RDMA Queue Pair QoS Allocation for Data Center Congestion Control

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing data center architectures face challenges in efficiently managing network traffic congestion and resource utilization due to the integration of disaggregated resources, leading to suboptimal performance and increased operational costs.

Innovation Solution

Implementing a data center architecture with chassis-less sleds and an orchestrator server that dynamically allocates and manages RDMA queue pairs based on quality-of-service parameters, ensuring efficient resource utilization and congestion control through intelligent workload distribution.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If disaggregated resources are integrated into data center architecture, then resource utilization efficiency is improved, but network traffic congestion and management complexity increase

Engineering Contradiction:
Improveresource utilization efficiencyVSAvoidmanagement complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The system segments RDMA queue pair management into distinct functional components: an orchestrator server for high-level coordination and sled controllers for local execution. This segmentation allows complex disaggregated resource management to be broken down into manageable units, improving resource utilization while controlling management complexity through distributed decision-making.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The orchestrator server acts as an intermediary between resource requests and disaggregated hardware resources. It receives resource allocation requests, makes intelligent decisions about queue pair allocation based on QoS parameters, and coordinates with sled controllers to execute allocations. This intermediary layer simplifies management complexity by centralizing decision logic while enabling efficient resource utilization through coordinated control.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If RDMA queue pairs are dynamically allocated based on QoS parameters, then network traffic congestion is reduced, but system complexity increases

Engineering Contradiction:
Improvenetwork traffic congestionVSAvoidsystem complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The system implements dynamic allocation of RDMA queue pairs based on real-time QoS parameters and network conditions. The orchestrator server continuously monitors resource usage patterns, adjusts queue pair allocations dynamically, and reconfigures network traffic routing. This dynamic approach reduces network congestion by adapting to changing conditions while managing system complexity through automated control algorithms.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system incorporates feedback mechanisms where the orchestrator server monitors network traffic patterns, resource utilization metrics, and QoS parameter compliance. Based on this feedback, it automatically adjusts queue pair allocations and traffic routing decisions. This closed-loop feedback control reduces network congestion by responding to actual conditions while managing system complexity through rule-based automated adjustments.

Inventive Principle:
Principle #23Feedback

3Loss of energy

If intelligent workload distribution is implemented, then operational costs are reduced, but computational overhead increases

Engineering Contradiction:
Improveoperational costsVSAvoidcomputational overhead
Core Design Contradiction:
Loss of energyVSUse of energy by moving object

Solution Approach 1:

The orchestrator server performs preliminary analysis of workload characteristics, resource capabilities, and QoS requirements before making allocation decisions. By pre-processing workload information and establishing allocation strategies in advance, the system reduces operational costs through optimized resource matching while minimizing computational overhead during actual execution through pre-computed allocation plans.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3771988B1Technologies for RDMA queue pair QOS management
Publication Date: 2025.07.16 INTEL CORP
  • EP3771988B1 patent drawingFigure 1
  • EP3771988B1 patent drawingFigure 2
  • EP3771988B1 patent drawingFigure 3

AI summary

Technologies for remote direct memory access (RDMA) queue pair quality of service (QoS) management are disclosed. In the illustrative embodiment, several queue pairs associated with a virtual machine on a compute sled may be created in a network interface controller of the compute sled. A QoS parameter such as a class of service identifier or a weighting may be assigned to each queue pair such that each queue pair has a different available bandwidth. The compute sled may also predict future RDMA queue pair bandwidth usage and adjust RDMA queue pair bandwidth allocation based on the prediction.