>You pretty much have to do it that way in a web app because it's stateless.

Yes, but I had an initial design where the connection was created at the oApp object and shared on every oApp.oProcess object (the one processing each hit), but this was not good (safe) as so expressed by a few experts on the topic here on the Universal Thread. So, it turned out that a connection is then created for each hit and cannot conflict with another. Initially, I created that design thinking that it would make it faster and avoid crowding the queue towards the backend. But, it turned out the other approach, the one which creates a connection for each hit, is safer, as fast and cleaner.

Now, after having taking another look at it. I see that every hit against a backend is enclosed in a Using / End Using syntax. So, basically, it's not a connection per hit but per database request. So, if I take a look at a specific Web site where I can see the stats, this means about five millions SQL Server connections per day. This includes all platforms and for all kinds of hits such as SQLExec(), insert, delete and update. Welcome the connection pooling because otherwise it wouldn't be as fast.