For what it's worth, David Reed's dissertation was on a
non-transactional distributed directory system. Bernstein and Goodman's
book "proved" that MVCC was serializable, which it most definitely was not.
Bernstein and Goodman also worked on a research project called SDD-1 to
see if it was possible to make a distributed transaction system by
applying an algebra to message queues of replication messages. The
answer was "yes" if you know the characteristics of all transactions in
the system, otherwise "no." But it did demonstrate the very interesting
aspect of playing games with time. (SDD-1 was done at my former
employer, Computer Corporation of America, which I stayed in touch with.)
The two major influences on MVCC were a CODASYL database system from
Prime that used page copy on write so reports wouldn't block updates and
SDD-1's games with time. I didn't run into David Reed's paper until the
academics reworked the MVCC Wikipedia page from disdain to one of their
own invention. David Reed does not claim that he invented transaction
MVCC, just multi-version book-keeping.
On 5/31/2022 3:37 PM, Pól Ua L. via Firebird-devel wrote:
https://en.wikipedia.org/wiki/Multiversion_concurrency_control#History
Szía Attila, és köszi!
Hi Attila, and thanks!
Pól...
Firebird-Devel mailing list, web interface at
https://lists.sourceforge.net/lists/listinfo/firebird-devel
--
Jim Starkey, AmorphousDB, LLC
Firebird-Devel mailing list, web interface at
https://lists.sourceforge.net/lists/listinfo/firebird-devel
