Endpoint Node Performance Probing for HPC Network Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance computing (HPC) systems face challenges in detecting and correcting communication performance issues, such as network congestion, due to their complex nature and high message rates, which traditional measurement techniques often exacerbate with increased overhead and system noise.
Innovation Solution
A system and method for monitoring communication performance in HPC networks that involves a source endpoint node and a destination endpoint node, using a performance probing engine to generate and compare timestamps for determining round-trip latency, allowing for accurate measurement without disturbing application behavior and enabling local congestion management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional performance measurement techniques (timing sections of code) are used, then network behavior attributes can be revealed, but measurement overhead increases and system noise is introduced
Solution Approach 1:
The patent extracts the measurement function from software-based timing approaches and implements it directly in network hardware (NICs). By placing timestamping capability at the network interface level, the measurement is separated from the application code path, eliminating the overhead of software timing while capturing actual network packet timing information.
Solution Approach 2:
The patent introduces hardware-based timestamping as an intermediary between the network packets and the measurement system. The NIC hardware automatically captures timestamps at packet boundaries without requiring software intervention, serving as a mediator that provides accurate timing information without the overhead of software-based measurement approaches.
2Measurement precision
If per-message code measurement approaches are used, then detailed performance data can be collected, but overhead increases dramatically with higher message rates
Solution Approach 1:
The patent enables the network interface card to perform measurement functions autonomously without requiring software processing for each message. The hardware automatically timestamps packets and can generate measurement data independently, allowing the system to handle high message rates without the software overhead that would otherwise limit productivity.
Solution Approach 2:
The patent replaces software-based measurement mechanisms with hardware-based timestamping. By substituting the software timing mechanism with hardware timestamping capability in the NIC, the system eliminates the performance penalty associated with software processing while maintaining detailed per-message measurement capability.
3Loss of information
If network probing technologies (ping, traceroute) are applied to HPC networks, then path information can be obtained, but overhead increases and high-fidelity measurements become difficult
Solution Approach 1:
The patent implements preliminary timestamping of packets at the source before they enter the network. By capturing the send timestamp in hardware at the moment the packet is transmitted, the system establishes a baseline measurement that eliminates the need for additional probing actions, thereby avoiding the overhead and potential interference that traditional probing methods would introduce.
Solution Approach 2:
The patent implements a feedback mechanism where the destination node returns the received timestamp to the source node. This feedback loop allows the source to calculate round-trip latency and verify packet delivery without requiring external probing tools, providing accurate path and performance information without the overhead of traditional network diagnostics.
4Adaptability or versatility
If software-based performance measurement is used, then flexibility in measurement approaches is maintained, but system noise increases and accurate detection becomes difficult
Solution Approach 1:
The patent extracts the measurement function from the software layer and places it in the hardware layer (NIC). This separation removes the measurement activity from the software execution path, eliminating the system noise that would otherwise be introduced by software timing operations while maintaining the ability to perform various measurement types through hardware configuration.
Data Source
AI summary
Technologies for monitoring communication performance of a high performance computing (HPC) network include a performance probing engine of a source endpoint node of the HPC network. The performance probing engine is configured to generate a probe request that includes a timestamp of the probe request and transmit the probe request to a destination endpoint node of the HPC network communicatively coupled to the source endpoint node via the HPC network. The performance probing engine is additionally configured to receive a probe response from the destination endpoint node via the HPC network and to generate another timestamp that corresponds to the probe request having been received. Further, the performance probing engine is configured to determine a round-trip latency as a function of the probe request and probe response timestamps. Other embodiments are described and claimed.