Hi Nadya,

>Actually, I've realized, that these tests are not quite accurate (since I was running them from the first to the last), so I re-write the testperfomance program a little bit. I'm running these tests from my local machine, but all tables are on the server. Right now I'm the only one (I assume), who is working on this server.

If you're the only one working on those tables, the server likely will aplly opportunistic locking, which means that it opens the tables exclusively for you (until another users wants to open these files) which means that the whole table can be cached on your local machine. This dramaticly will speed up queries, since all the data is already on your machine.

To have more reliable results, someone else should open these tables on another workstation. The server will ask your workstation to write its changes to the server and your workstation will clear its buffers. From that point each query will request the data directly from the server, (which means more network traffic). In this way you're simulating the application in a true working environment more closely. Benchmarking results are more reliable this way.

Well try to benchmark the query with SET DELETE ON/OFF and with and without the DELETED() Tag. You'll likely see the negative impact of the DELETED() tag.

Walter,