Subject Re: [IB-Architect] Super-transactions and Incremental Backup
Author Doug Chamberlin
At 6/21/00 09:22 PM (Wednesday), Jason Wharton wrote:
>In my mind I don't see where deleted and limbo records would be a problem.
>They just seem to fit in with all the rest of the stuff.

The big problem with deleted records is that there is no recording of the
fact they were deleted. When you apply an incremental or differential
stream delete actions have to be included in order for the results to match
the original results. Also, just like inserts and updates these deletes
have to be serial. Currently I don't think there is anything in the engine,
or more properly in the modifications we have been discussing, which makes
a lasting note of delete actions, especially if garbage collection happens
on record versions after the freeze point.

My only idea is to invent a "deleted" record fragment or type and leave
that mixed into the record versions. Ugly for sure so I have not seriously
suggested it.

>[snipped text]
>Is the paging architecture still valid in todays OS and hardware
>environments? To me at least, I wonder if perhaps operating systems aren't
>advanced enough such that the whole paging scheme can just be left up to the

I think that the pages referred to within the IB architecture are merely
buffers for working with record images, both compressed and uncompressed.
No matter what OS facilities you depend on for I/O you will always have to
allocate buffers as workspaces so I don't think this can be improved
significantly. Also, the architectural layering is valuable in and of itself.