skip to main content

Torrellas receives NSF-CNS grant for modern virtualization research


Joseph Park, Illinois ECE

Josep Torrellas
Josep Torrellas

Illinois ECE Professor Josep Torrellas and CSL Assistant Professor Tianyin Xu recently received an NSF - Computer and Neural Systems (CNS) grant for their project "Rethinking Architecture and Operating Systems for Modern Virtualization Technologies." This published work won the 2020 Architectural Support for Programming Language and Operating Systems (ASPLOS) Best Paper Award. The grant consists of $850,000 spanning four years. 

The NSF-CNS grant supports research and educational activities that invent new computing and networking technologies that explore new ways of using existing technologies, particularly with the fundamental properties of computer and network systems. The CNS also supports the computing infrastructure for experimental computer science and cross-divisional activities that "foster the integration of research, education, and workforce development."

The widespread deployment of modern virtualization technologies such as containers has changed cloud computing and enabled new uses such as server-less computing and function-as-a-service. The goal of their proposal is to rethink the computer architecture and the operating system (OS) in this environment so they are high-performing, secure, and scalable.

To improve performance, Torrellas' research is rethinking the virtual address translation mechanism in processors for parallelism. Boosting security, this research proposes new hardware and OS support to protect system calls. Torrellas aims to combine replicated virtual-to-physical page translations across processes and to attain scalability for running thousands of containers on the same machine. Their work is performed in collaboration with VMware.

"With the advent of containers in the cloud and of large memory sizes enabled by non-volatile memory, the way current computers perform virtual address translations is not scalable," said Torrellas. "We need to rethink it to take advantage of the memory access parallelism provided by modern processors."