In this episode of the Computer Architecture Podcast, hosts Dr. Suvinay Subramanian and Dr. Lisa Hsu welcome Professor Sarita Adve, the Richard T. Chang Professor of Computer Science at the University of Illinois at Urbana-Champaign. Professor Adve is a distinguished researcher whose work spans the entire system stack, from hardware and programming languages to operating systems and applications. She is known for her contributions to memory consistency models for C++ and Java, her innovative work in heterogeneous computing, and her leadership in the ACM SigArch Cares movement.
The central theme of this episode revolves around Professor Adve's recent and ongoing research in domain-specific systems for Augmented Reality (AR), Virtual Reality (VR), and Extended Reality (XR). She discusses the origins and motivations behind her group's work on the "Elixir" project, a fully open-source extended reality testbed designed to democratize XR systems research. The conversation delves into the unique challenges and opportunities in building systems for XR, including the complexities of hardware-software co-design, the need for new performance and quality metrics, and the importance of community collaboration to advance this cutting-edge field. Professor Adve shares personal anecdotes and insights gleaned from her extensive career, offering valuable perspectives on research, mentorship, and fostering an inclusive and supportive academic community.
Chapters
00:01:38 — Introduction and What Gets Professor Adve Up in the Morning
00:03:30 — Professor Adve's Personal Motivation and Journey into XR Research
00:04:40 — The Frustration with Toy Applications and the Shift to Real Systems
00:05:45 — The Genesis of the Elixir Project: From Seed Proposals to AR/VR Benchmarks
00:08:03 — The Challenges and Excitement of XR Systems Research: A "Real System"
00:09:31 — Key Challenges in XR: Orders of Magnitude, Multidisciplinary Nature, Co-design, and Metrics
00:11:03 — The Problem of Closed Systems and the Drive for Open-Source XR
00:11:52 — Diverse Challenges in Accelerating XR Applications
00:12:55 — Top-Down System View vs. Bottom-Up Component Acceleration in XR
00:14:53 — The Importance of Primitives and Cross-Layer Optimization in XR Accelerators
00:16:13 — The Difficulty of XR Research: Moving Beyond Traditional Consistency Models
00:16:48 — XR as a General-Purpose Framework and Standing on the Shoulders of Giants
00:20:09 — Defining Elixir: A Full-Blown Open-Source XR System
00:20:31 — Components of a Generic XR System: Perception, Visual, Audio, and Runtime
00:24:03 — The Elixir Consortium: Democratizing XR Research and Standardization
00:25:54 — Key Metrics in XR: Motion-to-Photon Latency and Its Complexities
00:28:33 — Variability in XR Systems and Scheduling Challenges
00:30:36 — Hardware-Software Co-design for XR: Intermediate Representations and HPVM
00:34:49 — The Importance of Pulling Together Different Layers of the Stack
00:48:18 — Professor Adve on the Cares Initiative: Goals and Impact
00:53:13 — The Viral Spread of the Cares Initiative Across ACM SIGs
00:56:47 — General Life and Career Advice: Being Comfortable Being Uncomfortable
00:58:05 — The Importance of People, Passion, and Paying It Forward in Academia
Takeaways
XR is a "Real System" Challenge: Building systems for AR/VR/XR is incredibly complex, requiring a holistic, top-down approach that considers the entire stack, from hardware to user experience, rather than just optimizing individual components.
The Need for New Metrics and User Studies in XR: Traditional performance metrics are insufficient for XR; end-to-end user experience, including factors like motion-to-photon latency and even qualitative aspects like "does the user barf?", are critical. This necessitates incorporating user studies into systems research.
Open-Source and Community Collaboration are Vital for XR Advancement: The Elixir project and its associated consortium aim to democratize XR research by providing an open-source testbed and fostering collaboration to develop standardized benchmarks and tools.
Cross-Layer Co-design and Generalizable Techniques: Effective XR systems require deep co-design across hardware and software layers. Identifying key "primitives" for acceleration, rather than just components, and developing generalizable techniques (like intermediate representations) are crucial.
The Cares Initiative Highlights the Importance of a Supportive Community: The Cares movement, co-founded by Professor Adve, aims to combat discrimination and harassment in academic conferences, fostering a more inclusive and respectful environment, which is essential for scientific progress.