RDMA Traffic Classification via ECN Markings and Queue Depth
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
RDMA workloads in network clusters fail to achieve desired throughput due to suboptimal congestion management protocols, particularly in RoCE environments, which are not tailored to the specific requirements of different traffic classes.
Innovation Solution
A framework that classifies RDMA traffic into distinct classes and applies customized processing settings, including ECN markings and queue depths, at both network device and NIC levels to optimize traffic handling, ensuring appropriate handling for bandwidth-sensitive and latency-sensitive traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If RoCE congestion management protocol is used for RDMA traffic, then network connectivity is maintained, but throughput performance is insufficient
Solution Approach 1:
The patent applies parameter changes by modifying ECN (Explicit Congestion Notification) markings and queue depth settings in the RoCE congestion management protocol. These parameter adjustments optimize the protocol's behavior for different traffic classes, enabling bandwidth-sensitive RDMA workloads to achieve higher throughput while maintaining reliable congestion control. The customized parameters allow the network device to better manage resource allocation and prevent congestion-induced throughput degradation.
2Productivity
If customized processing settings are applied for different traffic classes, then throughput is improved, but device complexity increases
Solution Approach 1:
The patent implements segmentation by dividing RDMA traffic into distinct traffic classes (e.g., bandwidth-sensitive and latency-sensitive) and applying customized processing settings to each class separately. This segmentation allows the network device to handle different traffic types with appropriate optimizations without requiring complete redesign of the processing architecture. The tag-based classification system enables selective application of ECN markings and queue depth settings only where needed.
Solution Approach 2:
The patent applies local quality by configuring different processing parameters (ECN markings, queue depths) specifically for bandwidth-sensitive traffic classes while using default settings for other traffic. This localized optimization ensures that complexity is introduced only in the specific areas where throughput improvement is needed, rather than uniformly across all traffic handling functions.
3Productivity
If resource allocation is optimized for bandwidth sensitive traffic, then network throughput increases, but other traffic may be affected
Solution Approach 1:
The patent implements dynamics by making resource allocation adaptive rather than static. The network device dynamically adjusts ECN markings and queue depth settings based on the traffic class identification. Bandwidth-sensitive traffic receives optimized resource allocation (higher queue depths, selective ECN marking) to maximize throughput, while other traffic classes maintain their original processing parameters. This dynamic, conditional approach ensures that throughput optimization for one class does not permanently degrade service for other classes.
Data Source
AI summary
Discussed herein is a framework that provisions for customized processing for different classes of traffic. A network device in a communication path between a source host machine and a destination host machine extracts a tag from a packet received by the network device. The packet originates at a source executing on the source host machine and whose destination is the destination host machine. The tag set by the source and indicative of a first traffic class to be associated with the packet, the first traffic class being selected by the source from a plurality of traffic classes. The network device determines, based on the tag, that the first traffic class corresponds to a bandwidth sensitive traffic and processes the packet using one or more settings configured at the network device for processing packets associated with the first traffic class.