Overview

A small utility to benchmark different approaches for building concurrent applications.

Pre-requisites

1. cargo - https://www.rust-lang.org/tools/install
2. python3.6+ with matplotlib

It generates the following files in the current directory:

• latency_histogram_{name}.png - X-axis latency in ms, Y-axis - counts for buckets
• latency_percentiles_{name}.png - X-axis - 0..100. Y-axis - latency percentile in ms
• latency_timeline_{name}.png - X-axis - a timeline in seconds, Y-axis - latency in ms, p50, p90 and p99
• request_rate_{name}.png - X-axis - a timeline in seconds, Y-axis - effective RPS (successes only)

where {name} is the --name (or -N) parameter value.

You may need to use --pythob/-p parameter to specify python3 binary, if it's not in /usr/local/bin/python3. E.g.

concurrency-demo-benchmarks --name async_30s \ --rate 1000 \ --num_req 100000 \ --latency "200*9,30000" \ --python /usr/bin/python3 \ async

Installation

cargo install concurrency-demo-benchmarks

Run batched/atomic/mutex increments benchmark

git clone https://github.com/xnuter/concurrency-demo-benchmarks.git cargo bench

Command line options

``` A tool to model sync vs async processing for a network service

USAGE: concurrency-demo-benchmarks [OPTIONS] --name --rate --num_req --latency [SUBCOMMAND]

FLAGS: -h, --help Prints help information -V, --version Prints version information

OPTIONS: -l, --latency Comma separated latency values. E.g. 200,200,200,500 -N, --name Name of the test-case -n, --numreq Number of requests. E.g. 1000 -p, --pythonpath Optional path to python3, e.g. /usr/bin/python3 -r, --rate Request rate per second. E.g. 100 or 1000

SUBCOMMANDS: async Model a service with Async I/O help Prints this message or the help of the given subcommand(s) sync Model a service with Blocking I/O

```

Output example: Latencies: p0.000 - 0.477 ms p50.000 - 0.968 ms p90.000 - 1.115 ms p95.000 - 1.169 ms p99.000 - 1.237 ms p99.900 - 1.295 ms p99.990 - 1.432 ms p100.000 - 1.469 ms Avg rate: 1000.000, StdDev: 0.000

Run sync demo

• 1000 rps
• 200ms latency, 10 endpoints

• 500 threads concurrency-demo-benchmarks --name sync_t500_200ms \ --rate 1000 \ --num_req 10000 \ --latency "200*10" \ sync --threads 500

• 1000 rps

• 600ms latency (stable)

• 500 threads concurrency-demo-benchmarks --name sync_t500_600ms \ --rate 1000 \ --num_req 10000 \ --latency "600*10" \ sync --threads 500

• 1000 rps

• 200ms latency but 30s for 10%
• 500 threads concurrency-demo-benchmarks --name sync_t500_30s \ --rate 1000 \ --num_req 100000 \ --latency "200*9,30000" \ sync --threads 500

Run async demo

• 1000 rps

• 200ms latency (stable) concurrency-demo-benchmarks --name async_200ms \ --rate 1000 \ --num_req 10000 \ --latency "200*10" \ async

• 1000 rps

• 600ms latency (stable) concurrency-demo-benchmarks --name async_600ms \ --rate 1000 \ --num_req 100000 \ --latency "600*10" \ async

• 1000 rps

• 200ms latency but 30s for 10% concurrency-demo-benchmarks --name async_30s \ --rate 1000 \ --num_req 100000 \ --latency "200*9,30000" \ async