In this insightful episode of the Computer Architecture Podcast, hosts Dr. Suvinay Subramanian and Dr. Lisa Hsu are joined by the distinguished Professor Mark Hill. A Professor Emeritus of Computer Science at the University of Wisconsin-Madison with 32 years of teaching experience, Professor Hill is renowned for his extensive contributions to parallel computer system design, memory system design, and computer simulation. He is the inventor of the widely used 3Cs model of cache behavior, a co-inventor of the SCDF memory consistency model, and a recipient of the 2019 Eckert–Mauchly Award. Notably, he recently completed his term as chair of the Computing Community Consortium (CCC).
The conversation delves into the exciting and increasingly necessary realm of cross-layer optimizations in computer architecture, driven by the slowing of Dennard scaling and Moore's Law. Professor Hill shares his perspective on why this approach is becoming critical for performance and energy improvements. The discussion explores the impact of industry verticalization, where companies increasingly control their entire hardware and software stacks, and how this trend both enables and challenges cross-layer innovation.
Professor Hill also offers valuable insights into fostering effective collaborations between academia and industry, drawing from his extensive experience working with 160 different co-authors and his leadership role in the CCC. He discusses strategies for identifying impactful research problems, the importance of communication and taking criticism, and the "meta-skills" crucial for success in the field. The episode also touches upon Professor Hill's personal journey into computer architecture, offering inspiration and practical advice for students and researchers alike. Since the recording, Professor Hill has joined Microsoft as a Partner Hardware Architect with Azure.
Chapters
00:00:00 — Podcast and Guest Introduction
01:27:16 — Welcome and What Excites Mark Hill: The Necessity of Cross-Layer Optimizations
01:37:621 — The Current Landscape: End of Dennard Scaling and the Rise of Cross-Layer Work
02:30:618 — Verticalization in Industry: Enabling Cross-Layer Optimizations
04:06:925 — Navigating Cross-Stack Work with Diverse Expertise
04:25:665 — The Challenge and Opportunity of Cross-Layer Collaboration: The "Direct Segments" Example
06:36:621 — Why "Direct Segments" Weren't Widely Adopted and the Role of Incremental Progress
07:38:828 — The Difficulty of Cross-Layer Optimization: Moving "All Seven Dwarves at Once"
08:17:118 — How Industry Verticalization (e.g., Apple) Enables Cross-Layer Innovation
09:59:415 — Drawbacks of Verticalization: Loss of Standardization and Competition
11:18:655 — The Cyclical Nature of Industry: Echoes of Past Vertical Integration
12:26:290 — Learning from Past Cycles: Applying Lessons to Current Cross-Stack Challenges
13:08:740 — Identifying Problems First: The "RAID Story" and Framing for Simplicity
16:41:610 — Academia's Role: Picking Opportunities and Scoping Problems
17:06:541 — How Academics Can Identify and Engage with Industry Problems: Trends, Internships, Sabbaticals
18:44:564 — Industry Engaging with Academia: Benefits (Talent, Influence) and Challenges (Time)
20:04:866 — Elements of Good Collaboration: Complementary Skills and Mutual Respect
20:27:913 — Mark Hill's Collaboration Philosophy: The Tortoise and the Hare, and Humility in Credit
21:12 — ,145] Trusting Your Gut and Developing a "Nose" for Research Directions
22:22:409 — The Importance of Contrarian Thinking and a Balanced Research Portfolio
23:04:795 — Visioning Work and the Computing Community Consortium (CCC) White Paper
23:44:594 — The "Herding Cats" Skill: Leading Collaborative Visioning Efforts
25:12:906 — "The Future is Here, Just Unevenly Distributed": Insights from the CCC White Paper
25:40:816 — Meta-Skills for Students: Beyond Technical Knowledge
26:16:46 — Cultivating "Fire in the Belly" and Essential Skills for Students
28:00:646 — The Crucial Role of Communication and Learning to Take Criticism
29:00:196 — Further on Communication and Criticism: Separating Work from Self
31:03:386 — The Architecture Community: Strengths (Industry Connection) and Areas for Improvement (Too Many Incremental Papers, Reviewing Strain)
32:14:996 — The Value of Simplicity in Research vs. "Safe" Incremental Papers
34:31:674 — Long-Term Impact vs. Incremental Work: A Discussion on Research Focus
35:56:980 — Professor Mark Hill's Journey into Computer Architecture: From Mechanical Adders to Academia
37:38:700 — Navigating Academia: Mark Hill's Personal Reflections and Career Path
39:40:950 — The Allure of Academia and the Evolving Landscape of Computer Architecture
41:03:386 — Mark Hill's Personal Journey: A First-Generation College Student's Path
41:50:906 — Overcoming Discouragement and Finding Your Path in Research: Different Journeys, No Pedestals
45:13:148 — Reflections on a Career in Academia and the Joy of the Field
47:41:716 — The Allure of Academia for a First-Generation College Student
49:50:906 — Concluding Remarks
Takeaways
Cross-layer optimization is becoming essential: With the end of Dennard scaling and slowing of Moore's Law, looking across hardware and software layers is critical for future performance and energy gains.
Industry verticalization is a double-edged sword: While companies controlling their full stack (like Apple or Google) can facilitate deep optimizations, it can also reduce broader competition and standardization that previously drove innovation.
Problem identification is paramount in research: Focus on identifying significant, real-world problems first, rather than starting with a solution and looking for a problem. As illustrated by the "RAID story," framing the problem correctly can make the solution almost obvious.
Effective communication and constructive criticism are vital skills: Researchers must not only discover but also effectively communicate their findings. Learning to give and receive criticism—separating critique of work from personal critique—is crucial for growth and collaboration.
"Fire in the belly" and a diverse research portfolio are key to academic success: Passion and intrinsic motivation are more important than just technical prowess. Academics should also cultivate a research portfolio that balances safer, shorter-term projects with more ambitious, longer-term, and potentially contrarian ideas.