Performance Analyzer

Performance Analyzer is an HTTP benchmarking tool capable of generating significant load from a single node or from a distributed cluster. It combines the capabilities of elixir to analyze the behavior of an application in different concurrency scenarios.

• Performance Analyzer
  • Install
  • Basic Usage
  • Run
  • Results
    • Examples

Install

mix do local.hex --force, local.rebar --force
mix do deps.clean --unused, deps.get, deps.compile
mix compile

Basic Usage

Open and edit config/performance.exs file to configure.

import Config

config :distributed_performance_analyzer,
  url: "http://httpbin.org/get",
  request: %{
    method: "GET",
    headers: [{"Content-Type", "application/json"}],
    # body: ~s|'{"data": "value"}'| --> If you don't use dataset values
    body: fn item ->
      # This is for dataset replacement
      ~s|'{"data": #{Enum.random(1..10)}, "key": "#{item.columnName}"}'|
    end
  },
  execution: %{
    steps: 5,
    increment: 1,
    duration: 2000,
    constant_load: false,
    dataset: "/Users/sample.csv",
    # dataset: :none, --> If you don't use dataset
    separator: ","
  },
  distributed: :none,
  jmeter_report: true

config :logger,
  level: :info

Property	Description
url	The url of the application you want to test. Make sure you have a network connection between two machines
request	Here you need to configure the HTTP verb, headers and the body of the request.
steps	The number of executions for the test. Each step adds the concurrency configured in the increment
increment	Increment in concurrency after each step
duration	Duration in milliseconds of each step
constant_load	Allows you to configure if the load will be constant or if the increment will be used to vary the concurrency in each step
dataset	The path to the csv dataset file
separator	Dataset separator (, ; :)
distributed	Indicates if it should be run from a single node or in a distributed way
jmeter report	Generates jmeter csv style report?

In the example above will be executed a test of 5 steps with an increment of 50:

1. Step 1: 50 of concurrency
2. Step 2: 100 of concurrency
3. Step 3: 150 of concurrency
4. ...

Each step will last 2 seconds.

Run

Docker:

https://hub.docker.com/r/bancolombia/distributed-performance-analyzer

docker run --rm -v <project_path>/config:/home/dpa/config -v <project_path>/dataset:/home/dpa/datasets bancolombia/distributed-performance-analyzer:latest

In the shell:

iex -S mix
or
iex --sname node1@localhost -S mix

To run IEX Execution:

alias DistributedPerformanceAnalyzer.Domain.UseCase.ExecutionUseCase
ExecutionUseCase.launch_execution()

Elixir Release:

mix release
_build/prod/rel/distributed_performance_analyzer/bin/distributed_performance_analyzer start

Results

After each step is executed you will get a table of results like the following:

####CSV#######
concurrency, throughput, mean latency (ms), p90 latency (ms), p95 latency (ms), p99 latency (ms), max latency (ms), mean http latency (ms), http_errors, protocol_error, error_conn, nil_conn
1, 9, 108, 110, 111, 117, 119, 108, 0, 0, 0, 0
2, 18, 107, 110, 110, 111, 111, 107, 0, 0, 0, 0
3, 28, 107, 109, 110, 111, 111, 107, 0, 0, 0, 0
4, 36, 108, 110, 111, 111, 111, 108, 0, 0, 0, 0
5, 45, 108, 110, 110, 111, 111, 108, 0, 0, 0, 0

JMeter format:

timeStamp,elapsed,label,responseCode,responseMessage,threadName,dataType,success,failureMessage,bytes,sentBytes,grpThreads,allThreads,URL,Latency,IdleTime,Connect
1684342000805,87,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,85,0,55
1684342000892,19,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,19,0,4
1684342000911,17,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,17,0,0
1684342000928,17,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,17,0,0
1684342000945,18,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,18,0,0
1684342000963,16,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,16,0,0
1684342000979,18,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,18,0,0
1684342000997,16,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,16,0,0
1684342001013,18,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,18,0,0
1684342001032,19,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,19,0,0
1684342001051,17,sample,200,OK,#PID<0.447.0>,TEXT,true,,24,2,0,0,GET -> http://localhost:8080/wait/10,17,0,1

Then, you can compare the attributes that are interesting for you. For example concurrency vs Throughput or Concurrency vs Mean Latency.

Examples