Here are some quick numbers from comparing the original code to the
patched code. Both the StringClumplet and the notify patches are
included in the second set of numbers. Note that this is not a
scientific test because I don't have an isolated maching.

Test Description: AutoCommit=Off - 10,000 inserts. Each insert has its
own commit. 1 select of the 10,000 records. Each record has 10 short
string fields.

Before:
Insert Time: 47,723
Select Time: 2,552


After:
Insert Time: 46,031
Select Time: 1,821

The biggest improvement was in the StringClumplet - setString and
getString functions are much faster and they are used frequently so even
a small increase should speed up most programs.

Does anyone else have a test they could run on the new code to compare
with previous results?


-----Original Message-----
From: David Jencks [mailto:davidjencks@...]
Sent: Thursday, October 17, 2002 5:18 PM
To: Firebird-Java@yahoogroups.com
Subject: Re: [Firebird-Java] Driver Speed Tests - Patch

This is definitely an improvement. I think the question is whether the
data is ever updated after the string is created. I think it can be if
you
clone a default cri and then modify it.

I wrote a slightly different version with a StringClumplet class that
never
needs to create a string, and committed it. Could you see how the
performance compares to what you came up with?

Many thanks for finding this bottleneck and figuring out what to do
about
it!

david jencks