>George,
>
>>snip<
>
>>The only place, therefore, where a performance hit
might be even noticable would be when a very large (say, 100,000+ records) were appended into a table at once.
>
>Several years ago, over on the CompuServe forums there was a fellow who worked for CBN as their lead IT manager. They typically added 60000 records in an 10 hour period in a system that was, if I recall correctly, fairly sophisticated relation-wise. He said that they had indeed had some issues with so many workstations trying to add records (ie. lock the header) at the same time that they determined that they should just pre-load the tables the night before. Apparently locking the header to put a token into the next available record was a more easily accomplished than adding the actual record, modifying the header plus all the locking. IOW, it made a difference in his particular case.
>
>Candidly, IMO if you're going to start adding 60,000 records in a day (a 10 hour day means that's 100 per minute) I think both pre-allocating records and using SQL start making a lot more sense.
DD,
Yep! I should clarify my statement a bit, however. It assumed a regular append
without key generation when compared to one using an auto-increment column. Naturally, when compared with with a PK generation routine to do the append, it should be much faster. So there is a performance gain as well.
George
Ubi caritas et amor, deus ibi est