💲UPI Payment System Design

Functional Requirements

• Users can link their bank accounts.

• Users can send and receive money using UPI IDs.

• Users can check their transaction history.

• Users can set a UPI PIN for secure transactions.

• Users can receive notifications for transactions.

• Users can receive payments using scanner.

Non Functional Requirements

• High availability and scalability to handle millions of transactions.

• Low latency to ensure real-time transaction processing.

• Strong security measures to protect sensitive financial data.

• Compliance with regulatory requirements.

https://drive.google.com/file/d/1qckD_J_v6kRAOTuZT2-rHeZkcHQ5a1wZ/view?usp=sharing

Envelope Calculations

In Jun 2024, 13.89 Billion transactions with amount of 20.27 Lakh crore with daily avg of 463 million transaction count.

Estimated Annual Transactions = 13.89 Billion * 12 = 166.68 Billion

We have daily average transaction count of 463 Million. But Lets consider peak situation with 1.3 times the average.

So Estimated Daily transactions per day would be around 463 * 1.3 = 601.9 Million transactions

Number of Transactions per hour: 601.9 / 24 = 25.08 Million

Number of Transactions per Second = 6967 ( Approx )

Capacity Estimations

In this sections we cover mostly about the Database storage requirement, network requirements, Server requirements and Bandwidth calculations.

Lets consider the transaction payload.

{
  "transactionId": "bc5fcc00-f47d-47b9-a424-5ca0bf93580a",
  "timestamp": "2024-07-17T14:30:00Z",
  "status": "SUCCESS",
  "amount": {
    "currency": "INR",
    "value": 1500.00
  },
  "sender": {
    "upiId": "sender@upi",
    "name": "Sunil Gudivada"
  },
  "receiver": {
    "upiId": "receiver@upi",
    "name": "Asish"
  },
  "note": "Payment for services",
  "metadata": {
    "merchantId": "6a107feb-e9cf-4db1-af27-c6a2e4bbff95",
    "transactionType": "P2P",  // P2P: Peer to Peer, P2M: Peer to Merchant
    "appName": "UPIApp"
  }
}

Size of this payload is about 373 B

Transaction storage calculation

Daily Storage: 463 Million * 373 B = 172.7 GB

Monthly Storage: 172.7 * 30 = 5.18 TB

Yearly storage: 172.7 * 365 = 63.04 TB

Network Bandwidth calculation

API Call Size:

Average Request/Response Size: 2 KB (including headers)

Daily API Traffic:

API Calls per Transaction: 3 (initiate, process, notify)

Total API Calls: 463 Million * 3 = 1.389 Billion

Data Transfer per Day: 1.389 Billion * 2 KB = 2.778 TB

Peak API Traffic:

Peak API Calls: 601.9 Million * 3 = 1.8057 Billion

Data Transfer per Day: 1.8057 Billion * 2 KB = 3.61 TB

Lets calculate the number of servers required.

Assume avg time taken by the request to complete around 50 Millisecond (MS)

Number of api calls at peak hours : 463 Million / 86400 = 5358.81 transactions per second

Throughput: 5358.81 * 50 = 267940.5 RPS

Consider one server is accepting 1000 Request per second

Number of servers: 267940.5 / 1000 = 267.9405 = 268 ( API Gateway )

Lets also calculate the number of servers required for the Backend services like User, VPA, Transaction, Bank Integrations.

Assume these microservices can process 200 request per second

Number of servers: 267940.5/ 200 = 1340 Servers

Components

Tech Stack

NPCI partnered with kyndryls cloud services for TARM Stack

Below details about the TARM Stack given by chatGPT. May be incorrect.

The TARM stack is an emerging modern web development stack that consists of:

1. Tigris: An open-source NoSQL database that supports both key-value and document models. Tigris is designed for high performance and scalability.

2. Appwrite: An open-source backend-as-a-service platform that provides a set of APIs to handle authentication, databases, storage, and server-side functions. It's a self-hosted Firebase alternative.

3. Remix: A modern JavaScript framework for building robust, dynamic, and SEO-friendly web applications. Remix emphasizes server-side rendering (SSR) and works seamlessly with modern JavaScript tools and libraries.

4. Meilisearch: An open-source, fast, and relevant search engine that can be integrated into web applications to provide powerful search functionalities.

Key Features and Benefits of TARM Stack:

1. Scalability and Performance: Each component in the TARM stack is optimized for high performance and can handle large volumes of data and traffic.

2. Developer-Friendly: The stack is designed to be easy to set up and use, with extensive documentation and community support.

3. Open-Source: All components are open-source, which provides transparency, flexibility, and cost-effectiveness.

4. Full-Stack Capabilities: The stack covers backend, frontend, database, and search functionalities, making it a comprehensive solution for web development.

References

https://economictimes.indiatimes.com/industry/banking/finance/total-upi-transactions-cross-100-billion-mark-in-2023/articleshow/106492755.cms?from=mdr https://www.npci.org.in/ https://www.itnews.asia/news/indias-npci-modernises-data-centres-using-kyndryls-cloud-services-592061 https://www.tigrisdata.com/ https://www.enterpriseitworld.com/kyndryl-to-target-a-market-size-of-530-billion/ https://www.linkedin.com/posts/kyndryl-india_kyndryl-and-npci-at-bharat-fintech-summit-activity-7029706712915836928-3mIx?utm_source=share&utm_medium=member_desktop