home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.cs.arizona.edu
/
ftp.cs.arizona.edu.tar
/
ftp.cs.arizona.edu
/
tsql
/
doc
/
tsql.mail
/
000049_rts _Thu Mar 25 09:06:17 1993.msg
< prev
next >
Wrap
Internet Message Format
|
1996-01-31
|
2KB
Received: from boojum.cs.arizona.edu by optima.cs.arizona.edu (5.65c/15) via SMTP
id AA17665; Thu, 25 Mar 1993 09:06:19 MST
Date: Thu, 25 Mar 1993 09:06:17 MST
From: "Rick Snodgrass" <rts>
Message-Id: <199303251606.AA29330@boojum.cs.arizona.edu>
Received: by boojum.cs.arizona.edu; Thu, 25 Mar 1993 09:06:17 MST
To: tsql@cs.arizona.edu
Subject: schema extensions
Ed Robertson makes the cogent argument that it is good design dictates
completing the eventual schema now, rather than starting with a small
schema and incrementally extending it in the future.
I and others have made the point that a larger initial schema is
counter to the goal of comprehensiveness. Specifically, if a larger
schema is adopted, then the number of possible queries grows very
quickly, making it difficult to get an initial version of the
benchmark completed. I feel strongly that we need to be very careful
not to be too ambitious this first time around. It is very important
that we produce *something* soon, that can then be generalized and
extended.
How about the following?
1. We design the schema to include the additional attributes. This
will involve some more discussion to ensure that we have the right
ones, and that we don't have too much "functional redundancy" (i.e.,
it doesn't seem that we need two time-invariant attributes, or two
user-defined attributes, or two multi-valued attributes). Perhaps Ed
or Patrick could produce a strawman proposal by augmenting the latex
prose that Christian sent out previously.
2. In the first draft, we present the schema in two parts. The first
will contain the entire schema. The second part, where we discuss
restrictions applied in this first draft, we identify the subset of
attributes that queries *in this first draft* could mention. This is
very similar to restricting ourselves to not include updates or
aggregates. In future versions, the set of attributes allowed in
queries could grow, until all are allowed to be used, at which time
the second part, on the restrictions, could simply be removed from the
document.
Could the others who have proposed benchmarks in the past comment on
how you think we ought to proceed?