PDL Talk Series
August 27, 2026
TIME: 12:00 noon to approximately 1:00 pm EDT
PLACE: Virtual - a zoom link will be emailed closer to the seminar
Aaron Ogus, Microsoft
Building Distributed Systems at ZettaScale: Moore's Law Meets Murphy's Law
Moore's Law delivered a doubling of computing capability roughly every two years. GPUs added another 100–1000× per processing element, and 100,000-GPU clusters promise another five orders of magnitude on top. Taken together, we have compressed something like fifty years of Moore's Law into the last 18 years, as a one-time step change. However, running cloud systems at this scale leads to situations you probably didn't learn about in CS class. Enter Murphy's Law: As distributed systems scale, operational behavior gets stranger. Some of it could have been anticipated. Much of it could not. Azure Storage was on the leading edge of building cloud scale storage systems, and many of the lessons of running at Peta-Scale, Exa-Scale, and now Zetta-Scale we learned by painful experience. This talk tells the stories behind some of those experiences, and will help the audience understand considerations for the design and operation of cloud scale systems, and cloud scale storage systems. In the next few years we can expect single jobs running on well beyond 100,000 processing elements, and every new scale point will bring complications nobody saw coming. I hope that those listening will come away caught up on some of the last 18 years hard-won scaling lessons.
BIO: Aaron Ogus is a Technical Fellow in Microsoft Azure Storage, where he sets architectural direction across Azure Storage and Compute and works with a large team on the co-design of hardware and software for storage, compute, and networking at hyperscale. He has worked on Azure Storage since its inception in 2008, driving innovations that reshaped the industry: the first datacenter-class SSDs, adoption of commodity switching silicon and Layer-3/BGP datacenter fabrics, RDMA-over-Ethernet at regional scale, DPU adoption. Aaron helped bring Tape as the Archive media of choice to the new generation of CSPs, and has helped drive archival research including Project Silica. Aaron joined Microsoft in 1991 and contributed to Windows 95 networking and file services, DirectX, Xbox, and Forza Motorsport before moving to the cloud. He holds a BASc in Systems Design Engineering and Computer Engineering from the University of Waterloo, has over 34 patents, and co-authored the USENIX ATC 2012 Best Paper on erasure coding in Windows Azure Storage. Outside work, Aaron was an SCCA regional Formula Mazda champion and a World Poker Tour televised final-table runner-up.
CONTACTS
, PDL Co-Director
RMCIC 2208
, PDL Co-Director
(412) 268-3064
GHC 9109
Executive Director, Parallel Data Lab
(412) 268-5485
PDL Administrative Manager
(412) 268-6716