[Date Prev][Date Next][Thread Prev][Thread Next][Author Index][Date Index][Thread Index]

:zz: Sequence we'll implement Slices

Hi all--

We need slices, i.e. some way to bring in
 files with lots of cells.  No way can we get
 serious about Big Data without slices.
 (I don't trust virtual memory in as anarchically
 fine-grain an environment as ZZ, especially
 when you have to climb over all the cells
 in a row to get to the end.)

So Slices are how we will package up collections
 of cells to swap in bunches (using the conventional
 file mechanism, sigh).

Slice 0 will always be resident, others
 can be explicitly summoned and purged.

My defining paper on this from last July
 describes what I really want.

However, it'll be necessary to work toward that.

So in today's conclave with Andrew, we decided
 the order is--

0.  One-point slices: they are connected just at
 one cell.  This allows them to be co-resident
 and not much more.

1.  Hard-edged slices: while slices may have
 many individual connections, each is contiguous--
 with no hopping and cloning across the border.

First version: an edge of a cell in slice 0 may
 have only one connection-- either to a cell in
 its own slice, or to ONE other cell elsewhere
 in the universe.

2.  Hard-edged slices with preferences from
 outside: cells in other slices may choose a cell
 in cell 0 to connect to, and if two cells in two 
 different slices choose the same cell, the one
 with the lower slice no. wins.

(This introduces the preflet, an expressed preference.)
 Each cell may name no more than one other cell
 in each direction of each dimension.
 Preflets need not be requited.

When a slice arrives, preflets are resolved;
 when a slice leaves, preflets resume their
 Pining state.  (So a preflet is latent/dormant while
 a cell is resident.)

If a cell is moved, its preflet presumably changes.

3.  The preflet is expanded to four modes
? MUST connect to
? WANT to connect to
? MUST be higher/lower than
? WANT to be higher/lower than

This is the final preflet scheme expressed last year.
 What I especially like is that it 
? allows cells from other slices to interpose/
 insinuate betwixt others
? implements locked links
? implements sequential links
? is minimalistic

All best, Ted

Theodor Holm Nelson, Visiting Professor of Environmental Information
 Keio University, Shonan Fujisawa Campus, Fujisawa, Japan
 Home Fax from USA: 011-81-466-46-7368  (If in Japan, 0466-46-7368)
http://www.sfc.keio.ac.jp/~ted/  (Professorial page)
Permanent: Project Xanadu, 3020 Bridgeway #295, Sausalito CA 94965
 Tel. 415/ 331-4422, fax 415/332-0136  
http://www.xanadu.net (see also Professorial page, above)
PERMANENT E-MAIL: ted@xxxxxxxxxx
QOD 99.03.25
"One event is an anomaly. Two is a coincidence.  Three is a trend."  
-- Anon.