Scaling Billion‑Level Ads: Architecture Lessons from Sogou’s Senior Engineer

Background

The article originates from a CSDN interview conducted at the 2015 CSDN Annual Technology Summit in Beijing, featuring Liu Jian, a senior architect at Sogou responsible for the commercial platform’s core infrastructure.

Interview Highlights

Career and Focus Liu holds a Ph.D. in Computer Science from Beihang University, previously worked at IBM China Research, and now leads Sogou’s commercial platform architecture, covering data storage, distributed computing, service‑orientation, security, and advertising billing.

Path to Architecture He emphasizes that becoming an architect requires learning ability, curiosity, and hands‑on practice. He studied service‑oriented computing, read extensively on software architecture, and experimented with SEDA, which shaped his macro‑level thinking.

Architecture of Sogou’s Commercial Platform

The platform supports over a billion online ads across search, network, and brand promotion. Its architecture must address distributed high‑concurrency, massive data volume, strong consistency, and rapid fault recovery.

Base Platform Includes SOA framework, database sharding, data storage, caching, offline/online processing, and distributed coordination. Core open‑source components are Spring, Redis, MySQL, MongoDB, Hadoop, Kafka, Storm, and Zookeeper.

Base Services Provide generic, business‑agnostic capabilities such as unique ID generation, data sync, authentication, authorization, configuration, distributed scheduling, messaging, and media services.

Core Services Implement business‑specific functions like billing and audit.

The design mirrors mainstream internet infrastructure, emphasizing high availability, performance, scalability, and low cost. Real‑time ad delivery is critical; any outage can cause revenue loss, so fault detection, traceability, and rapid recovery are prioritized.

Advice for Engineers Aspiring to Be Architects

Start by optimizing the architecture of current projects to improve personal technical depth.

Continuously learn and exchange new technologies, ideas, and architectural patterns.

Stay aware of mainstream technologies, evaluate their fit for your projects, and apply them when appropriate.

Time‑Management Tips for Technical Professionals

Liu recommends batching interruptive tasks (e.g., email, meetings) into dedicated time slots, preserving uninterrupted periods for design and development, using a TODO list to finish daily items, and selecting meetings based on relevance to one’s work.

Emerging Languages and Technologies

His team primarily uses Java, C++, Python, and scripting languages. They are particularly interested in Scala and Apache Spark for efficient big‑data and stream processing, leveraging existing Java assets while gaining higher productivity.

Learning Strategies for Rapidly Changing Tech

He advises focusing on foundational software theory—databases, operating systems, data structures, algorithms—while adopting a problem‑driven learning plan, combining study with practice, and engaging with domain experts.

Evolution Stages of the Platform

Horizontal scaling of computation.

Horizontal scaling of storage.

Service‑oriented transformation.

Adoption of stream‑processing.

Each stage brought specific challenges and lessons, which Liu shares as practical experience for architects dealing with fast‑moving business demands.