Introduction to Hurd translators

debian.org

23 pointsvezzy-fnord4 comments

This concept of translators looks very similar to FUSE (Filesystem in Userspace) in Linux, I wonder how it differs, is it easier to write?

And how about the context switch cost? being used to Linux, I'd think that the execution flow has to jump back and forth between kernel and user space to call the translators, and again and again if you chain them?

FUSE is more general, I think, insofar as it provides a filesystem abstraction for anything you like. Whereas this looks more like a layer that sits above an existing filesystem (like a FUSE overlay).

(Edit: Emphasis on the "I think"!)

FUSE is less general, as it only allows implementation of filesystem drivers. HURD translators should allow implementation of almost arbitrary drivers in userspace (such as device drivers or network stacks).

HURD is a micro-kernel so almost everything that traditionally lives in the kernel on Linux systems would be implemented as a translator in a HURD system (apart from very low-level virtual memory and CPU management).

Neat :) Thanks for the correction

Reply on news.ycombinator.com