Computer Systems & Security Group, USTC

Home
Publications
Projects
Courses
Members

PGO: Profile-guided Optimizations for Android

Overview:

The kernel is the core of the operating system, responsible for the execution of core functions such as CPU resource scheduling, process management, and file system. Improving the performance of the kernel can speed up program execution running on it. Static compiler optimization is a traditional method to improve the running speed of the program. Most of the static compiler optimization methods are based on heuristics' algorithm, this optimization method is sometimes not suitable for actual running scenarios, even produces negative optimization. In order to overcome the limitations of static optimization, related researchers have gradually explored and summarized a technology that utilizes important information at runtime to guide compilation optimization, which is often referred to as PGO.

PGO is an effective method to improve program performance. At present, this technology has been applied in large open source projects such as PHP, Chrome browser, .NET, Firefox and Linux kernel, and Significant performance gains have been achieved. The Android kernel is based on the Linux kernel. Some researchers have applied the PGO technology to the Android kernel and achieved significant performance improvements.

Unfortunately, using PGO to optimize the Android kernel will consume a lot of manpower. First, the kernel needs to be instrumented. In this process, about 20 modifications to the Android kernel source code are required. If manual instrumentation is used, it usually takes at least half an hour to complete, and it is prone to errors. Second, it is necessary to complete the compilation of the kernel, replacement, collection of profile files, optimization and recompilation, etc., which is quite cumbersome and time-consuming. There are many Android applications. Third, in order to improve the overall performance of the system, it is necessary to collect profile files in many typical scenarios, that is, the above process needs to be repeated many times, and the demand for manpower is further expanded. Last, due to frequent updates of Android applications, the previous optimization may fail after the application update. Therefore, the entire process needs to be re-executed to optimize the Android kernel. In this case, the demand for manpower will reach a new height, and it is extremely unfavorable to be applied to actual production.

The goal of the PGO project is to investigate the PGO technology in Android kernel. To be specific, this project aims to: :

  • investigate the key point to make PGO technology achieve better optimization result on Android platform
  • develop practical PGO framework for Android platform.

Publications:

Members:

  • Qiliang Fan
  • Baojian Hua
  • Jiali Liang
  • Zhizhong Pan
  • Shaobo Su
  • Keyuan Zong