D
dbeavon
Guest
Here are the results of the test on Windows. As George says, some of the _ActServer numbers aren't being reported properly but I think we can rely on "queries received" which is still reported on Windows, as it was on HP-UX. before index added =============== │Messages receive: 0 │ │ Messages sent: 0 │ │ Bytes received: 0 │ │ Bytes sent: 0 │ │Records received: 0 │ │ Records sent: 3953 │ │Queries received: 93 │ │ Time slices: 3991 │ after index added ================= │Messages receive: 0 │ │ Messages sent: 0 │ │ Bytes received: 0 │ │ Bytes sent: 0 │ │Records received: 0 │ │ Records sent: 3953 │ │Queries received: 27 │ │ Time slices: 3953 │ Once again, the round-trips are different by a factor of three. I create and start the database the same way as in HP-UX: prodb mysports sports2000 -newinstance proserve mysports -Mm 32000 -B 3000 -L 50000 -d mdy -T c:\temp -S 5205 My Windows repros always run slightly faster than HP-UX, probably because of modern Intel Xeon CPU. As you can see above, the number of network round-trips (with and without index) was still different by a factor of three, and the overall elapsed times were 30ms (with an index) and 50ms (without). It is interesting that the number of network round-trips was not correlated to the amount of time that elapsed. I decided to do yet another set of tests running the client/server code *local* to the database (but still not self-service). Below are the types of results I was seeing, whether the index was used or not. Notice that when the code is run locally, there is a much higher number of queries that are generated. ┌───────────────────────────────────────┐ │Messages receive: 0 │ │ Messages sent: 0 │ │ Bytes received: 0 │ │ Bytes sent: 0 │ │Records received: 0 │ │ Records sent: 3953 │ │Queries received: 238 │ │ Time slices: 3750 │ └───────────────────────────────────────┘ However when running locally, the elapsed time results were basically the same (ie 30 ms and 50 ms for the "with index" and "without index" scenarios.) I guess that indicates that the latency of the network itself is negligible and the database server is solely responsible for performance issues. When running my tests in shared memory, the elapse time results go all the way down to 15 ms and 25 ms.
Continue reading...
Continue reading...