Performance of 4.0

[Gabor Boros]

Hi All,

With an internal testing system I see the below numbers (same database
structure, same data, same tests). Thousands of tests (today's numbers,
Test:19288, Checks:256340) executed with 2.5, 3.0 and 4.0 SuperServer on
three different machines (Win64).

Version changing(altering, etc.) on the test databases:
2.5 - 12:42
3.0 - 15:18
4.0 - 21:00

Fill database with test data:
2.5 - 01:47
3.0 - 01:40
4.0 - 02:11

Run all tests:
2.5 - 31:40
3.0 - 36:06
4.0 - 43:56

I try to find what and why slow with 4.0 (and 3.0). Any idea how?
Cannot show the real system, so just a dumb example for demonstration
(restored same 2.5 backup with 2.5 and 4.0).

execute block as
declare variable dependent_name type of column
rdb$dependencies.rdb$dependent_name;
declare variable i integer=0;
begin
for select distinct d.rdb$dependent_name from rdb$dependencies d
where d.rdb$dependent_type in (1,2,5)
into dependent_name
do i=i+1;
end

2.5
Elapsed time = 0.11 sec
Fetches = 286262

4.0
Elapsed time = 0.213 sec
Fetches = 148899

execute block as
declare variable dependent_name type of column
rdb$dependencies.rdb$dependent_name;
declare variable i integer=0;
begin
for select d.rdb$dependent_name from rdb$dependencies d where
d.rdb$dependent_type in (1,2,5)
into dependent_name
do i=i+1;
end

2.5
Elapsed time= 0.08 sec
Fetches = 286262

4.0
Elapsed time = 0.150 sec
Fetches = 148899

select count(1) from rdb$dependencies d where d.rdb$dependent_type in
(1,2,5)

2.5
136244

4.0
138729

10ms vs. 20ms. Not a big difference, but if executed thousand times...

Executed a random test with Firebirds's trace:
update or insert into ... (same table and executed many times while the
test running, not test the update or insert into just modify the test
data while testing)
2.5 - 11 ms, 5 ms, 5 ms, 5 ms
4.0 - 17 ms, 10 ms, 10 ms, 10 ms

I am not happy with performance of 3.0 and very unhappy with 4.0.
No performance testing before releases?

Gabor

[Gabor Boros]

2021.06.23. 14:15 keltezéssel, Gabor Boros írta:
> With an internal testing system I see the below numbers (same database
> structure, same data, same tests). Thousands of tests (today's numbers,
> Test:19288, Checks:256340) executed with 2.5, 3.0 and 4.0 SuperServer on
> three different machines (Win64).

Executed a very small subset of the tests (three times and report the
best result per case), 2.5.9 vs. 4.0.1.2520.

2.5 SS - 1 min 36 sec 199 ms
4.0 SS - 2 min 2 sec 910 ms
4.0 CS - 2 min 9 sec 939 ms
4.0 SC - 2 min 9 sec 598 ms

Gabor