Chapter 4

A network filesystem's whole pitch is that a remote store looks like a local one: the same open, read, write, and rename, against bytes that live on someone else's disks. The abstraction is genuinely useful and genuinely leaky, and the leaks are where production incidents come from—a POSIX call that no longer means quite what it did, latency that turns a working access pattern into a pathological one, and a server that can disappear mid-write.

The chapter builds up in three movements, echoing the shape of Filesystem Fundamentals. First the landscape: what distributed and network filesystems are, how NFS, Lustre, GPFS, and WekaFS differ in architecture and failure behavior, and where object stores stop being filesystems at all. Then a deep dive into NFS close-to-open consistency, the rule most often assumed rather than understood. With the cast assembled, we compare along the same three dimensions— synchronization guarantees, resiliency guarantees, and performance characteristics—and close with the observability story for a filesystem you can only reach over the wire.

The tools themselves live in The Systems Toolkit; here we apply them. By the end you'll know which guarantees a given mount actually gives you, which ones your code was quietly relying on, and how to tell the difference while an incident is in progress.

# Chapter Sections