All,
I think I may be onto something here. As I was discussing with Frank
Radice, I felt that there were indications that W2k Server was caching
writes to disk far too long. I kept looking for where read/write caching
is controlled and could not find it. Finally, I dug deeply into my RAID
controller's properties and found the control there. I have disabled write
caching for now. This may well cause some performance penalties on writes,
but I hope it will overcome other issues.
What got me to thinking about the caching was evidence such as User A
making a change to the data that it would take several minutes for User B
to see. I know this is not an R:Base problem, but if A's writes are being
cached, it will take B longer to see the change. Of course, B's reads are
also being cached by the controller, so the read cache would take even
longer to be updated with A's write. One must trust that the controller
will handle the refresh of the read cache given the now-immediate write.
Frank,
I'd suggest exploring this issue on your server as an explanation for your
new row issue.
Emmitt
