Deconstructing Lustre
by Phil Schwan
The Lustre cluster file system provides a POSIX-compliant file system API for clusters of almost any size.
It’s designed to scale along all axes: given sufficient hardware, it can offer gigabytes-per- second access to multi-petabyte storage arrays. Most of the world’s largest supercomputers use Lustre to access their storage, including the largest clusters in North America, Europe, and Asia.
This makes Lustre relatively unique, not just by file system standards, but also in what it requires from its operating system environment.
In this hour we’ll tour the highlights of Lustre’s design, including:
- how Lustre differs from most other cluster file systems
- locking and concurrency
- a challenging recovery problem or two
- Linux kernel integration and associated nightmares
Refreshments will be provided after the seminar, thanks to iVEC.
About Phil
Phil is a visiting iVEC Industry & Government Fellow, consulting in the areas of open source business, software engineering process, and storage for high-performance computing.
He’s been actively involved in open source and open development since 1992, contributing to a variety of technologies including enterprise desktop, Linux, privacy and security, network file systems, and consumer internet.
He co-founded CFS, co-developed its award-winning open-source Lustre file system, and oversaw the company’s first four profitable years as Chief Executive. Lustre is used on most of the world's largest Linux supercomputers, providing high-speed, POSIX-compliant storage for governments, universities, and Fortune 100 companies worldwide.
At the Mozilla Corporation he directed an overhaul of localization for Firefox 2.0, shipping the web browser simultaneously in 23 languages for the first time. Approximately 20% of the world’s internet population uses Firefox as their primary browser, including a growing proportion in languages other than English.
Previously, Phil executed the open-source strategy for zero-knowledge systems, one of the most widely-discussed providers of privacy and security software and services. He also helped develop a toolkit for private electronic cash and one of the first commercial anonymous IP networks.