the configuration that delivers optimal performance while maintaining high hardware utilization across the data center.
Data Source
AI summary
Embodiments herein describe a computing system which is reconfigurable into different server configurations that have different numbers of sockets. For example, the computing system may include two server nodes which can be configured into either two independent servers (i.e., two 2S servers) or a single server (i.e., one 4S server). In one embodiment, the computing system includes a midplane which is connected to processor buses on the server nodes. When configured as a single server, the midplane connects the processor bus (or buses) on one of the server nodes to the processor bus or buses on the other server node. In this manner, the processors in the two server nodes can be interconnected to function as a single server. In contrast, the connections between the server nodes in the midplane are disabled when the server nodes operate as two independent servers.