Programmable Switch Interconnect for Network Adaptability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Datacenters face challenges in efficiently configuring and managing diverse tenant needs for compute resources, network bandwidth, and fault tolerance across various network architectures, including LAN, WAN, and SAN, while maintaining locality and scalability.
Innovation Solution
A programmable switch interconnect system using circuit and packet switches, controlled by a programmable controller, which configures uplink and downlink ports, bandwidth allocations, and failover capabilities to create customized network topologies such as folded Clos networks, allowing for efficient resource allocation and fault tolerance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If diverse network configurations are provided for different tenant needs, then adaptability is improved, but device complexity increases
Solution Approach 1:
The switch assembly is designed as a universal platform that can be dynamically reconfigured to serve multiple tenant needs and network architectures. The circuit switches and packet switches work together to provide various network topologies (LAN, WAN, SAN, folded Clos networks) through a single multi-functional device, eliminating the need for separate specialized switches for each tenant requirement.
Solution Approach 2:
The system employs dynamic reconfiguration capabilities where the controller can programmatically adjust the switch fabric connections in real-time. This allows the same physical switch assembly to adapt its topology and connectivity to match changing tenant requirements, workload demands, and network conditions without requiring manual reconfiguration or hardware changes.
2Productivity
If network topology is reconfigured to maintain locality, then network performance is improved, but reconfiguration time increases
Solution Approach 1:
The system pre-establishes multiple switch fabric configurations and connection profiles that are optimized for different workload types and tenant requirements. When a reconfiguration is needed, the controller can rapidly switch between pre-configured states rather than building configurations from scratch, significantly reducing reconfiguration time while maintaining optimal network performance.
Solution Approach 2:
The reconfiguration process is designed to minimize service disruption by maintaining active connections during topology changes. The controller orchestrates switch fabric reconfiguration in a way that keeps data paths active throughout the transition, ensuring continuous network operation and preventing performance degradation during reconfiguration events.
3Productivity
If bandwidth allocation is increased for specific tenants, then network performance is improved, but loss of energy increases
Solution Approach 1:
The system dynamically adjusts bandwidth allocation parameters based on actual network conditions, tenant priorities, and workload demands. Rather than provisioning fixed high-bandwidth connections that consume energy continuously, the controller modifies bandwidth parameters in real-time to match actual needs, ensuring high performance when required while reducing energy consumption during low-utilization periods.
Data Source
Figure 1
Figure 2
Figure 3A~3B
AI summary
A technique includes using circuit switches to selectively couple packet switches of a switch assembly to the port connectors of the assembly.