ngnfs is a new distributed posix file system for Linux that aims to provide low latency and high concurrency by reducing the machinery between persistent data at rest and the network endpoints operating on it.

ngnfs is the result of the author's experience with classic parallel and clustered file systems in production at scale. ngnfs is primarily designed to avoid the coarse coherency regimes and heavy server structures of those systems that introduce latency and concurrency bottlenecks. Instead ngnfs is built around network protocols that provide block-granular concurrency and distributed atomic transactions.

The talk will highlight the problems that large scale systems encounter, the core components of the design that address these challenges, the state of the implementation and initial performance results, and explore the opportunities for collaboration on a project in its early stages.