Hello Clark,
Monday, July 05, 1999 you wrote:
C> In my understanding, pointer based approaches like you
C> are recommending have been implemented in several prototype
C> objected oriented databases. They have been shown to be
C> orders of magnitude slower than set oriented techniques,thus
C> many OO databases are implemented as wrappers over
C> relational systems!
I can't guess where you got such information. Contrarily,
I know at least one (commercial) network database server which
orders of magnitude faster than ANY SQL server. It simply no
match to them. That experience is exactly what made me write
to Postgres mailing list. As I wrote (maybe to hackers' list)
pointer lookup takes ideally three CPU commands - read,
multiply, lookup, whereas index scan takes dozens of them and
puts a strain on optimizer's intellectual abilities, and
as we have seen it can hardly choose the optimum way of
performing a join. In pointer-field case optimizer can be quite
dumb, because there is only one way to perform a query.
Best regards, Leon
