THIS IS A CONTINUATION OF N00014-14-1-0549 De-bloating Software via Automated Selection of Libraries
Abstract
Statement Of WorkThe proposed research will investigate and develop called Library Auto-Selection (LAS). LAS is a new type of dynamicoptimizations that exploits the developer?s insights as well as the compiler and runtime system support toautomatically remove inefficiencies during the program execution. The primary tasks in the proposed project are:1. Developing a performance specification language.2. Developing LAS compiler and runtime system support.3. Developing techniques to reduce overhead.4. Evaluating LAS on object-oriented applications at different levels.ObjectiveThe PIs proposes to improve the state of the art of automated bloat removal by developing a novel optimization technique called Library Auto-Selection (LAS). LAS is a new type of dynamic optimizations that exploits the developer?s insights as well as the compiler and runtime system support to automatically remove inefficiencies during the program execution.ApproachThis research is geared toward simplifying software/program execution which has a positive effect on the software performance and reducing the attack surface, improving the efficiency and security of our cyber infrastructure. The PI approaches this goal by developing Library Auto-Selection (LAS), which include performance specification language which allows developers to describe their insights on the performance tradeoffs among the implementations of the data structure/algorithm. The source code and the performance specification language will then be used to performcompilation and optimization. A runtime profiling system is also generated for monitoring usage of theprogram/algorithm and its data structure, and to allow for online (real-time, in execution) switching to better performing, reduce overhead, implementation of the algorithm and data structure. The PI plans to evaluate LAS on various applications including large-scale web servers, data-intensive frameworks, and mobile apps to study its effectiveness. Merit/ONR RelevanceThe proposed research is well in line with one of the direction of ONR basic science research focus on future software, especially on software complexity reduction. The propose research is trying to address the runtime redundancies and inefficiencies that significantly impact their performance, security and scalability. Dr. Xu proposes to significantly improve the state of the art of automated bloat removal by developing a novel optimization technique called LibraryAuto-Selection (LAS). LAS is a new type of dynamic optimizations that exploits the developer?s insights as well as the compiler and runtime system support to automatically remove inefficiencies during the program execution. This research is geared toward simplifying software execution which has a positive effect on the software performance and reducing the attack surface, hence improving the efficiency and security of our cyber infrastructure. A secure and efficient cyber infrastructure enhances the potential success for Navy missions.Progress??? Finished the Fa??ade compiler and runtime system development??? Finished a preliminary study of using the technique to improve scalability of several Big Data systems??? Finished the PerfBlower tool development for performance problem amplification??? Finished the GraphQ system development for efficient graph query answer??? Finished the ITask system development that can significantly improve the performance and scalability of data-parallelsystems??? Published papers in ASPLOS???15, USENIX ATC???15, ECOOP???15, and SOSP???15??? Two tools were released:o PerfBlower (https://bitbucket.org/fanglu/perfblower-public)o GraphQ (https://bitbucket.org/wangk7/graphq)
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 26, 2018
- Source ID
- N000141612149
Entities
People
- Guoqing Xu
Organizations
- Naval Information Warfare Center Pacific
- Office of Naval Research
- United States Navy