One folder, streamed from several servers at once

You give the SpliceFS client one host name. It logs in there, gets the list of every server behind that name, and from then on asks all of them. It fetches the list again every minute, so servers can join and leave.

Finding files. When you open a folder, the client asks every server what it has there and shows you the combined answer. A server that does not have the folder says so, which costs almost nothing. The client asks it again the next time, because the folder may turn up there later.
Reading files. A file is read only from the servers that said they have it. Each of them sends different blocks at the same time, so their speeds add up. The client puts the blocks in order and keeps them in a cache on local disk.
Slow and failing servers. The client measures every server as it reads. A slow server gets less work, and only blocks that are not needed yet. A server that fails is rested, for one second at first and for at most thirty, and then tried again, so it carries its share again as soon as it recovers.

See it in pictures.

Status: alpha, step one. Read-only is the first step toward a complete distributed filesystem. The first 0.x versions are published as packages. It has not run in production. See how it compares with mature systems today.

Native packages

Debian12, 13deb, apt repository
Ubuntu22.04, 24.04, 26.04deb, apt repository
Red Hat Enterprise Linux
Rocky Linux AlmaLinux
8, 9, 10rpm repository, tested on Rocky Linux
Amazon Linux2023rpm repository
openSUSE Leap16.0rpm repository
Alpine Linux3.22, 3.23, 3.24apk repository
FreeBSD14, 15pkg, server and client. Built and tested, not published yet.
macOS (Apple silicon)Homebrew tapToday the formulas build from source.