🚀 Performance Testing Essentials for Web Apps

🎯 Purpose

The purpose is to rapidly identify the major bottlenecks in an app’s response time and resource usage under specific load conditions.

🌱 Origin

The Performance Testing Tool ab (Apache Bench) was created by Adam Twiss in 1996, and the modern, rate-controlled alternative hey, developed by Jaana Burcu Dogan (rakyll) in 2016.

🧠 Essentials

Focus on three areas: Goals, Measurement, and Analysis.

1. Core Performance NFR Types

• Latency (Response Time): Ensuring the server responds quickly enough for the end-user. The target should be measured at the P90 or P95 percentile (e.g., 90% of requests must be $\le 300\text{ ms}$), while the system is subjected to a specific concurrency level.
• Capacity (Throughput & Resource): Ensuring the system can handle the expected number of requests without running out of CPU or memory. Key metrics are Requests Per Second (RPS) and sustained CPU/Memory usage under target load.

2. Common Load Profiles

• Normal Load: A controlled, low-to-moderate rate of requests matching typical daily usage (e.g., 2 RPS). Best for validating Latency NFRs.
• Stress/Peak Load: Pushing the system to its breaking point to find its maximum sustainable RPS. Best for validating Capacity NFRs and finding bottlenecks.

3. Measurement Tools

Two key CLI tools: one for speed (stress) and one for control (load).

4. Analysis Essentials

Focus on two primary sources of data to diagnose issues:

• Client-Side (Load Tool Output): Check the P90/P99 latency values in the hey output to see if the NFRs are met.
• Server-Side (System Monitor): While the test runs, use htop or watch + ps on the server to ensure CPU and Memory remain below the defined limits.

🛠️ hey Hands-On

Usage of hey for controlled NFR verification because it allows rate limiting.

• Step 1: Linux Installation hey github repo

# 1. Download the executable binary (use current link from github repo)
wget https://hey-release.s3.us-east-2.amazonaws.com/hey_linux_amd64 

# 2. Make it executable
chmod +x hey

# 3. Move it to a system path
sudo mv hey /usr/local/bin/

# 4. Verify
hey -h

• Step 2: Run Backend and save its Process-Id

# Open new terminal (T1) and change to BE dir
cd [workspace/backend]
# Run BE
node server.js

# Open new terminal (T2) and Save BE PID of Node.js server to variable
PID=$(pgrep -f "node server.js")

• Step 3: Execute Load-Test using Performance NFR Params
  • Response Time: P90 $\le 300\text{ ms}$
  • Resource Usage: CPU $\le 15%$ and Mem $\le 300\text{ MB}$ under a load of 2 Concurrent Users for 10 minutes.
  • Simulation of a load of $\approx 1\text{ RPS}$ per user, totaling $600$ requests over $10$ minutes.

# -n 600: Total 600 requests
# -c 2: 2 concurrent workers (simulated users)
# -q 1: Rate limit of 1 query per second *per worker* (Total 2 RPS)
# -m POST: HTTP method
hey -n 600 -c 2 -q 1 -m POST http://localhost:3000/api/spin

• Step 3: Resource usage Verification during Test Run

# Go to T2 and Monitor (for 10min) CPU and Memory (RSS in KB: Resident Set Size, used physical mem)
watch -n 5 "ps -p $PID -o %cpu,%mem,rss"
# %CPU < 15%; RSS < 307200KB (300MB).

• Step 4: Response Time Verification after Test Run

• check latency distribution in hey output. 90% in [X] ms is le 300 ms

🛠️ ab Hands-On

• Step 1: Linux Installation

# 1. Install
sudo apt install apache2-utils

# 4. Verify
ab -h

• Same Steps as Tool hey

• Only missing cmd option is -q