Where is Memory Disaggregation 5 years after CXL 1.1?
Abstract:
CXL promises a lot including increased memory bandwidth/capacity and cross-node pooling and sharing. Within Microsoft Azure, we’ve been part of the push to develop and deploy CXL in datacenters since the start. After wins as well as wrong turns, we understand key use cases and incremental deployment steps.
This talk will review technical challenges and learnings from the perspective of a major cloud platform. We will show how research and industry-standard artifacts have supported this journey including our focus on small CXL pools without switches (Pond) and hardware-codesigned tiering (Memstrata). We also describe ongoing work on advanced use cases of CXL pools with a focus on deployable hardware, which outlines current industry and research frontiers.
Short Bio:
Daniel is a Principal Researcher at Azure Systems Research and an Affiliate Assistant Professor in Computer Science at the University of Washington. His research focuses on improving the efficiency, sustainability, and reliability of cloud platforms. Before joining Microsoft Azure, he was a member for the Systems Research group at Microsoft Research. He is the recipient of an ACM ASPLOS 2023 distinguished paper award, USENIX OSDI 2023 best paper award, 2021 ACM SOSP Best Paper Award, 2018 Mark Stehlik Fellowship at Carnegie Mellon University, and best paper awards at IFIP Performance and ACM WiSec.