CRII: SHF: A Novel Address Translation Architecture for Virtualized Clouds

Due to its cost effectiveness and energy efficiency, persistent memory (PM) has been increasingly deployed in cloud environments where PM can be virtualized and utilized as both memory and storage devices in virtual machines. Address translation from virtual (a virtual memory address or a file logical address) to physical (a physical location on memory or storage) is critical to the performance of both PM memory system and PM storage system. The state-of-the-art address translations, however, negatively impact the overall system performance in cloud computing because they were originally designed for dynamic random access memories (DRAM) virtualization and traditional storage systems such as hard disk drives (HDDs) and solid state drives (SSDs). This project aims to fundamentally change how address translation is done in the cloud environments specifically beneficial to PM memory and PM storage. The idea is to bypass and short-circuit the lengthy translation process of traditional techniques by leveraging unique observations through extensive experiments. The success of this project has potential to advance next-generation cloud computing and contribute to educational activities by integrating research outcomes into new curriculum and teaching emerging cloud computing technologies to both college and K-12 students. This project introduces XLANE, a general, effective, and scalable address translation architecture for virtualized clouds with high efficiency. The goal is to fundamentally change the traditional address translation techniques while retaining full compatibility to the default designs. XLANE includes two techniques: direct memory translation (DMT) and direct file translation (DFT). DMT enables the reduction of the number of sequential memory accesses from 24 to two in each address translation for memory virtualization. DFT achieves the first direct file mapping design for virtualizing ultra-low latency storage devices like PM with efficient direct file translation. XLANE is a general approach that benefits not only PM systems but also compute express link (CXL)-based memory systems, tiered memory systems, and many others. This project will enable modern clouds and data centers to run emerging data-intensive applications (e.g., graph computing and artificial intelligence) on virtualized PM and CXL-based memory with exceptional performance and scalability, as well as low overhead and energy consumption. This project is jointly funded by the Software and Hardware Foundation (SHF) core research program and the Established Program to Stimulate Competitive Research (EPSCoR). 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.