CSR: Medium: Performant Architecturally Diverse Systems via Aspect-oriented Programming

点击翻译按钮获取中文摘要

Architecturally diverse computers are systems that incorporate hardware accelerators such as graphics engines, and reconfigurable logic to achieve some tangible benefits in terms improved performance, and decreased power are difficult to program. This project seeks to dramatically improve the programmability of architecturally diverse computers by enabling them to tune themselves, freeing the application developer to focus on correctness issues rather than performance issues.This will be accomplished by using aspect-oriented programming techniques to expose tuning knobs that have direct impact on the benefit of interest, notably, performance or power consumption. Understanding the impact of these tuning knobs will be achieved using queueing network models which are automatically derived from the application's specification. The queueing models will be used to guide an automated control process that seeks to maximize the benefit of interest by adjusting the tuning knobs.The project will utilize a number of computational science applications as test cases, providing benefits to both of the computational biology and data science communities. Both graduate and undergraduate students will be involved in the research, and several university-level programs will be used to help attract members of underrepresented populations.Relevant data, code, and other supplementary materials will be placed in the university library's permanent repository at https://openscholarship.wustl.edu/data/. This repository will provide persistent, curated access to both code and data, including support for DOIs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.