Multi-Protocol Optical Switches for Pooled Data Center Resources
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face inefficiencies due to latency and bandwidth limitations in twisted-pair copper cabling, leading to underutilization of physical resources and inability to manage diverse workload types, as networking components are not equipped to handle both high-performance computing and other network traffic.
Innovation Solution
Implementing a data center design with shallower sleds for improved airflow, optical fiber connections for high-bandwidth and low-latency networking, and multi-mode optical switches that support multiple link-layer protocols, enabling resource pooling and dynamic reallocation based on workload demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If optical fiber connections are implemented, then bandwidth and latency performance is improved, but device complexity and manufacturing difficulty increase
Solution Approach 1:
The network switch is designed to handle multiple types of network traffic simultaneously - both HPC protocols (InfiniBand, Omni-Path) and standard Ethernet protocols - through a unified optical switching fabric. This multi-functionality allows the same infrastructure to serve diverse workload types without requiring separate specialized networks, thereby achieving high performance without proportionally increasing complexity
Solution Approach 2:
The patent introduces an optical switching fabric as an intermediary layer between compute devices and pooled hardware resources. This optical fabric mediates communication between components, providing high-speed connectivity while abstracting the complexity of direct optical connections from individual devices. The switching fabric handles protocol conversion and routing, simplifying the overall system architecture
2Speed
If physical hardware resources are included locally in each compute device, then speed of access is improved, but adaptability and resource utilization deteriorate
Solution Approach 1:
The patent segments hardware resources (GPUs, FPGAs, storage, memory) from compute devices and organizes them into separate pooled resource modules. This segmentation allows resources to be independently managed and dynamically allocated to different compute devices based on workload demands, improving both adaptability and utilization while maintaining fast access through optical connectivity
Solution Approach 2:
The patent transitions from a traditional one-to-one mapping between compute devices and hardware resources to a many-to-many relationship enabled by the optical switching fabric. This dimensional change in resource allocation allows any compute device to access any pooled resource dynamically, achieving high resource utilization while maintaining performance through the high-speed optical network
3Productivity
If networking components are specialized for HPC protocols, then productivity for HPC workloads is improved, but adaptability to handle other network traffic deteriorates
Solution Approach 1:
The network switch incorporates multiple protocol support within a single device, including HPC protocols (InfiniBand, Omni-Path) and standard Ethernet protocols. The optical switching fabric is designed to handle different protocol types simultaneously, allowing the same infrastructure to optimize HPC workloads while also supporting general-purpose networking, thereby achieving both high productivity and adaptability
Solution Approach 2:
The patent applies different protocol handling capabilities to different ports or pathways within the network switch. HPC-optimized pathways use specialized protocol processing for maximum performance, while other pathways handle standard Ethernet traffic. This local differentiation allows the system to maintain high productivity for HPC workloads while preserving adaptability for diverse network traffic types
Data Source
AI summary
Technologies for switching network traffic include a network switch. The network switch includes one or more processors and communication circuitry coupled to the one or more processors. The communication circuitry is capable of switching network traffic of multiple link layer protocols. Additionally, the network switch includes one or more memory devices storing instructions that, when executed, cause the network switch to receive, with the communication circuitry through an optical connection, network traffic to be forwarded, and determine a link layer protocol of the received network traffic. The instructions additionally cause the network switch to forward the network traffic as a function of the determined link layer protocol. Other embodiments are also described and claimed.