On Mon, Jun 07, 1999 at 10:39:35AM -0700, Erik E. Fair wrote:
> I recommend that you look at the PRs filed against this class of problem; I
> did a quick survey and found:
> 
> kern: 495 2978 3459 4084 7668
> bin: 2404 3686
> 
> I specifically recommend that you look at 3790, 3791 and 3792 which are
> Koji Imada's proposed solutions (three different implementations!) to this
> problem, which have unfortunately languished since he died in a motorcycle
> accident later that year.

Yesyes. we looked at them, specifically at Imadas proposals..
The problem is, Imada did the work MUCH before UBC was envisioned. I can test
any of this since two months ago, but before I dropped in any of the solutions,
I wanted to rather ask Chuck Silvers for his opinion, as UBC might impose
additional restrictions.

Note that the proposals by Chuq are essentially:

1) Imadas middle solution (fixed in-cache "block" size, drivers and filesystems
  translate if necessary.

2) Imadas last solution (make the in-cache block size the device sector size)

3) Imadas middles solution with (basically) DEV_BSIZE switched to 1. I don't
   know why he didn't consider it; having looked at the code, I would have
   proposed it to him.

I, too, vote for 3) (if no radical problems are associated with it).

	-is