Overview

At PhonePe, our in-house API gateway handles over a million requests every second. When you operate at this scale, you encounter performance challenges that are impossible to spot during testing. Using eBPF as our debugging tool, we not only solved these issues but also saved millions in yearly infrastructure costs.

We’ll share three real stories about finding and fixing performance bottlenecks:
Case Study 1: The Growing Infrastructure Problem
Why were we adding 5 new 18-core instances every week? We’ll show how we used eBPF to find the root cause and stop this expensive growth.

Case Study 2: The Aerospike Mystery
Our Aerospike database was handling only 6,000 writes per second per node, when it should have been doing 10 times more. Learn how we uncovered and fixed what was holding it back.

Case Study 3: Working Smarter, Not Harder
By switching from synchronous to asynchronous processing, we reduced our instances count from 200 to just 20. We’ll explain how we made this massive improvement.

Takeaways

• How to use eBPF to find performance problems in production
• Real examples of scaling challenges and their solutions
• Practical ways to improve system performance
• How to turn performance fixes into cost savings

Audience

• Site Reliability and DevOps Engineers
• Engineering leaders
• Cloud architects and engineers