Re: Row locking and serializability

Assuming the SQL 92 definition of 'serializable' to be:

    If there is an interleaved history then the     history can be unwoven to produce two     non-interleaved transactions which (in at     least one order) produce exactly the same     result

then I have to agree that Oracle does not implement serializable transactions simply by its command

    set isolation level serializable;

I am very interested by the description of 'predicate locks' being one of the necessary conditions for serialisable transactions - this answers my 'logical if not physical' question about Ingres locking pages that do not exist when it reads currently existing rows that match a query.

Can I assume also that not only do readers need to block writers, but writers need to block readers ? It tends to suggest, of course, that the only possible implementation of serialisability is one which serialises all transactions ;)

Jonathan Lewis
Yet another Oracle-related web site: www.jlcomp.demon.co.uk Received on Tue Jun 15 1999 - 05:17:38 CDT