EAGER:Stochastic Thermodynamics of Distributed Computation

点击翻译按钮获取中文摘要

Society is witnessing explosive growth in the deployment of distributed computational systems, especially in networked system architectures, and their accompanying software protocols. One of the main performance challenges in these systems is poor energy efficiency; they dissipate too much heat. Similarly, a major goal of chip design has been to minimize the heat dissipation of individual nanoscale components and/or to apply phenomenological laws to reduce the heat dissipation of the global architecture. However, theoretical computer science research has focused on the design of algorithms that minimize time and memory costs per computation. To date there has been little use of first principles of physics to analyze how the global structural organization of computers affects their heat dissipation. Thus, the goal of the proposed research is to develop an understanding of how the physical features of a distributed computational system (e.g., communication rates, network topology, and heterogeneity) affect its performance, specifically the relationship between its heat dissipation, space and time resource costs, robustness, speed, and computational power. Besides the technical work the project will also undertake: (i) outreach via new coursework in the context of this project, (ii) support of mentoring students/postdocs thereby increasing the pool of technically trained workforce, and (iii) public and media outreach for increasing public understanding of science.This exploratory project will aim to start to fill this gap in scientific understanding by exploiting the recently developed tools of stochastic thermodynamics. The PI will use investigate how the following features of the global physical architecture of a computer affect its cost: (1) communication rate and accuracy of information transmission between subsystems, (2) characteristics of the network topology of the architecture, particularly hierarchy and modularity, (3) heterogeneity among the individual subsystems. Specifically, the project will investigate how these design features affect trade-offs between time, memory, and heat costs; desired computational capabilities; and robustness against environmental fluctuations or faulty subsystems. The investigations will involve computer simulations, which will help reveal patterns and trends in these trade-offs, as well as theoretical analyses, which will help uncover the mathematical basis of the observed pattern.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.