[OpenAFS-devel] Re: OpenAFS-devel digest, Vol 1 #814 - 2 msgs

[Harald Barth]

> >       File server backend: It would be nice if there would be a
> >       possibility to add different storage backends.
> 
> What, exactly, do you mean by a back end?

No, I'm not talking about the HW, I'm talking how AFS uses the OS on
the fileserver to finally store the files. OpenAFS can today save
files in two ways:

	* A special vice partition with hidden inodes containing extra
	info. Example: Solaris. This kindof misuses the vendor ufs. If
	you do a vendor fsck or turn on logging (Solaris) you are
	toast. Pro: Quite fast. Con: The toast and portability issues.

	* A file tree in a case sensitve file system where what would
	be an inode names above is translated into directories and file
	names. Pro: Protable. Con: Slow meta operations.

Which one is used is chosen compile time on per server basis. These
two could use some company. Possible extensions: Raw partition,
reiserfs (squid)cache API, tape or migrating backends (*). To make
tinkering with different storage alternatives wortwhile, an interface
would be useful. Milko (the Arla server) has such an API behind the
scenes. Unfortunately there hasn't been manpower enough to do anything
with Milko recently.

(*) MRAFS has support for putting the files into (SGI) DMF, it is
probably smarter to interface to an existing file migration solution
than to build your own.

Harald.