What is it about?

Existing storage technologies do not provide a sustainable and cost-effective way of storing data for the long term, as their media lacks the longevity and durability required. This paper presents Silica: the first storage system designed and co-optimized from the media up to the service level for sustainable cloud archival storage. The system is underpinned by quartz glass, an extremely durable and resilient media that allows data to be left in situ indefinitely, with data lifetimes of thousands of years, and this paper focuses on all the computer systems aspects of building a cloud-scale storage system around the core Silica Write and Read technology.

Featured Image

Why is it important?

Our system ushers in a new era of sustainable, and cost-effective storage for long-lived archival data. The media offers unprecedented durability unmatched by any incumbent storage technology. The storage system we've built around the core technology is the first of its kind, incorporating insights derived from studying real archival workloads, and taking a completely clean-slate design approach to building the best possible cloud-scale archival service. The paper does not only talk about a singular design point, bur rather discusses and quantifies many trade-offs in a broad design space of how archival storage systems should be built.

Perspectives

I'm extremely proud of this work, as it's a collective and ongoing effort of more than 150+ person-years spanning ambitious research in computer systems, machine learning, laser-processing physics, free-space optics & microscopy, chemistry, electrical & mechanical engineering, industrial design, and more! Designing and prototyping Silica has really allowed us to enable humanity's knowledge to be preserved in a sustainable and cost-effective way for future generations by re-thinking how cloud storage systems are built.

Ioan Stefanovici
Microsoft Corp

Read the Original

This page is a summary of: Project Silica: Towards Sustainable Cloud Archival Storage in Glass, October 2023, ACM (Association for Computing Machinery),
DOI: 10.1145/3600006.3613208.
You can read the full text:

Read

Contributors

The following have contributed to this page