Re: Read-write model

From: Jonathan Leffler <>
Date: Sat, 03 Apr 2004 07:35:48 GMT
Message-ID: <oPtbc.12675$>

Karsten Wutzke wrote:
> I have an important question about scheduling and the underlying
> read-write model:
> Is it right, that the read-write model dictates, that there are no
> redundant read or write operations on the same database object per
> transaction, or better, only one read or write on a specific database
> object per transaction???
> TA1 = r(x) w(y) r(x) w(x) -> the 2nd r(x) would be invalid

There are few (if any) DBMS that would give problems with the sequence shown. It seems rather ridiculous to suggest that a given row may only be read once per transaction. For example, if you select data using two different search criteria on the same table, it is perfectly reasonable for a row to match both criteria.

If TA2 = r(x) w(x) r(x) w(x), then I believe that Oracle used to return the same (non-modified) value for both r(x) operations, which led to some problems with one of the TPPC benchmarks which required the DBMS to read the modified data. However, in the majority of DBMS, the second r(x) would return the value stored by the first w(x).

Do you have a reference for where there is a read-write model that precludes multiple reads?

Jonathan Leffler                   #include <disclaimer.h>
Guardian of DBD::Informix v2003.04 --
Received on Sat Apr 03 2004 - 09:35:48 CEST

Original text of this message