Collaborative Research: SHF: Medium: Hardware and Software Support for Memory-Centric Computing Systems

现代数据中心应用程序正变得越来越耗费内存，并受到传统冯·诺伊曼架构限制的严重制约。Compute Express Link(CXL)的新兴硬件生态系统带来了前所未有的机遇，可以实现全新的内存计算模式，有效缓解未来以内存为中心的应用程序的冯·诺伊曼瓶颈。然而，为了充分释放这一新兴技术的潜力，必须充分解决几个基础研究挑战。该研究项目采用整体和连贯的方法来开发解决方案，以正面应对挑战，为内存中计算基础设施的未来铺平道路，并从根本上影响以内存为中心的应用程序。此外，该项目的学术活动通过向学生提供培训机会、丰富课程和课堂教学以及促进教育和外联活动来扩大其影响。该项目是基础研究的先锋，旨在克服阻碍内存计算实现的三个关键挑战：内存资源碎片化、内存共享的体系结构支持不足以及硬件和软件模型之间的低效分离。利用新兴的CXL技术，该项目采用了系统的设计方法来解决这些复杂的问题。它涉及跨越系统堆栈内多个层的全面努力，特点是开发先进的硬件功能，优化系统中的内存资源利用，并将硬件支持整合到通用编程平台中，以加快应用程序的速度。该项目的研究为满足未来以内存为中心的计算领域中数据密集型应用程序的迫切需求的基础研究奠定了基础。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Modern data-center applications are becoming increasingly memory-intensive and severely constrained by the limitations of the traditional von Neumann architecture. The emerging hardware ecosystem of Compute Express Link (CXL) opens an unprecedented opportunity to enable a radically new in-memory computing paradigm that can effectively mitigate the von Neumann bottleneck for future memory-centric applications. However, to fully unlock the potential of this emerging technology, several fundamental research challenges must be adequately addressed. This research project takes a holistic and cohesive approach to develop solutions that tackle the challenges head-on, paving the way for the future of in-memory computing infrastructure and fundamentally impacting memory-centric applications. Furthermore, the academic activities in this project extend their impact by providing training opportunities to students, enriching curriculum and classroom teaching, and contributing to educational and outreach initiatives. This project spearheads the fundamental research aimed at overcoming three pivotal challenges that hinder the realization of in-memory computing: fragmented memory resources, insufficient architectural support for memory sharing, and inefficient separation between hardware and software models. Leveraging the emerging CXL technology, this project adopts a systematic design methodology to address these intricate issues. It involves comprehensive efforts spanning multiple layers within the system stack, featuring the development of advanced hardware functionalities, the optimization of memory resource utilization in the system, and the integration of hardware support into general programming platforms to expedite applications. The research undertaken in this project lays the groundwork for fundamental studies that cater to the pressing demands of data-intensive applications in the realm of future memory-centric computing.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.