More about HKUST
Optimizing distributed build time in large industrial build cluster
The Hong Kong University of Science and Technology Department of Computer Science and Engineering MPhil Thesis Defence Title: "Optimizing distributed build time in large industrial build cluster" By Mr. Bangyan DU Abstract: Large IT companies encounter significant compilation demands due to their extensive codebases, particularly when managing and hosting them in a multirepo approach. For instance, WeChat has over 1,500 code repositories totaling more than 800 GB, covering various services. With more than 800 frontline developers making over 4,000 code sub- missions daily, the average compilation demand exceeds 20,000 requests. Distributed build clusters encounter two main challenges: resource scheduling and the absence of dis- tributed preprocessing. Firstly, effective resource scheduling is crucial for optimizing the use of existing cluster to compile more C++ files quickly. The second challenge stems from the lack of distributed preprocessing in a build cluster, primarily due to the high costs associated with uploading all related header files from the local master machine to the re- mote slave machines for a given C++ source file. To address the first issue, we propose a resource prediction model that estimates the compilation resources required for each C++ source file. This model facilitates the allocation of adequate remote slave machines and helps eliminate typical errors like Out of Memory errors and Deadline Exceeded errors in the build cluster. For the second issue, we present a two-stage distributed preprocessing solution and further minimize the huge number of system calls of GCC C++ compiler to help reduce the preprocessing time on the local master machine. Our experience report not only sheds light on WeChat’s build cluster but also offers insights into effective practices of build time reduction for large-scale IT companies. Date: Tuesday, 27 May 2025 Time: 1:00pm - 3:00pm Venue: Room 2128B Lift 19 Chairman: Dr. Shuai WANG Committee Members: Prof. Charles ZHANG (Supervisor) Dr. Ziyi LIU