Re: Reinventing the TransRelational Model?

From: x <>
Date: Thu, 13 Apr 2006 15:16:43 +0300
Message-ID: <e1lfe7$3is$>

"paul c" <> wrote in message news:s0a%f.6055$7a.3904_at_pd7tw1no...

> I too suspect whatever it does is not entirely specified in the patents.
> I vaguely remember something to the effect of the "method" not being
> affected by different implementations that "somebody skilled in the art"
> might think of. Apparently (I read somewhere) the patent lawyers avoid
> words like 'algorithm' because those can't be patented.

You can look at the referenced work in the patents.

> Still from the patents, my limited understanding is that it certainly
> goes into more detail than just describing a column-based store. The
> zig-zag structure/method is certainly unusual but I don't know if it's
> novel. I also don't think that indexing all values is novel nor
> combining an index with links that replace the common row-based impl'ns.

Nothing is new.

> When I looked at trm from the point-of-view of a memory-based 'store',
> as opposed to the disk-based strawman that was critized, I could see no
> way to avoid some pretty heavy updating overhead, unless a lot more
> pointer plumbing was used, plus a fair amount of churning for
> projections on tables with lots of columns. My desire was to minimize
> storage as well as to avoid a garbage collection component. Maybe there
> are clever ways to deal with the updating overhead that are beyond me
> but I concluded that for ten million rows and hundred million values and
> a bunch of btrees, i could get away with less than 5 byte storage
> overhead per value, getting the 'adjacency' effect for free, albeit at
> increased cost to add or delete columns.

There are these things called succint data structures. Knuth describe an interesting algorithm in one of his books. If I remember correctly, it is cited in the patent.

> I also noticed that Mr. Bayer, co-inventor of the btree which was never
> patented, has applied for a patent on his clever new 'all-column' (my
> words) index - i forget the name but it might be 'z+tree'. Where would
> we be today if he'd patented the btree or if Mr. Hoare had patented
> quicksort?

I think he patented some of the algorithms like the tetris sort, not the ideea of z-tree.
I think this kind of trees are used in data clustering.

> The ctree project seemed pretty derivative to me, for example it's
> occurred to lots of people to stick compression routines on top of
> indexes.

Compression is the key for reducing complexity. The ideea is to compress the data and to use data as an index. To me nothing of the above looks new.
Maybe the mix is unique, but I don't know all the work that has been done in all the research projects in the past for implementing the relational model. Maybe these ideeas were not implemented at the time because of the hardware technology available at that time or just overlooked by the ones in charge. Received on Thu Apr 13 2006 - 14:16:43 CEST

Original text of this message