From: David Wood (jbevren_at_starbase.globalpc.net)
Date: 2008-04-16 02:50:24
On Tue, 15 Apr 2008 ruud.baltissen@apg.nl wrote:
> Hallo Jim,
>
>
> > But, I like Craig's idea in another context. Use u0>h* to
> > move to different virtual "heads" in the partition.
>
> 1) The problem is that we HAVE to write 512 bytes when writing a sector.
> So when writing a sector of 356 bytes, we first have to read the whole
> 512 bytes sector (but we save only the 256 bytes of the other sector)
> and then write both sectors back to the disk. The problem is the memory
> for the "hidden" sector. There are programs, database PRGs most of the
> time, that eat up all the buffers, thus leaving us with no buffer to
> temporary saving that extra sector.
> 2) You cannot copy files from one "head" to the other "head" in a normal
> way.
>
> Then I rather prefer a FS that knows about this extra sector and will
> reserve memory for it as needed. And running the risk that some PRGs
> find themself with less buffers then expected.
I'd mentioned this before, and figure it may need to be mentioned again. A
given 'aware' FS can have a scratch sector. This sector, being 512 bytes
and at a fixed location relative to the beginning of the filesystem, can be
used to save two buffers that are otherwise in-use long enough to work with
(de)blocking a 512byte sector.
IE: (psuedocode follows)
*Save buffers 0/1 to scratch sector
*read sector to be modified
*modify sector
*save sector to be modified
*read buffers 0/1 from scratch sector
*resume
While it would indeed be slower, it would only affect writes. Most disk
operations are reads, so the impact imho would be minimal. If some of us
are concerned about doing this with flash media, it's a valid concern.
however, any decent flash device will have proper sector wear leveling to
the extent that premature failure would be a nonissue imho.
Message was sent through the cbm-hackers mailing list
Archive generated by hypermail pre-2.1.8.