What is ....

[Kurt Wall]

On Thu, Feb 09, 2006 at 10:53:55AM -0500, David Bandel took 73 lines to write:
> On 2/9/06, Ben Duncan <bns at meta3.net> wrote:
> > Ok, thanks everyone. The reason I asked is DBM stuff I am
> > copying from, often to refer to bit-mappin' the index (WTFIT?) and
> > using a bit-map for telling what "fields" are indices.
> >
> > I can see the usage from when disk space was a premium (Kinda like the
> > old COMP-3 stuff in Cobol), but as another question:

As I said, it's not about disk consumption (anymore), but access speed
and efficiency, which should always be a concern.

> Hmmm.  Many years ago when I did real programming, I remember breaking
> bytes into nybbles for memory conservation.  Even today the rational
> would be for pointers you might need to use often, this is still the
> best way because no matter how much memory you have, the coveted 0th
> page (where your registers are) still has limited memory and is still
> the fastest access.  I doubt that has changed.  Loading stuff there
> boosts access times drastically for stuff you're needing to access
> constantly.
> 
> I'm sure Kurt could update/fix any errors in the above (it has been
> years as I said).  But the principals still apply even if my
> terminology or use thereof is outdated.

I could expand, but you're spot on. Stack memory is always dear and
as I suggested in another message, data locality is important. Because
the X86 architecture is famously register-starved, cache hits are even
more important on X86 than on, say, SPARC which is register-rich.

> > C: Apply the K.I.S.S. principal ...
> 
> The above reads to me like:  fast, cheap, good:  pick any two.

Yep.

Kurt
-- 
This will be a memorable month -- no matter how hard you try to forget
it.