Scenario: Run multiple latency benchmarks with different sizes and rates

[tbroden84]

WHY are these changes introduced?

Extended #34 to benchmark latency instead of throughput.

Part of making multiple runs: #30

WHAT is this pull request doing?

• Uploads multiple latency runs script
• Collect and present the results in a summary table

HOW was this pull request tested?

Tested for c8g.large for 20 seconds over message sizes [16, 64, 256, 512, 1024] and rate limits [10, 100, 1000, 10000, 50000, 100000, 200000, 500000]

https://gist.github.com/tbroden84/79ba926c519f7a52d64d690088297abd

[tbroden84]

Interesting observation on small rate limit where latency can be really high. From the output it's during the first second of the run. Perhaps lavinmqperf needs a warmup period or discard -N number of starting measurements.

[jage]

Interesting observation on small rate limit where latency can be really high. From the output it's during the first second of the run. Perhaps lavinmqperf needs a warmup period or discard -N number of starting measurements.

IIRC the latest version should have a "ready" gate, e.g. publisher won't start until consumer is ready, maybe we need to revisit that. Can you reproduce it locally?

[viktorerlingsson]

I wonder if it would make a difference using a lower prefetch-value when running the latency tests 🤔