There is one comment I would like to state, on the issue of
a sequential scan beeing faster than an index scan. It is actually
often
true in a singel user system that an index scan is more expensive
than a sequential scan.
As long as we have table level locks this is also true for heavyly
concurrent access.
Here comes the disadvantage:
Once row or page locks will be implemented, the sequential scan
cost should be reconsidered, since then readers will often be
waiting
for updaters, that are actually updating data, that is irrelevant
for the
reader. The average wait time will have to be added to the sequ.
scan
cost.
Andreas
