SHF: Small: The Chip Is the Network: Rethinking the Theoretical Foundations of Multicore Architecture Design

项目ID：0916752标题：芯片就是网络：重新思考多核体系结构设计的理论基础PI名称：Radu Marculescuu研究所：卡内基-梅隆大学摘要最近的cmos技术进步允许在同一片上系统(SoC)上集成数十或数百个单独可编程的处理元件，以及大量专用存储器。在这种多处理器系统中，各个处理节点可以通过片上网络(NoCs)进行通信和协调。因此，一个主要的挑战是确定以严格的方式设计和优化这种片上网络的数学技术。传统的排队和马尔可夫链缓冲区分配方法在一定程度上是有帮助的，但捕获流量的变异性是一个主要问题。从这些总体思想出发，本项目引入了一种新的统计物理方法来分析多处理器SoC的性能。更准确地说，我们利用玻色气体和通信网络中的信息流之间的类比，发展了一种全新的网络流量的数学描述。这种将网络视为气体的新建模范例可以进一步用于开发高效的片上缓冲区分配算法。新的设计方法使更高效的多处理器SoC的开发成为可能，这些SoC通过从娱乐到游戏、安全以及生物和基因工程的各种应用对社会产生巨大影响。更广泛地说，该项目的结果通过提高对设计和控制复杂系统所需的网络概念的理解水平，对其他研究团体产生了重大影响。

Project ID: 0916752 Title: The Chip Is the Network: Rethinking the Theoretical Foundations of Multicore Architecture DesignPI Name: Radu MarculescuInstitution: Carnegie-Mellon UniversityABSTRACTRecent advances in CMOS technology allow the integration of tens or hundreds of individually programmable processing elements, together with large amounts of dedicated memory, on the same system-on-chip (SoC). In such multiprocessor systems, individual processing nodes can communicate and coordinate via networks-on-chip (NoCs). Therefore, a major challenge is to determine the mathematical techniques for designing and optimizing such on-chip networks in a rigorous manner. Traditional queuing and Markov chain approaches to buffer allocation are helpful to a certain extent, but capturing the traffic variability represents a major problem. Starting from these overarching ideas, this project introduces a new statistical-physics approach for performance analysis in multiprocessor SoCs. More precisely, we develop a completely new mathematical description of network traffic using an analogy between a Bose gas and the information flow in the communication network. This new modeling paradigm where networks are seen as gases can be further used to develop efficient on-chip buffer assignment algorithms. The new design methodology enables the development of more efficient multiprocessor SoCs which have a dramatic impact on society via applications ranging from entertainment to gaming to security and to bio- and gene engineering. More broadly, the results of this project impact significantly other research communities by improving the level of understanding of networking concepts needed to design and control complex systems.