On 11 mai, 01:02, paul c <toledoby...@[EMAIL PROTECTED]
> wrote:
Hi paul,
[Snipped]
> I concluded that one could accomplish the same 'physical' effect if
> every 'column' were indexed.
Yes.
> However, in certain cases, it seemed to me that even if Codd was
> approving of it (or so I remember reading), it denied his goal of
> 'symmetric exploitation', reason being that it depended on projections
> of columns that were 'adjacent' (my term) in the 'zig-zag' organization
> of physical columns, eg., if the zigzag connected column a to b and b to
> c but the query involved only columns a and c, then the result wouldn't
> have what I think of as a 'pleasing' order, unless a sort were invoked.
Tarin failed to deal with solving the mathematical counterpart of zig
zags. He failed to minimize mathematically the number of operations
necessary to reconstitute the tuple. He thought that separating the
layers should de facto be *sufficient* but he failed to see that it
was just *necessary*. I have proved mathematically that a storing
system that imposes *any* kind of ordering to represent a binary
relation necessarily has an equally increasing of operations to
reconstitute tuples as more un-ary domains intersect with domain
defined by the set of all tuples. That fundamental problem defeats in
fact the entire purpose of data independence.
Based on the study of Tarin's work and then mistakes, I have expanded
to try to avoid transrelational model shortcomings by designing a
radically different appoach to how relation can be reconstituted at
run time.
> I realize that a pleasing order is counter to relational dogma but when
> it comes to the mundane domains such as dollars and dates that I've been
> indoctrinated in all my life, I find ordering of those familiar domains
> to be just as powerful, helpful and convenient as pretty much any logic
> I'm aware of. As far as I can tell, Codd's symmetric exploitation could
> be achieved if every permutation of 'columns' were encoded in the
> trans-relational intermediate layer.
Permutation are logical operations that are in fact not efficient as
far as number of operations involved to reconstitute relation domain
elements. Among other things, symmetric exploitation is impossible
when any binary relation is encoded as is on the physical disk.
I have created a method that we can name relation reduction that
allows to minimize operations at elementary level. The method is
based on the study of the relation****p between the cardinality of each
attribute domain element and relation domain tuple. I also designed a
subsequent new physical encoding scheme far more efficient than any
encoding scheme involving a direct encoding. I do not want to
discredit Tarin's work because it is probable that I would have never
thought about that without studying TRM and seeing where it fails but
I can say with certainty that TRM is *still* a direct image system.
The relation reduction does not put in fact put *any* ordering as a
prerequisite onto encoding information onto the physical layer which
brings it in coherence with the logical layer while keeping both
layers separate. The consequences are numerous and express true data
independence:
The larger is the tuple set the more efficient the encoding. I can
actually see sometime file sizes *decrease* when the number of tuples
to be represented increases: the more logical information is encoded
the smaller the physical counterpart of that information. I believe
this is true data independence.
I must give a strong credit to Tarin's work as it was innovative
enough to trigger new questions which led me here.
[Snipped]
|
