Comparative Study of Backend Architectures

This document compares three architectures based on performance metrics from a load test. The three architectures evaluated are:

1. Architecture 1: Monolithic: In a monolithic architecture, the entire application is developed and deployed as a single unit. All components of the system (like the user interface, business logic, database management, and any other backend services) are tightly coupled and run within a single process. Here, all CRUD operations are handled by a singleton server, connected to a singular DB.

2. Architecture 2: Monolithic with Segregated Databases: This architecture is similar to the monolithic architecture, but with a segregated database layer. It still operates as a single, tightly-coupled application, but the database is separated into different sections or services to manage data for different functionalities more effectively.

3. Architecture 3: Microservices with API Gateway: Microservices architecture is a decentralized approach where the application is divided into a set of independent, self-contained services. Each service is designed to handle a specific business functionality and operates autonomously. An API Gateway is used to route requests from clients to the appropriate microservices. Here, the architecture consists of 2 servers, connected to their respective seperate Databases to handle 2 different categories of API requests. The requests are proxied dynamically to their respective servers by an API Gateway. (NB: This might be a scrapy version of what a true Microservices architecture might be in production; since I reproduced this just from the definition)

Overview

Each architecture underwent the same load test with Virtual Users (VUs), and the performance was measured based on various factors such as response time, error rates, HTTP request success rates, and session length.

Key Considerations:

• Databases are empty before running batch tests.
• Only the required server of the architecture being tested for is running in the background.

System Specifications:
The tests were performed on:

• Processor: AMD Ryzen 7 5700U with Radeon Graphics
• Base Speed: 1.80 GHz
• Cores: 8
• Logical Processors: 16
• RAM: 16.0 GB
• GPU: AMD Radeon(TM) Graphics (Dedicated 2 GB, Shared 6.9 GB)

Key Metrics Analyzed

• Errors: Connection refused (ECONNREFUSED), Timeout errors (ETIMEDOUT).
• HTTP Response Codes: Success (200), Creation success (201).
• Request Rate: The number of requests made per second.
• Response Time: Minimum, Maximum, Mean, Median, P95, P99.
• User Sessions: Completed sessions, session length.

Tabular Summary

Metric	Architecture 1: Monolithic	Architecture 2: Monolithic with Segregated DB	Architecture 3: Microservices with API Gateway
Total Time	2 minutes, 10 seconds	2 minutes, 6 seconds	2 minutes, 9 seconds
Errors
- ECONNREFUSED	1391	1096	N/A
- ETIMEDOUT	658	N/A	12
HTTP Requests (Total)	6300	6300	6300
Request Rate (req/sec)	54/sec	55/sec	55/sec
Successful HTTP Responses
- HTTP 200	1783	2553	3122
- HTTP 201	2468	2651	3166
Response Time (ms)
- Mean	1550.3	1581.1	1106.8
- Max	9998	9711	10003
- P95	8692.8	6187.2	6312.2
- P99	9607.1	7260.8	8692.8
User Sessions
- Completed	4251	5204	6288
- Failed	2049	1096	12
Session Length (ms)
- Mean	1554.5	1585.1	1111.5
- Max	10002.8	9716.1	10009.7
Success Rate	67.5%	82.7%	99.8%

Observations & Inference

• Architecture 1 (Monolithic) shows the highest error rates, especially with ECONNREFUSED (1391) and a relatively high number of failed user sessions (2049). However, it maintains a reasonable request rate of 54 requests per second with a mean response time of 1550.3 ms.

• Architecture 2 (Monolithic with Segregated Databases) reduces the number of ECONNREFUSED errors (1096) and improves the number of successful HTTP requests, but the mean response time is slightly higher at 1581.1 ms. The failure rate is also lower than in Architecture 1.

• Architecture 3 (Microservices with API Gateway) performs the best overall, with minimal timeout errors (12) and the lowest failure rate (12). It also achieves the fastest mean response time (1106.8 ms), although the response times at higher percentiles (P95 and P99) are slightly higher compared to the monolithic architectures.

In conclusion, Architecture 3 (Microservices with API Gateway) stands out as the most efficient architecture in terms of response time, error handling, and user session success. It strikes the best balance between performance and scalability, making it a better choice for high-demand systems. However, the choice of architecture should depend on the specific needs and complexity of the application, as well as the trade-offs related to implementation and maintenance.