Re: lack of raw disk devices

Matti Aarnio (matti.aarnio@tele.fi)
Tue, 6 Jan 1998 19:00:04 +0200 (EET)


The direct device access (buffered or not) applications have
one benefit which allows them to beat filesystem based file
access in speed. Namely the usual indirected-blockpointer
scheme at EXT2 becomes more and more slow when the file grows.

Therefore we can't seriously suggest obsoleting raw-device
access as long as we can't provide efficient means for
continguous file allocation and access with minimum meta-data
slowdown. Hmm.. I will think about this...

/Matti Aarnio <matti.aarnio@tele.fi>

From: linux kernel account <linker@nightshade.z.ml.org>
> On Tue, 6 Jan 1998, Petrucci, Joseph wrote:
> > Wouldn't this solution be just as difficult to implement as Raw
> > devices???
> [snip poorly quoted stuff about letting the FS determin read buffering on
> a file by file basis]
>
> No, it would be more difficult.. But it would be 'neat' and have very
> good benifits.. A extra flag on the open call to set caching behavior
> would be great.
>
> Raw devices would have comparitivly few benifits. Furthermore, this
> feature would make raw devices absolutly unnessassary (just open the /dev/
> file with the approiate open flags)..
>
> This feature would have great uses for MM apps and such..
>