Parallel Data Laboratory Talks
- Remote Access - Zoom
- Virtual Presentation - ET
- PHILLIP GIBBONS and MOHAMMAD BAKHSHALIPOUR
- Phillip: Professor, Computer Science Department and Department of Electrical & Computer Engineering
- Mohammad: Ph.D. Student, Department of Elecrical & Computer Engineering
- Carnegie Mellon University
► Speaker 1 → Phillip Gibbons, Carnegie Mellon University, Professor, CSD/ECE
Processing-in-Memory: Theory and Practice
As computational resources become more efficient and data sizes grow, data movement is fast becoming the dominant cost in computing. Processing-in-Memory (PIM), an idea dating back to at least 1970, is now emerging as a key technique for reducing costly data movement, by enabling computation to be executed on compute resources embedded in memory modules. While there has been considerable recent work on the systems/architecture/technology side of PIM, there has been very little work addressing the theory/programming/algorithm side. Open problems include: How does/should programming/algorithm design on PIM systems differ from traditional shared/distributed settings? What are the fundamental limitations/trade-offs in using processing-in-memory?
This work-in-progress talk highlights our very early results on addressing these questions. As a driving application kernel, we focus on a novel PIM-friendly index structure supporting parallel inserts/deletes, point queries, and range queries. Our index addresses head-on the inherent tension between minimizing communication and achieving load balance in PIM systems. The talk will also report on our early experiences implementing our ideas on UPMEM’s 2,560-module PIM system.
Phillip Gibbons is a Professor in the Computer Science Department and the Electrical & Computer Engineering Department at Carnegie Mellon University. He received his Ph.D. in Computer Science from the University of California at Berkeley in 1989. Prior to joining CMU, Gibbons was a researcher at Bell Laboratories and Intel Research Pittsburgh, and co-director of the Intel Science and Technology Center for Cloud Computing. His research areas include big data, parallel computing, distributed systems, databases, and computer architecture. His 200+ publications span systems and theory, and have been cited 27,000+ times with an h-index of 82. He recently won the ACM Paris Kanellakis Theory and Practice Award for pioneering the foundations of streaming data analytics. He was founding Editor-in-Chief for the ACM Transactions on Parallel Computing, Associate Editor for the Journal of the ACM and other journals, and program or area chair for a number of conferences in parallel/distributed computing, sensor networks, and databases. Gibbons is both an ACM and IEEE Fellow.
► Speaker 2 → Mohammad Bakhshalipour, Carnegie Mellon University, Ph.D. Candidate
Algorithm/Hardware Co-design for Mobile Robot Path Planning
I will present RACOD, an algorithm/hardware co-design for real-time mobile robot path planning. RACOD consists of two main components: a hardware accelerator for speeding up the costly collision detection operations, and an algorithm-level speculation technique for aggressively parallelizing the robot path search algorithm. By affording multiple collision detection hardware accelerators and speculating up to a similar amount of parallelism, RACOD significantly improves the performance of path planning for mobile robots. This work demonstrates how Prescriptive Memory, which enables conveying semantic information from applications to platforms, can be used to significantly speed up an AI-in-the-wild application kernel.
Mohammad Bakhshalipour received his M.Sc. and B.Sc. in Computer Engineering from Sharif University of Technology. Since 2019, he has been working toward a Ph.D. degree at Carnegie Mellon University. His research interests include prescriptive memory, hardware/software co-design, robotics in the wild, and data prefetching.