Interview with Zhao Haiping on Asynchronous Processing and Performance Optimization in Distributed Systems

Zhao Haiping , who joined Facebook in 2007 when it had fewer than 50 engineers, worked on software performance and architecture, creating the HipHop project that rewrote PHP to achieve a 5‑6× speedup and saved billions of dollars. After moving to Alibaba in March 2015 as a researcher in the Technical Assurance Department, he focused on solving performance issues in Alibaba's large‑scale systems.

InfoQ: Welcome back to China. Can you introduce why you joined Alibaba and what attracted you? Zhao Haiping: I was drawn by the massive scale of Alibaba’s traffic, especially events like Singles' Day, which present unique performance challenges not seen in US companies. Managing resources for such peaks while maintaining high utilization is a difficult technical problem that aligns with my interest in large‑scale system performance and stability.

InfoQ: Your new role involves addressing infrastructure performance issues? Zhao Haiping: Yes, we work on performance, stability, capacity, architecture, and operations. Improving performance increases capacity, and better monitoring enhances stability and operational convenience. When we identify architectural problems, we adjust the design accordingly.

InfoQ: How do you achieve accurate monitoring and profiling in such large distributed systems? Zhao Haiping: A powerful profiling system is essential to know where CPU, memory, disk I/O, or network bandwidth is consumed across machines. My first task was to improve Alibaba’s monitoring and profiling to provide real‑time visibility of software performance, allowing developers to see resource usage of their code online and address issues themselves.

InfoQ: Can you talk about the HipHop for PHP project and its origins? Zhao Haiping: HipHop started from a need to convert a frequently changing PHP function into C++. The goal was to create a tool that could translate PHP code to C++ without slowing down development, addressing PHP’s high CPU consumption caused by dynamic function calls. By moving many calls to static compilation, we achieved significant speedups, eventually reaching 5‑6× performance gains.

InfoQ: How did HipHop evolve into HHVM? Zhao Haiping: HHVM replaced static C++ compilation with a JIT approach that translates PHP to intermediate bytecode, then to machine code at runtime, reducing binary size and deployment issues. I left the HipHop team before HHVM’s development, focusing instead on asynchronous processing in distributed systems.

InfoQ: What are your thoughts on companies creating their own programming languages? Zhao Haiping: Languages should not be treated as sacred; they are part of the software system and should be chosen or created only when existing languages cannot meet specific needs. Creating a language makes sense only if it solves concrete problems that cannot be addressed by libraries or existing languages.

InfoQ: Any advice for Chinese developers? Zhao Haiping: Improve communication skills, be willing to express ideas assertively, and master English to broaden your perspective. Technical ability alone is insufficient; effective communication and language proficiency are crucial for career growth.