Cisco Posts Rust-Written PuzzleFS File-System Driver For Linux

[phoronix]

Phoronix: Cisco Posts Rust-Written PuzzleFS File-System Driver For Linux

PuzzleFS is a next-generation container file-system for Linux with fast image building, direct-mount support, and other container-optimized features being worked on by Cisco engineers. And it's written in Rust...

https://www.phoronix.com/news/Experimental-Rust-PuzzleFS

[lucrus]

Wow, I mean just, wow! That's exactly what the whole world was looking for: a new filesystem for the Linux kernel.

[luben]

Looks similar goals with ComposeFS: https://lwn.net/Articles/922851/ so it may hit similar issues for inclusion in mainline

[schmidtbag]

Wow, I mean just, wow! That's exactly what the whole world was looking for: a new filesystem for the Linux kernel.

Yeah but it's written in Rust so half the users here are going to be like "finally! Now we can throw out every other filesystem!" because if it isn't written in Rust, it might as well not exist!
/s

[lucrus]

Yeah but it's written in Rust so half the users here are going to be like "finally! Now we can throw out every other filesystem!" because if it isn't written in Rust, it might as well not exist!
/s

Some 20 years ago, when I was a more or less novice software developer, I tried writing a NFS server in Java. It seemed like a wonderful idea, because it was backed by a relational MySQL database, where the tables became directories and records became text files named after their primary key value... so a simple file indexer could provide full text search capabilities over the entire database contents... except file indexers are slow as hell on their own even when running on real hardware, let alone a network filesystem emulated through SQL queries running in a JVM of the era... but overall, when I think about that project nowadays, it's still more interesting to me than most filesystems out there. Because you see, it was written in Java...

BTW, the entire thing (file indexer, NFS server, JVM, MySQL and data storage) was assumed to happily run on a single server (actually a common desktop pc) with a single hard disk...

[discordian]

The first thing I'll write after learning rust, will be a testing and benchmarking Plattform. Even if it's just 30 lines of code and doesn't work at it should head the frontline at Phoronix.

[Vistaus]

Looks similar goals with ComposeFS: https://lwn.net/Articles/922851/ so it may hit similar issues for inclusion in mainline

Why isn't Cisco cooperating with that project?

[hsiangkao]

Why isn't Cisco cooperating with that project?

Just technically speaking, I think the question might be "why not make a comparsion with a 5-year exist kernel project EROFS (already have chunk deduplication and compression) and work yet another one into Rust rather than collaborate?"

[cynic]

Some 20 years ago, when I was a more or less novice software developer, I tried writing a NFS server in Java. It seemed like a wonderful idea, because it was backed by a relational MySQL database, where the tables became directories and records became text files named after their primary key value... so a simple file indexer could provide full text search capabilities over the entire database contents... except file indexers are slow as hell on their own even when running on real hardware, let alone a network filesystem emulated through SQL queries running in a JVM of the era... but overall, when I think about that project nowadays, it's still more interesting to me than most filesystems out there. Because you see, it was written in Java...

BTW, the entire thing (file indexer, NFS server, JVM, MySQL and data storage) was assumed to happily run on a single server (actually a common desktop pc) with a single hard disk...

that sounds like an awesome project!
Modern JVMs and PostgreSQL would provide 1000% performance improvement!

did you released your code somewhere?

[EphemeralEft]

I’d like to take this opportunity to announce my new project, koOLfS (pronounced “scowl fees”, and the capitalization is important). It’s a modern CoW file system written in COBOL with the goal of having all features of BTRFS plus the ability to destroy all data if you create a file named “a.out”. If you’re worried about some programs using that name by default, stop using those shitty programs.

koOLfS will of course be incompatible with BTRFS and the first version should be available in late 2025. Contributors are welcome, but I have some very specific coding style requirements that you can read about in the 60 page document on my GitHub.