Roman,

> I have found a problem in your test case. You are executing the test
> in auto-commit mode. Auto-commit mode is not the best one from the
> performance point of view, because driver fetches the complete result
> set locally. See my modification to your code below.

I've repeated the tests in not auto-commit mode. Unfortunately your
suggestion I would say didn't help. Shortly, the effect of turning off
auto-commit mode is in decseasing time spent on establishing the
connection and prepare statement at the expense of time spent on
fetching data from the Result Set. As regarding total times than
the JayBird in not auto-commit mode produced almost the same test results
as in auto-commit on Linux, and on Windows even worse. While IC2.5
kept best results on Linux though not so good on Windows.
By the way in my multi-tier server application I use not auto-commit mode.
Please, find all figures below.

So, I guess the problem is not with my code beacause it complies with
JDBC2 and 3 specifications. The problem is with JayBird however.

Seeing such bad thing, my next questions are:
1. Can JayBird delopers group improve situation? If yes, then when?
2. From whom I can expect support on IC2.5 since I'm going to use
IC2.5 on RH8.0 for a while but met some installation problems.
On Slackware 8.1 I didn't have such problems. Will I find support on
borland.public.interbase.interclient? Are there other discussion groups
on IC?

Thanks,
Sergei Volin.
-------------------------------
NOT AUTO-COMMIT MODE
Interclient 2.5/Windows JayBirdRC3/Windows
time1 time2 time3 time1 time2
time3
10536 0 941 10385 30
1412
10625 90 842 400 20
1452
10565 10 911 10445 20
1382
10395 30 1412

Interclient 2.5/Linux JayBirdRC3/Linux
time1 time2 time3 time1 time2
time3
331 6 284 315 24
18307
332 5 284 316 23
18316
332 6 286 316 23
18343