CAREER: Redefining Memory Virtualization in Clouds

NSF Award Search · 01002930DB NSF RESEARCH & RELATED ACTIVIT · $586,027 · view on nsf.gov ↗

Abstract

Modern cloud applications, such as those involving artificial intelligence, have become increasingly memory intensive. These applications often require large amounts of memory to achieve high performance. Due to its poor scaling properties, traditional dynamic random-access memory (DRAM) has become a bottleneck and a major infrastructure cost in clouds, where DRAM is virtualized to serve applications running in virtual machines (VMs). To address the DRAM scalability issue, emerging and future memory (EFM) such as Compute Express Link (CXL)-based memory has demonstrated high potential. EFM will encompass heterogeneous memory with multiple memory tiers and distinct characteristics such as cost and volatility. Traditional memory virtualization was primarily designed for virtualizing homogeneous volatile DRAM. It will incur high overhead, lack mechanisms for reducing cloud memory costs, and offer limited usability when used for virtualizing EFM. This CAREER project will redefine memory virtualization for EFM, aiming to significantly reduce cloud memory costs, while offering high performance and usability for modern cloud applications. This project incorporates innovative techniques to minimize virtualized EFM address translation overhead, virtualize slow memory as fast memory in EFM virtualization, and improve VM live migration performance. The success of this CAREER project is expected to enable data centers utilizing current and future cloud systems to achieve high performan

Key facts

NSF award ID
2440456
Awardee
University of Rhode Island (RI)
SAM.gov UEI
CJDNG9D14MW7
PI
Weiwei Jia
Primary program
01002930DB NSF RESEARCH & RELATED ACTIVIT
All programs
CAREER-Faculty Erly Career Dev, EXP PROG TO STIM COMP RES
Estimated total
$586,027
Funds obligated
$339,666
Transaction type
Continuing Grant
Period
07/15/2025 → 06/30/2030