core-to-core-latency-plus

CPU Core-to-Core Latency Tools +

Description

This program is used to calculate the communication latency between CPU cores.

Specifically, it fixes two threads on two different CPU cores and measures the latency by having the threads perform a series of compare-and-swap operations through the CPU cache coherence protocol.

This fork aims to improve the usability of the program and collect more CPU benchmark data for display.

*** Note: Due to MacOS removing related interfaces, the core_affinity library that this program depends on cannot bind threads to specific cores on MacOS systems, nor can it obtain the current core ID at runtime. Therefore, the test results of this program on MacOS are strictly invalid (they may all be scheduled to the same one or a few cores). It is recommended that Mac users install a Linux system for testing ***

How to run

How to run:

Download pre-compiled version and run

Please download from https://github.com/KCORES/core-to-core-latency-plus/releases.

for Linux, please run:

core-to-core-latency-plus -b 1 --upload

for Windows, please open CMD, cd to file folder and run:

core-to-core-latency-plus.exe -b 1 --upload

Note that the --upload option will upload your benchmark results to https://core-to-core-latency.kcores.com.

You can view your results and generate a heatmap using the link generated after the upload is complete.

Considering the substantial size of the historical benchmark dataset (creating a Jupyter Notebook from the "results" folder results in a file of approximately 170MB), we highly recommend uploading the data to the platform for immediate heatmap visualization. For offline access to historical data, please consult History Results.

install with cargo and run

$ cargo install core-to-core-latency-plus
$ core-to-core-latency-plus -b 1 --upload

Leaderboard

For more results, please visit: https://core-to-core-latency.kcores.com

History Results

History Results

License

MIT