Ali-Reza Adl-Tabatabai is a Senior Director of Engineering at Uber, where he leads the team responsible for the tools and application frameworks that Uber developers use daily. Ali received his Ph.D. in Computer Science from Carnegie Mellon University.

This talk will describe the Pegasus workflow management system and how the system first designed for distributed, collaborative science in 2001 has evolved to support of a variety of applications executing in heterogenous environments. The talk will illustrate core Pegasus principles and how they enabled science workflows to start moving to the cloud in 2008. It will also describe how the needs of applications in astronomy, gravitational-wave physics, earthquake science, and other domains led the research and development of the workflow management capabilities. Finally, the talk will touch upon the challenges ahead.

Ewa Deelman received her PhD in Computer Science from the Rensselaer Polytechnic Institute in 1998. Following a postdoc at the UCLA Computer Science Department she joined the University of Southern California’s Information Sciences Institute (ISI) in 2000, where she is serving as a Research Director and is leading the Science Automation Technologies group. She is also a Research Professor at the USC Computer Science Department and an AAAS and IEEE Fellow. The USC/ISI Science Automation Technologies group explores the interplay between automation and the management of scientific workflows that include resource provisioning and data management. Dr. Deelman pioneered workflow planning for computations executing in distributed environments. Her group has led the design and development of the Pegasus Workflow Management software and conducts research in job scheduling and resource provisioning in distributed systems, workflow performance modeling, provenance capture, and the use of cloud platforms for science.

Imagine some product team inside Google wants 100,000 CPU cores + RAM + flash + accelerators + disk in a couple of months. We need to decide where to put them, when; whether to deploy new machines, or re-purpose/reconfigure old ones; ensure we have enough power, cooling, networking, physical racks, data centers and (over longer a time-frame) wind power; cope with variances in delivery times from supply logistics hiccups; do multi-year cost-optimal placement+decisions in the face of literally thousands of different machine configurations; keep track of parts; schedule repairs, upgrades, and installations; and generally make all this happen behind the scenes at minimum cost. And then after breakfast, we get to dynamically allocate resources (on the small-minutes timescale) to the product groups that need them most urgently, accurately reflecting the cost (opex/capex) of all the machines and infrastructure we just deployed, and monitoring and controlling the datacenter power and cooling systems to achieve minimum overheads - even as we replace all of these on the fly. This talk will highlight some of the exciting problems we're working on inside Google to ensure we can supply the needs of an organization that is experiencing (literally) exponential growth in computing capacity.

John Wilkes has been at Google since 2008, where he is working on automation for building networks for warehouse scale computers. Before this, he worked on cluster management for Google's compute infrastructure (Borg, Omega, Kubernetes). He is interested in far too many aspects of distributed systems, but a recurring theme has been technologies that allow systems to manage themselves. He received a PhD in computer science from the University of Cambridge, joined HP Labs in 1982, and was elected an HP Fellow and an ACM Fellow in 2002 for his work on storage system design. Along the way, he’s been program committee chair for SOSP, FAST, EuroSys and HotCloud, and has served on the steering committees for EuroSys, FAST, SoCC and HotCloud. He’s listed as an inventor on 50+ US patents, and has an adjunct faculty appointment at Carnegie-Mellon University. In his spare time he continues, stubbornly, trying to learn how to blow glass.

AI services are widespread across many services at Meta. Organic demand for these workloads continues to grow, in addition to scaling of the workload complexity in terms of compute, data, memory, and communication requirements. This separates the design of AI platforms from traditional compute platform, in that the scaling of the complexity outpaces the rate at which the underlying compute, memory, and networking technology improves generation over generation, requiring end to end codesign and optimization of the HW and SW. This talk overviews some of the most relevant AI workloads at Meta, their compute requirements, and how it impacts the design of the underlying system.

Ehsan K. Ardestani is a Technical Lead Manager at Meta working on codesign of AI platforms. Ehsan has been working in the industry on architecture of several generations of general purpose and application specific processors, with more recent focus on computing systems, for about a decade.

Deep learning (DL) has been utilized in many user-facing applications, such as augmented reality (AR). Training DL models often require access to multiple GPUs, which can be expensive to set up on-premises. This talk will describe recent work in speeding up cloud-based DL training by utilizing cheap transient servers and devising policies to manage training sessions effectively. Despite the advancement in VM-based distributed training, it can still be daunting for DL practitioners to select the proper amount of resources and configure the training cluster. To alleviate the management complexity, the talk will be about a recent work for fast and cost-effective serverless-based DL training. Lastly, the talk will end with a use case leveraging an edge-based DL model to support mobile AR and challenges in conducting controlled experiments.

Dr. Tian Guo is an Assistant Professor of Computer Science at Worcester Polytechnic Institute (WPI). She earned her Ph.D. degree from the University of Massachusetts Amherst. Her research interests include designing systems mechanisms and policies to handle trade-offs in cost, performance, and efficiency for emerging applications. Specifically, she has worked on projects at the intersection of systems and machine learning, secure machine learning, cloud/edge resource management, big data frameworks, deep learning inference, distributed training, neural architecture search, and AR/VR. Her recent work focuses on improving system support for deep learning and on the practical applications of deep learning in AR/VR. She is the recipient of the National Science Foundation CRII award, the MMSys 2020 best paper award, and the 2022 award for Outstanding Achievement by a Young Alum from The Manning College of Information and Computer Sciences, UMass Amherst.