Parallel Data Laboratory Summer Talk Series - Ioan Stefanovici

— 1:00pm

Location:
Virtual Presentation - ET - Remote Access - Zoom

Speaker:
IOAN STEFANOVICI , Principal ResearcherFuture AI Infrastructure TeamMicrosoft Research Cambridge
https://www.microsoft.com/en-us/research/people/iostefan/

A Tale of Two Extremes: Storing Humanity's Knowledge in Glass, and AI's Data in New Memory

This talk will address two important and timely problems at opposite ends of the data temperature spectrum in today's hyperscaler environments: archiving humanity's knowledge for eternity in glass and storing AI's data in efficient memory.
The first part will focus on Project Silica, which is the first cloud storage system for archival data underpinned by quartz glass, an extremely resilient media that allows data to be left in place indefinitely, thereby eliminating the inefficient, wasteful, and costly migrations of data required with today's magnetic storage technologies. The hardware and software of Silica have been co-designed and co-optimized from the media up to the service level with sustainability as a primary objective. This design follows a cloud-first, data-driven methodology underpinned by principles derived from analysing the archival workload of a large public cloud service. Silica can support a wide range of archival storage workloads and ushers in a new era of sustainable, cost-effective storage.

The second part will introduce Managed-Retention Memory: a new class of memory for the AI era. AI clusters today are the largest uses of High Bandwidth Memory (HBM). However, HBM is suboptimal for AI workloads for several reasons; it is overprovisioned on write performance, but under provisioned on density and read bandwidth, and also has significant energy per bit overheads. It is also expensive, with lower yield than DRAM due to manufacturing complexity. We propose a new class of memory: Managed-Retention Memory (MRM), which is optimized to store key data structures for AI inference workloads. We believe that MRM may finally provide a path to viability for technologies that were originally proposed and optimised to support Storage Class Memory (SCM). These technologies traditionally offered long-term persistence (10+ years) but provided poor IO performance and/or endurance. MRM makes different trade-offs, and by understanding the workload IO patterns, MRM foregoes long-term data retention and write performance for better potential performance on the metrics that are important for these workloads.

—

Ioan Stefanovici is a Principal Researcher in the Future AI Infrastructure team at Microsoft Research Cambridge. His research focuses on inventing the infrastructure for AI systems and he is particularly interested in novel, disruptive memory and storage technologies. Previously, he spent 8+ years working on Project Silica, which developed the first-ever archival storage technology designed and built from the ground up for the cloud, by using femtosecond lasers to store data in glass, and polarization microscopy + ML to read it back. He received his PhD in Computer Science from the University of Toronto in 2016, and was also a Research Fellow at Corpus Christi College, at the University of Cambridge (2016-2018). Ioan lives in London (UK) with his wife, their grumpy 13-year old cat, and 2 ever-curious and playful Maine Coons.

Zoom Participation.  See announcement. 

For More Information:
karenl@andrew.cmu.edu

Add event to Google
Add event to iCal