Resource Disaggregation: The Past 5 years, and The Next 10 Years

Abstract: Server-centric architectures, that pack a small amount of CPU, memory and storage resources on individual servers, have been the mainstay for decades. However, the end of Dennard's scaling and the demise of Moore's Law has led to surfacing of several fundamental limitations of such architectures---the dreaded "memory wall", limited resource density, poor resource utilization, long time-to-adoption of emerging technologies, to name a few. As practitioners are realizing these limitations and are beginning to explore alternatives, it is upon us to guide the design of next-generation hardware and software infrastructure.

In this talk, I will reflect on my (ongoing) journey to realize resource disaggregation---an architecture that offers compute, memory and storage resources as individual pools, slices of which can be "stitched" over a network fabric. I will discuss how decoupling of resources in such disaggregated architectures introduces many new systems and algorithmic challenges, some progress towards resolving these challenges, the current state of affairs, and how our journey so far has led to uncovering a myriad of new research questions in systems, networking, architecture, security, and theory.