RDMA Fabric Controller for Dynamic Congestion Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network fabrics, such as RDMA networks, face challenges in achieving optimal transport efficiency in multi-tenant environments due to static hardware configurations and the inability to dynamically address hardware optimizations, limitations, health, and configuration.
Innovation Solution
A real-time feedback control mechanism is implemented to dynamically configure hardware components within the network fabric based on performance metric data and flow information, allowing for optimal transport efficiency across multiple tenants.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If hardware components in network fabric are configured statically prior to deployment, then device complexity is reduced and ease of manufacture is improved, but adaptability to different workloads and multi-tenant environments deteriorates
Solution Approach 1:
The patent implements dynamic reconfiguration of hardware components (switches, routers, network interface cards) in the network fabric by introducing a controller that receives feedback about network conditions and workload requirements, then modifies operational parameters of hardware components in real-time. This transforms the static configuration into a dynamic system that adapts to changing multi-tenant workload demands without requiring physical reconfiguration or redeployment.
Solution Approach 2:
The controller changes operational parameters of hardware components based on feedback information about network congestion, workload characteristics, and performance metrics. By modifying parameters such as bandwidth allocation, routing paths, and quality of service settings dynamically, the system achieves adaptability to different workloads while maintaining the same physical hardware infrastructure.
2Productivity
If network fabric is optimized for a single tenant or specific application, then transport efficiency is improved for that tenant, but adaptability to multiple tenants with competing priorities deteriorates
Solution Approach 1:
The network fabric controller provides universal management capabilities that serve multiple tenants with different workload requirements through a single system. The controller analyzes feedback from various tenants and dynamically allocates resources to optimize transport efficiency for each tenant's specific application needs while maintaining overall network performance, making the infrastructure universally adaptable to diverse workloads.
Solution Approach 2:
The system applies different configuration policies and optimization strategies to different segments of the network fabric based on local workload requirements. Each tenant's traffic flow receives customized handling with appropriate quality of service parameters, routing priorities, and bandwidth allocations tailored to their specific application needs, while the overall network maintains efficient operation across all tenants.
3Reliability
If ECN marking and notification packets are used to manage congestion, then packet loss is reduced, but device complexity and protocol overhead increase
Solution Approach 1:
The patent introduces a controller as an intermediary that centralizes congestion management functions. Instead of requiring complex ECN marking and notification packet exchanges between hosts and network devices, the controller receives feedback about network conditions and directly manages congestion by adjusting hardware component parameters, simplifying the protocol stack while maintaining lossless transmission through centralized control.
Solution Approach 2:
The system implements a feedback mechanism where the controller continuously monitors network conditions and workload requirements, then uses this feedback to dynamically adjust congestion management strategies. This feedback loop enables the controller to proactively manage congestion before packet loss occurs, replacing complex reactive ECN protocols with simpler proactive control based on real-time network state information.
Data Source
AI summary
Described herein is a controller that is communicatively coupled with a network fabric. The controller obtains performance metric data of one or more hardware components included in the network fabric. The controller collects flow information of one or more workloads that are executed on the network fabric. Further, the controller applies a configuration policy to the one or more hardware components of the network fabric based on the performance metric data and the flow information of the one or more workloads. The application of the configuration policy modifies at least one operational parameter of the one or more hardware components of the network fabric.