>Ed-
>
>>I see nothing weird here. Transaction and Buffering are different species.
>
>What I meant by weird was why one would want to implement transactions on unbuffered tables since it would make tables very unavailable. JMO.

Do you mean 'unavailable' in as sense that multiple locks will interrupt other users working with data? I do not think that's critical, because transaction is dynamical, i.e. locks will be applied for short period of time and SET REPROCESS will take care of this. Anyway, locks will be applied to real tables, not to buffers.