zfs set primarycache=none foo/bar

?

Though this will amplify reads as even metadata will need to be fetched from disk, so perhaps "=metadata" may be better.

* https://openzfs.github.io/openzfs-docs/man/7/zfsprops.7.html...

I'm curious what your workflow is that not having any disk caching would have acceptable performance.

A workflow where the person doesn't understand that RAM isn't wasted and it just their utility to show usage is wrong. Imagine being mad at file system cache being stored in RAM.

The problem with ARC in ZFS on Linux is the double caching. Linux already has a page cache. It doesn't need ZFS to provide a second page cache. I want to store things in the Linux page cache once, not once in the page cache and once in ZFS's special-sauce cache.

If ARC is so good, it should be the general Linux page cache algorithm.

I maybe wrong, but I remember (circa 2011) that any access to ZFS on Linux entirely ignored page cache buffers unless you use mmap, so by "default" only binaries and libraries get double cached.

ARC is better than page cache in linux right now. It's not used by linux because:

1) Linus irrational hate towards ZFS (every rant I read shows clearly that he has next to zero knowledge about ZFS)

2) Patents and Licensing

Also, it's a linux's problem that it's doing that, not ZFS - nothing is cached twice on other platforms that run ZFS. Why? See reason #1.

Are you really saying: “design of user land code Y should be in the kernel”?

ZFS has been run just fine on a system with 1GB of ram. Ram issues with ZFS are just FUD.

>If ARC is so good, it should be the general Linux page cache algorithm

Not possible until IBM's patent expires

Well patent 6,996,676 was filed in November 2002, which should mean it's expired now?

I guess there's a few others listed in various places that were filed up to 2006-2008, I'm not sure how important they are.