Adaptive Queue-Pair Switching for Latency and Memory Tradeoffs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance computing systems face challenges in optimizing latency and memory consumption due to the complexities in managing queue pairs (QPs) for efficient data transmission between nodes, particularly with RC and DC QPs, which affect applications like artificial intelligence and mathematical calculations.
Innovation Solution
A system and method that adaptively manages QPs by monitoring access rates and counts to peers, switching between RC and DC QPs based on communication patterns, establishing RC QPs for frequently accessed peers and terminating them for less frequently accessed peers to optimize memory usage and reduce latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RC QPs are used for all peers, then communication reliability is improved, but memory consumption increases
Solution Approach 1:
The patent applies local quality by differentiating QP management based on peer characteristics. Frequently accessed peers use RC QPs for reliable communication, while infrequently accessed peers use DC QPs to save memory. The system dynamically adjusts QP types based on access patterns, applying different connection qualities to different peers rather than using a uniform approach.
Solution Approach 2:
The patent changes the QP type parameter (RC vs DC) based on communication frequency. The system monitors access rates and dynamically transitions QPs between RC and DC modes, adjusting the connection parameter to optimize the balance between reliability and memory consumption based on actual usage patterns.
2Quantity of substance
If DC QPs are used for all peers, then memory consumption is reduced, but communication latency increases
Solution Approach 1:
The patent applies local quality by assigning different QP types to different peers based on their access frequency. Hot peers (frequently accessed) are assigned RC QPs with dedicated resources for low-latency communication, while cold peers (infrequently accessed) use DC QPs to conserve memory. This localized optimization prevents the latency penalty from affecting all communications.
Solution Approach 2:
The patent implements dynamic QP management where the system continuously monitors communication patterns and adjusts QP types accordingly. Peers can transition between RC and DC QP modes based on changing access patterns, allowing the system to adapt to dynamic workloads and minimize latency for actively communicating peers while maintaining memory efficiency.
3Speed
If RC QPs are established for frequently accessed peers, then communication speed is improved, but device complexity increases
Solution Approach 1:
The patent implements self-service by having the system automatically monitor communication patterns and make decisions about QP type selection. The QP management system observes access frequencies and autonomously transitions QPs between RC and DC modes without requiring manual intervention or complex external control, reducing the perceived complexity for users while maintaining optimization capabilities.
Data Source
AI summary
A system is described which monitors accesses to one or more peers using queue pairs (QPs). Each access is associated with a respective peer of the one or more peers. The system identifies, based on one or more of a rate and a count of the monitored accesses to the one or more QPs, a first peer of the one or more peers. In response to identifying the first peer, a reliable connection QP is established for the first peer of the one or more peers.