BENCH.md

Benchmarking Guide

Requirements for running the benchmark:

• Cargo and rustc (installation instructions)
• Python

You can run the benchmarks from the eval/ folder.

zk@zk eval~: python3 generate_report.py

This will create a report.txt file that creates measurements for the different configurations.

Customization

You can perform a benchmark across different configurations. In the python script you find the following section:

for cache_line in [0, 128]:
    for queue_size in [15]:
        for prod in [1, 3, 5, 9]: # change this per machine
            for dummy in [0, 500]:
                for chunk_size in [1]:

As you can see, we're iterating over all possible parameter configurations. The parameters are explained in the following.

variable	meaning
cache_line	The alignment of the queue indices (should be either 0 or your cache line size)
queue_size	queue capacity = 2 ^ queue_size, should be <= 16 currently because of u16 indices
prod	Number of concurrent producers
dummy	Number of dummy instructions inserted between every push to reduce contention
chunk_size	Number of bytes submitted in every push operation by consumers

You can set the parameters to suit your particular system.