Re: Row locking and serializability

RE: "the proposition that Oracle has correctly interpreted the SQL92 definition of serializability"

Here is an extract from the paper referenced by Vadim Tropashko and Hal Berenson who have been contributing to this thread. The paper was published in the proceedings of SIGMOD 95 and is called "A Critique of ANSI SQL Isolation Levels". Hal Berenson is one of the authors of the paper, the others being Phil Bernstein of Microsoft, Jim Gray of U.C. Berkeley, Jim Melton of Sybase, Elizabeth O'Neil of UMass/Boston and Patrick O'Neil also of UMass/Boston. The comments made by these researchers lead to the conclusion that the proposition (that Oracle has correctly interpreted the SQL92 definition of serializability) is false.

• begin extract ---

ANSI SQL defines four levels of isolation by the matrix of Table 1. Each isolation level is characterized by the phenomena that a transaction is forbidden to experience (loose or strict interpretations). However the ANSI SQL specifications do not define the SERIALIZABLE isolation level solely in terms of these phenomena. Subclause 4.28, "SQL-transactions", in [ANSI] notes that the SERIALIZABLE isolation level must provide what is "commonly known as fully serializable execution." The prominence of the table compared to this extra proviso leads to a common misconception that disallowing the three phenomena implies serializability.

• end extract ---

Disclaimers: (1) My employer may have opinions very different from mine. (2) My opinions may prove to be significantly incorrect. (3) Oracle itself is the final authority on the capabilities of the Oracle product line.

In article <Yxr83.7714$G01.107914_at_news3.giganews.com>,   "David Cressey" <dcressey_at_valinet.com> wrote:
> I was agreeing with the proposition that Oracle has correctly
interpreted
> the SQL92 definition of serializability, not that Oracle's
implementation
> agrees with what is commonly assumed about serializability.
>
> More imortantly, as Hal Berenson's reply suggests, the SQL92
> interpretation
> is not a good criterion for an implementation that is free from
flaws. I
> would add, that what is commonly assumed about serializability, at
least
> in my working environment, would form an even worse criterion for
> assessing serializability.
>
> Regards,
> Dave Cressey
>
> Michael Ortega-Binderberger wrote in message ...
>
> >I disagree. The fact that as the original mail suggests, you can
> >construct and example where it breaks down, means that Oracles
> >serializability is not what is commonly assumed.
>
>

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't. Received on Sat Jun 12 1999 - 18:42:41 CDT