Bill Studenmund wrote:

> On Fri, 26 Mar 1999, Bill Sommerfeld wrote:
> 
> > > Can you think of anywhere else we can store the data?
> > 
> > 5) in another file in the lower-level filesystem, possibly indexed by
> > inode number.  This is akin to the LFS "ifile", as well as how AFS
> > stores some of its metadata.
> 
> Hmmm. I like this. I found out from Jason why this idea was nixed (people
> who aren't here anymore thought it would be too slow). But we'll look at
> it.

If the payload is small enough, one idea that comes to mind is storing
the payload in the extra inode, similar to how short symlink names are
stored in the inode.  A new file type could be created, and the changes
to dump/restore would be minimal (except for inode number re-matching
mentioned in another message).  Inode number re-matching could me
made easier by adding a new call (or mode to open()) which allocates
consecutive inodes for you, and sets the inode-link field in the first
inode.  If the inodes are consecutive, you also don't have to worry
about an extra seek/read to get the payload.

Just random ideas...

Simon.