API GatewayWhy Perfomance and Load Testing Matters(2)

2) Preparations for Performance and Load Testing 

As mentioned earlier, the API Gateway acts as the entry point for API requests and responses. The subject of performance and load testing is the API Gateway with proxy functionality.

The objective of the test is to determine how many requests can be processed per second and how much system resources (CPU, Memory) are utilized, allowing for comparison.

1. Preparing the API Gateway

   • Dedicated server preparation
     • CPU: 4CPU
     • Memory: 8G
   • Test Functions
     • Proxy: Receives API requests from the client, forwards them to the API server, and delivers the response back to the client.
   • Additional setup of Tomcat + Proxy Servlet for comparison testing

2. Preparing the API

   • Dedicated server preparation
     • CPU: 4CPU
     • Memory: 8G
   • API
     • Generates and returns a 4 KB JSON response upon receiving a request.

3. Preparing the Client

   • Dedicated server preparation
     • CPU: 4CPU
     • Memory: 8G
   • Using nGrinder Open Source Testing Tool
     • Prepare nGrinder Controller, Agent, Monitor 

• Notably, nGrinder provides a monitor for measuring system resources of the target server, enabling post-test UI-based analysis of TPS, CPU, and memory usage. Additionally, the use of separate Agents allows for increasing concurrent users.

3) Performance and Load Test Execution and Results

• Test Environment
  • Concurrent users: 100 users (5 processes * 20 threads each)

    Test duration: 5 minutes

    • Test 1
      [Client] — https → [OSORI API Gateway] — https → [API Server: 4k JSON]

    • Test 2
      [Client] — https → [Tomcat + Proxy Servlet] — https → [API Server: 4k JSON]

• Test Results
  • OSORI API Gateway
    • Average TPS: 11,177
    • Requests/responses processed in 5 minutes: 3,310,757
    • Average CPU usage: 76%
    • Average Memory usage: 0.3G (excluding 1.5G used by the Monitor process)
  • Tomcat + Proxy Servlet
    • Average TPS: 6,232
    • Requests/responses processed in 5 minutes: 1,845,405
    • Average CPU usage: 91%
    • Average Memory usage: 2.7G (excluding 1.5G used by the Monitor process)

4) Summary

OSORI demonstrated nearly twice the TPS performance compared to Tomcat. The number of requests/responses processed in 5 minutes shows that OSORI handled approximately 1.5 million more than Tomcat. (If each API request is assumed to generate 10 KRW, this would result in an additional revenue of 15 million KRW.)

Another noteworthy aspect is that while processing more requests, OSORI’s CPU usage was 15% lower, and its memory consumption was only 11% of Tomcat’s.