vPIM – Virtualization of Processing-in-Memory

Published in Compas 2022, 2022

Recommended citation: Dufy Teguia, Alain Tchana, vPIM – Virtualization of Processing-in-Memory, Compas 2022 https://brisco007.github.io/files/vPIM_Virtualization_of_Processing_in_Memory.pdf

Download abstract here

For decades, computer scientists have been interested in in-memory computing, also known as Processing In Memory (PIM). PIM refers to a broad category of technologies that enhance memory with computing capabilities, such as placing processors near the memory. This allows bringing computation closer to data, thereby avoiding data movement between RAM and the processor. Research on PIM has spanned over a decade, but few proposed designs have been developed, built, or commercialized. That being said, UPMEM has recently launched a promising product: DRAM enhanced with processors. The goal of this work is to virtualize this hardware to enable its use in the cloud. We will use Firecracker, developed by Amazon, coupled with KVM to perform the virtualization of the product. We will provide support for PIM to virtual machines, applying the split-driver method for this purpose.