home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.cs.arizona.edu
/
ftp.cs.arizona.edu.tar
/
ftp.cs.arizona.edu
/
tsql
/
doc
/
tsql.mail
/
000039_atuzhili@squar….stern.nyu.edu _Tue Mar 16 14:40:27 1993.msg
< prev
next >
Wrap
Internet Message Format
|
1996-01-31
|
4KB
Received: from SQUARE1.STERN.NYU.EDU by optima.cs.arizona.edu (5.65c/15) via SMTP
id AA16071; Tue, 16 Mar 1993 12:38:16 MST
Received: by square1.stern.nyu.edu (4.1/1.34)
id AA04330; Tue, 16 Mar 93 14:40:27 EST
Date: Tue, 16 Mar 93 14:40:27 EST
From: atuzhili@square1.stern.nyu.edu (Alexander Tuzhilin)
Message-Id: <9303161940.AA04330@square1.stern.nyu.edu>
To: tsql@cs.arizona.edu
Subject: Benchmark initiative
We would like to make two comments on the proposal to develop a
comprehensive set of natural language queries as a test of "goodness"
of various query languages and algebras.
First, we feel that a certain classification of various queries has to
be established before the set of queries is proposed. As an *initial*
suggestion, we can classify temporal queries as follows.
| HISTORICAL | ROLLBACK | BITEMPORAL
--------------------------------------------------------
UNGROUPED | | |
| | |
GROUPED | | |
| | |
TEMP. AGGREGATES | | |
| | |
SCHEMA VERSIONING| | |
| | |
OTHER FEATURES | | |
--------------------------------------------------------
Then we can place one or several queries in each cell of this matrix.
We want to stress that this is only an *initial* proposal to classify
various temporal queries which should be elaborated further. Perhaps,
the two-dimensional matrix is not sufficient. Perhaps, this matrix
should be reorganized as a three-dimensional cube
(historical/rollback/bitemporal is one dimension, grouped/ungrouped
another, and temporal aggregates and schema versioning the third).
Perhaps, it can be an n-dimensional cube.
The second comment is that the development of the benchmark should not
be a substitute for a rigorous *theoretical* study of expressive
powers of various temporal query languages and algebras. It is not
entirely clear what the goal of the benchmark is. What seems
necessary is a kind of typography of temporal data models as suggested
by the above table and discussion. For example, one data model can be
grouped bitemporal, another be ungrouped historical with aggregates.
Perhaps, the benchmark can be useful in developing such a typography.
Each type of data model should support a class of queries the model
embodies and should have its own standard of completeness. We believe
that this standard should be developed in the terms of an appropriate
logic (as in the classical relational case) rather than trying to
determine expressive power "by consensus" (we would not want to say
that one language is more expressive than another if it can express
95% of the benchmark queries and the other one only 87%).
We addressed this issue in our paper "On Completeness of Historical
Relational Query Languages" (Working Paper IS-91-41 from Stern School
of Business, NYU; also to appear in TODS) where we proposed a measure
of completeness of various query languages. We introduced and
disussed the semantic difference between grouped and ungrouped data
models and explained the difference between queries on grouped and
ungrouped temporal data models. For the ungrouped temporal data
models (e.g. TQuel), we proposed to use either temporal logic, or
first-order logic with explicit references to time, or the temporal
algebra TA (proposed in the paper "A Temporal Relational Algebra as a
Basis for Temporal Relational Completeness" (Tuzhilin & Clifford,
VLDB'90)), all of which have the same expressive power for the
discrete linear models of time, as a basis for the ungrouped temporal
completeness. For the grouped models (e.g. Gadia, Tansel, HRDM), we
proposed to use the calculus L_h that we defined in that paper. We
also examined completeness of various temporal query languages and
algebras proposed in the literature.
We invite comments and discussion of these issues from the temporal
database community.
Jim Clifford
Al Croker
Alex Tuzhilin