BigQuery backend, query optimization / materialized views.Google Auto sharder and load balancer for stateful services that handles billions of QPS.One-suite storage system for Google applications, providing various features including storage, backup, indexing, batch data processing, GDPR support, etc..

Storage and Serving Infrastructure for planet scale Google Applications, e.g., Google Docs.

Topic Sentiment Analysis. • Research techniques to analyze public opinion expressed upon certain topics on the Web like product reviews, weblogs, etc. This is related to various web applications and could be a key feature of future search engines. Data mining, Statistical Topic Model and Natural Language Processing techniques are investigated in it.

Content based chunk storage scheme• Utilize the Rabin Footprint hashing technique to intelligently divide long document into chunks, which avoids duplicate update of storage in case of document editing.• Design a benchmark to test the proposed method on various kinds of Google documents. The results show that write bandwidth can be reduced by about 40%.• Present the design ideas in the intern poster session and obtain “Excellent” in final intern evaluation.

Distributed high throughput database design • Participate (3 persons) designing a backend database storing hundreds of millions of changeful user profile data and a high throughput frontend system responding to thousands of concurrent requests within milliseconds latency rate. • Prototyped the backend in memory distributed database system, the first release was finished in about 4 months’ time, which is scalable to data volume, workload increase and new applications.Huge volume OLAP system • Modified logging mechanism of bulk insert, reducing load time by about 1/3.• Improved different aggregated reports on same dimension to use only one data table, saving both pipeline and database loading time.• Designed and implemented the time partition based database for a group of tables each having millions of rows per month.• Implemented and optimized stored procedures by rewriting query and building indexes.