CAREER: MemMax: Maximizing Cyberinfrastructure Memory Utilization via Hardware Acceleration for OS-level Memory Utilization Management

超级计算机和计算数据中心等网络基础设施系统的性能增长对美国经济和福祉至关重要;它有助于提高科学和工程生产力，这对美国的全球竞争力至关重要，并有助于发展美国数字经济，这是美国整体经济的重要组成部分。直到最近，摩尔定律还在指导着硬件资源在相同成本下的指数级增长，并且一直是网络基础设施性能增长背后的关键驱动因素。随着摩尔定律预测的增长放缓，持续的性能提升需要网络基础设施最大限度地利用其可用的硬件资源，特别是计算内存。目前，与物理和/或理论上可能的情况相比，网络基础设施系统通常仅利用其存储器的一小部分。由于存储器主要用作性能增强器，因此存储器利用不足会导致性能不佳和对额外计算资源的不必要投资。该项目似乎通过确定如何更好地使用超级计算机和云计算系统中的内存资源来解决这个问题，从而提高这些系统的效率和性能。MemMax项目积极吸引研究生和本科生参与，沿着也拓展到K-12学生。MemMax探索如何共同设计CPU和操作系统，以最大限度地提高网络基础设施系统的内存利用率，从而以用户透明的方式大幅提升性能，最高可达4倍。MemMax由两个研究方向组成，一个针对HPC系统，另一个针对云系统，因为这两种类型的系统有不同的原因导致其内存利用不足。该研究方法包括用于表征现有系统行为的真实系统测量、用于验证MemMax功能正确性的硬件原型设计以及用于量化MemMax实现的性能改进的架构模拟。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估而被认为值得支持。

The performance growth of cyberinfrastructure systems, such as supercomputers and computing data centers, is important to US economy and well-being; it helps boost scientific and engineering productivity, which is vital for US global competitiveness and helps grow the US digital economy, which is an important part of overall US economy. Until recently, Moore’s Law guided the exponential growth of hardware resources under the same cost and has been a key driving factor behind cyberinfrastructure performance growth. As the growth predicted by Moore’s Law slows down, sustaining performance gains requires cyberinfrastructure to maximize the utilization of its available hardware resources, especially computing memory. Currently, cyberinfrastructure systems often only utilize up to a small fraction of their memory compared to what is physically and/or theoretically possible. As memory primarily serves as a performance enhancer, memory under-utilization causes under-performance and unnecessary investment in additional computing resources. This project seems to address this problem by identifying ways to better use memory resources in supercomputers and cloud computing systems, thereby increasing the efficiency and performance of these systems. The project, MemMax, actively involves both graduate and undergraduate students, along with outreach to K-12 students.MemMaX explores how to co-design CPU and OS to maximize memory utilization in cyberinfrastructure systems to boost their performance both user-transparently and substantially by a factor of up to 4. MemMax consists of two research thrusts, one targeting HPC systems and targeting cloud systems, as these two types of systems have different causes for their memory underutilization. The research methodology consists of real-system measurements to characterize the behavior of existing systems, hardware prototyping to valid the functional correctness of MemMax, and architectural simulations to quantify the performance improvement MemMax achieves.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.