The China Computer Federation (hereinafter “CCF”)-Tencent Rhino Bird Fund has announced the winners for outstanding projects of 2021 recently. Associate Professor Zhang Feng of the School of Information at Renmin University of China was awarded the highest honor of the year, the Excellent Project Award. He was also the recipient of the 2021 TPDS Best Paper Award. Zhang has been working tirelessly and progressively in the cross-field of database and system architecture since he came to work in the university.
1. Conducting authentic, prospective and challenging database system research through industry-university collaboration
Recently, there is a growing trend of exploring big data and artificial intelligence, with applications such as AI-powered painting and ChatGPT receiving significant public attention. As big data technology continues to evolve, universities must seize opportunities and pursue innovative research aimed at improving human life quality. To this end, Zhang Feng and the Tencent database team have collaborated on the development of a novel database storage engine that can process, e.g., retrieve and update compressed data at the storage level, thus substantially improving the efficiency of database systems.
As early as in 2017, Zhang and his team had recognized that storage space and processing time would become bottlenecks in the development of big data storage and analytics technologies. They then proposed the idea of "direct computing on compressed data", and later presented related thoughts, challenges and solutions at the 2018 VLDB International Conference. Since then, he had made breakthroughs in multiple dimensions such as direct processing algorithms, performance and system implementation. In 2021, he saw participating in the CCF-Tencent Rhino Bird Fund program as a great opportunity to transform their achievements and worked with the Tencent database team to develop a new type of database storage engine for direct processing of compressed data.
Zhang considers industry-university collaboration a highly effective model for collaborative innovation. On the one hand, the database team of Renmin University of China has profound technological accumulations, which serve as the technical foundation for collaboration and a strength of the university. On the other hand, emerging and evolving industrial needs call for system implementation and application of technological achievements. Only by focusing on authentic, prospective and challenging database system research can real-world problems be solved.
Zhang said: “This award is a recognition of our efforts. Through the collaboration, I have gained insight into the industry-university synergy integrating cooperation, application, and talent training. I have also gained a deeper understanding of industrial needs and developed an industrial perspective for my research, which means a lot to me.”
2. Addressing new challenges in big data computing and empowering cloud database technology
During the collaboration, Zhang Feng published 8 papers on top-tier conferences and journals such as SIGMOD and VLDB, and filed for 3 patents. His achievements are primarily applications in various practical scenarios. He explained his work: “We found that graph relations are often more capable of expressing social network data than traditional relational databases, but they occupy enormous space and produce a lot of redundancy. Therefore, we applied our direct processing techniques to graph data. For example, common friends on our social network can be expressed with a single rule rather than be stored repeatedly in the database. Likewise, corresponding analysis deals with only one rule and reuses intermediate results to save time. Other application scenarios besides graph data include stream data and GPUs.”
The development of IoT and cloud computing technologies has led to the rapid growth of cloud-edge-terminal architecture, with various heterogeneous embedded devices connecting to the Internet. However, these devices have limited capacity and weaker performance, hence they need to transfer the machine learning tasks to third-party cloud service platforms with faster GPU servers. This poses new security challenges. For that, Zhang Feng proposed a multi-party computing-based GPU secure machine learning framework in his paper “An Efficient Parallel Secure Machine Learning Framework on GPUs”, one of the earliest studies in utilizing GPUs for secure machine learning framework around the globe, and winner of the 2021 TPDS Best Paper Award.
Compared with the secure machine learning methods that do not involve GPUs, Zhang's methods improve performance by over 30x speedup while ensure data security. During the research, he encountered technical problems such as parallelization of complex computing patterns, CPU-GPU data transfer overhead within nodes, and inter-node dependencies, and proposed a series of solutions including profiling-guided adaptive GPU optimization, overlapping of computing and memory accessing within nodes, and compressed inter-node data transmission. His work alleviates the limitations of these hardware and enables them to tackle larger amount of data and perform more effectively in data transmission and storage.
As a top-tier journal recommended by the CCF, IEEE TPDS has a high academic reputation in parallel and distributed computing. In 2021, TPDS published 297 papers, among which a Best Paper and a Best Paper Nominee were chosen according to conventions.
3. Scaling new heights: a tradition of the database team of Renmin University of China
When talking about his own learning and growing experiences, Zhang Feng has always emphasized his mission and responsibility as a member of the database team. Zhang’s previous research focused on computer system architecture, but after coming to Renmin University of China, he devoted himself to the cross-field of system architecture and database technology, and delved into direct computing of compressed data based on grammar rule parsing. After identifying the research problems, he has been exploring ways to systematically deepen the research, e.g., integrating, parallelizing and indexing with new hardware devices such as GPUs.
On this journey, numerous people from the database team of Renmin University of China have blazed the trail. Zhang Feng shared: “Both my doctoral mentor at Tsinghua University, Professor Chen Wenguang, and my postdoctoral mentor, Professor Du Xiaoyong, have emphasized conducting practical and useful research. Combining theory with practice and conducting useful research are also the research principles of the database team.” In the 1980s, China had no database products with independent intellectual property rights, and the domestic market was monopolized by foreign companies. The database team has been actively contributing to large-scale computer system projects in the country, placing great importance on developing database software with high technical and financial demands, and developing domestic database products, gradually bringing database application from peripheral scenarios to core sectors. “As a young generation, we should both inherit the traditions of our predecessors while at the same time innovate, to create a better future for Renmin University of China in the field of database”, said Zhang Feng proudly.
Zhang Feng and students
While focusing on his own research, Zhang Feng also shoulders the task of passing on the responsibility and mission of the database team to the next generation. He has been the directing adviser of the Turing Class of the School of Information since 2019, and has noticed that students lack hands-on skills in study and life. As a teacher, he actively encourages students to spend time in the laboratory at the undergraduate stage. He said: “I have invited undergraduates to our research projects, and since then many of them have published high-level academic papers.” Through early laboratorial training, many undergraduate students of the School of Information have published top-tier papers in the field of data science, which have been recognized by domestic and foreign peers. They have been admitted into world-famous universities, such as Renmin University of China, Tsinghua University and Peking University for doctoral study. Students from the Turing Class of 2019 have even published top papers at ASPLOS and VLDB, and got PhD offers from UIUC and other world-renowned institutions. In combination with discipline advantages, the School of Information has developed a "3+X" computer science curriculum to cultivate students' scientific research interests. The project of "Data-Centered Training of Leading Computer Talents: Teaching Reform in the Turing Class" was awarded the first prize of Beijing Higher Education Achievement Award in 2022.
Zhang Feng keeps moving forward and scaling new heights. He conducts application-oriented research to solve real world problems in database system development, and honors traditions of the database team of Renmin University of China while blazes the trail for future endeavors.
