Re: Memory file systems

[Hubert Feyrer <hubert%feyrer.de@localhost>]

On Tue, 9 Feb 2010, Matthew Mondor wrote:
> A main difference if I understand, is that mfs uses a backend closer to
> a disk file system and can also be tied to kernel images using
> mdsetimage.  mfs also has size restrictions defined at configuration
> time.

I don't think those three statements are true:

 * mfs uses (well) memory as storage, and if that runs short, it uses
   the system's virtual memory system as backing store, just like tmpfs.
   The VM system uses a swap partition in both cases, and can be also
   bound to a swap file, but that doesn't depend on tmpfs or mfs.
 * mdsetimage has nothing to do with mfs. Read mdsetimage(8) and md(4).
 * you can restrict the size of mfs, but that's done at mount time, not
   system config(8) time.


> tmpfs on the other hand might theoretically grow in size more
> gracefully and provide enhanced performance, but I honestly didn't try
> tmpfs much myself...

The important point is that tmpfs can *shrink* again, and free ressources 
previously allocated. Which mfs can't do.

Why does mfs exist? Because it was there first (for quite a long time!), 
and noone bothered to throw it out yet.

There may be a number of corner cases where mfs can do things that tmpfs 
can't do (yet?), but I don't know them off-hands. See the mount_tmpfs(8) 
and mount_mfs(8) manpage's BUGS sections for some of them.


 - Hubert