PhonePe Processes 100M Daily Payments with 99.999% Uptime Using Aerospike

PhonePe is India’s leading fintech super app, processing over 100 million transactions daily and serving 380 million registered users and 30 million merchants. Launched in 2016, the company has played a pivotal role in India's shift to digital payments, supporting Unified Payments Interface (UPI) transactions, merchant payments, financial services, and personalized engagement features across its platform. To maintain real-time performance and uninterrupted availability at this scale, PhonePe required an infrastructure capable of handling millions of transactions per second with ultra-low latency.

Explosive growth, database scaling, and fraud prevention

PhonePe’s rapid expansion to 380M users and 30M merchants required a fault-tolerant, low-latency infrastructure to handle 100M+ daily payment transactions while ensuring real-time fraud prevention and high availability.

• Unprecedented growth in digital payments

  Transaction volume surged to 3B per month, straining existing systems and demanding real-time processing at massive scale.

• Scaling limitations of legacy databases

  Relational and big data stores like HBase couldn’t support real-time updates or handle peak loads, causing payment processing bottlenecks and increasing operational complexity. HBase’s block cache was difficult to manage, limiting application-level control over caching and slowing performance.

• Frequent failures and costly downtime

  Disk and memory failures every 15 days led to weeks-long rolling restarts, while failover and disaster recovery remained complex.

• Diverse use cases needed real-time performance

  Fraud detection queries took too long, limiting real-time blocking. Personalization and engagement systems lagged, reducing user experience quality.

Aerospike powers a high-performance, fault-tolerant data layer

PhonePe deployed Aerospike to power its digital payments ecosystem, ensuring real-time transaction processing, fraud detection, and personalized user engagement while reducing infrastructure costs. It now operates hundreds of Aerospike clusters and thousands of server nodes across three on-premises data centers, managing over 160TB per site and processing millions of queries per second (QPS) with ultra-low latency.

• Unmatched speed and reliability

  Aerospike delivers sub-millisecond response times for payments and fraud detection, with active-active replication ensuring seamless failover and zero replication lag

• Efficient storage and infrastructure

  Hybrid Memory Architecture (HMA) and workload-specific tuning enabled an 8x server footprint reduction—lowering costs while sustaining millions of QPS

• Replaced HBase for better performance

  Aerospike replaced key HBase workloads, reducing latency, hardware needs, and complexity, while serving as a high-performance cache for feature store use cases

• Flexible consistency models

  PhonePe uses Aerospike AP (availability-partition tolerant) mode for high availability and SC (strong consistency) mode for transactional integrity, optimizing for workload-specific performance and reliability

• Seamless cross-data center replication

  Aerospike XDR supports both unidirectional and bidirectional replication, maintaining data consistency across geographically distributed sites

• Built-in event streaming and observability

  Kafka connectors enable real-time event streaming, while Prometheus and Grafana support monitoring, anomaly detection, and alerting

99.999% uptime, scalability, and efficiency

By optimizing infrastructure with Aerospike’s high-performance architecture, PhonePe achieved faster query response times and reduced operational complexity while scaling to billions of transactions.

• 99.999% uptime

  Ensures uninterrupted digital payments across India with active-active availability.

• 5x higher throughput

  Powers fraud detection and feature store lookups with sub-millisecond latency, even at peak scale.

• 8x reduction in server footprint

  Reduces infrastructure costs while sustaining millions of QPS across diverse real-time workloads.