| Subject | Firebird 2.5 linux udf executing time much slower then FB 1.5 |
|---|---|
| Author | |
| Post date | 2013-10-10T09:25:26Z |
Hello
I test Firebird 2.5. Some of my procedures work
slower then with FB1.5 but mostly in linux.
I discovered that one of the reason was slower
execute of udf .
On windows (local test) the difference between
FB2.5 and 1.5 was acceptable.
Firebird.2.5.2 Classic (Superserver
too) 64 bit (tested with binary and also with compiled source)
buffers 2048, FW off
Linux CentOS release 6.2 (Final), kernel 2.6.32-220.7.1.el6.x86_64, Intel(R) Xeon(R) CPU E31220 @ 3.10GHz
client IbExpert
client IbExpert
I tested it with two servers -second red
hat.
I created very simple udf to test
with one function: int fn_one() { return 1;}
I declared this function as MD_ONE
DECLARE EXTERNAL FUNCTION MD_ONE RETURNS
INTEGER BY VALUE ENTRY_POINT 'fn_one' MODULE_NAME
'udflib';
I created procedure
CREATE PROCEDURE XX_2( S INTEGER, C
INTEGER) RETURNS ( R INTEGER)
AS
declare d integer;
begin
r = s;
while (r < c) do begin
d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one();
d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one();
r = r + 1;
suspend;
end
end
AS
declare d integer;
begin
r = s;
while (r < c) do begin
d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one();
d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one(); d = md_one();
r = r + 1;
suspend;
end
end
and executed it
select count(*) from xx_2(0,
80000)
Average execute time was over 3.5 sec
In FB1.5 it last about 0.2 sec
The more udf functions is executed the diffrence is biger.
I now that loop is big but in practice there are construction like
that
for select from xx do begin
for select from yy do begin
for select from zz do begin
d= md_one;
end
end
end
I need advice - what can I do to speed up execution.
Is it a problem with configuration FB or linux ?
Regards,
Mariusz Duch