The Demikernel Datapath Architecture for Microsecond-scale Datacenter Systems

Zoom link

Abstract: Datacenter systems and I/O devices now run at single-digit microsecond latencies, requiring nanosecond-scale operating systems. Traditional
kernel-based operating systems impose an unaffordable overhead, so recent kernel-bypass OSes (e.g., Arrakis, Ix) and libraries (e.g., Caladan, eRPC) eliminate the OS kernel from the I/O datapath. However, none of these systems offer a general-purpose datapath OS replacement that meet the needs of microsecond-scale systems.

This talk proposes Demikernel, a flexible datapath OS and architecture designed for heterogenous kernel-bypass devices and microsecond-scale datacenter systems. We have built two prototype Demikernel datapath OSes and show that minimal effort is needed to port existing microsecond-scale systems. Once ported, Demikernel lets applications run across heterogenous kernel-bypass devices with nanosecond-scale overheads and no code changes.

Bio: Irene Zhang is a Principal Researcher at Microsoft Research. Her work focuses on datacenter operating systems and distributed systems, especially making new datacenter hardware technologies more widely usable by highly-demanding datacenter applications. Irene completed her PhD in 2017 at the University of Washington, where her PhD thesis focused on distributed systems that span mobile devices and cloud servers. Her thesis work received the ACM SIGOPS Dennis Ritchie doctoral dissertation award and the UW Allen School William Chan Memorial dissertation award. Before her PhD, Irene was a member of the virtual machine monitor group at VMware, where she worked on memory resource management and virtual machine checkpointing.