Contacts
IBSurgeon

INSERT/UPDATE/DELETE test

A single-threaded SQL benchmark that measures INSERT, UPDATE, and DELETE throughput on Firebird 2.5, 3.0, and 4.0. Results depend mainly on disk, CPU, RAM, and Firebird cache size. Use it as a quick hardware check - good hardware typically lands in the upper 30% of published results.

Published results

INSERT/UPDATE/DELETE test results on various disks, Firebird versions, and computers
Graph of collected runs. Full table: insert_update_delete_results.xls.

How to run the test

  1. Copy the SQL script below into a file, for example c:\temp\dml-basic-benchmark.sql (Linux: /tmp/dml-basic-benchmark.sql).
  2. Change the database path in the first create database line. The sample uses Firebird 3 TCP (localhost:e:\inserttest4.fdb). You can switch to XNET (xnet://Disk:\path\database.fdb) or embedded (disk:\path\database.fdb).
  3. Keep the same firebird.conf you use in production. To tune it first, use an optimized configuration.
  4. Run isql (on some Linux packages: isql-fb). The test creates a table with several indices, then performs 1 million INSERT, UPDATE, and DELETE operations.
  5. The database grows to about 3.6 GB - delete it after the test.

isql command

Windows

Command 
isql -i c:\temp\dml-basic-benchmark.sql

Linux

Command 
isql -i /tmp/dml-basic-benchmark.sql

Test SQL script

Copy into a .sql file, then edit the connection string and SYSDBA password if needed.

dml-basic-benchmark.sql 
create database "localhost:e:\inserttest4.fdb" user "SYSDBA" password "masterkey" page_size 16384;

set term ^;
execute block as
begin
– ######################################################
rdb$set_context(‘USER_SESSION’, ‘ROWS_TO_HANDLE’, 1000000);
– ######################################################
execute statement ‘drop sequence g’;
when any do
begin
end
end
^
set term ;^
commit;


create sequence g;


recreate table test(
id int
,grp smallint
,pid int
,dts timestamp
,code_sml varchar(15)
,code_med varchar(150)
,code_lrg varchar(1500)
,code_unq char(16) character set octets
,constraint test_pk primary key(id)
,constraint test_unq unique( code_unq )
);


create index test_pid on test(pid);
create descending index test_dts on test(dts);
create index test_dml on test(code_sml);
create index test_med on test(code_med);
create index test_lrg on test(code_lrg);
commit;


set bail on;
set list on;
set stat on;


set term ^;
execute block returns( inserted_rows int, elap_ms int )
as
declare i int = 0;
declare t timestamp;
begin
inserted_rows = rdb$get_context(‘USER_SESSION’, ‘ROWS_TO_HANDLE’);
t = ’now’;
while ( i < inserted_rows ) do
begin
insert into test(id, grp, pid, dts, code_sml, code_med, code_lrg, code_unq)
values(
gen_id(g,1)
,rand() * 10
,rand() * 1000
,dateadd( rand()*1000000 second to timestamp ‘01.01.2019 00:00:00’ )
,lpad(’’, 15, ‘QWERTY’ )
,lpad(’’, 150, ‘QWERTY’ )
,lpad(’’, 1500, ‘QWERTY’ )
,gen_uuid()
);
i = i + 1;
end
elap_ms = datediff(millisecond from t to cast(’now’ as timestamp));
suspend;
end
^
commit^


execute block returns( updated_rows int, elap_ms int )
as
declare i int = 0;
declare t timestamp;
begin
updated_rows = rdb$get_context(‘USER_SESSION’, ‘ROWS_TO_HANDLE’);
t = ’now’;
while ( i < updated_rows ) do
begin
update test set
grp = rand() * 10
,pid = rand() * 1000
,dts = dateadd( rand()*100000 second to timestamp ‘01.01.2019 00:00:00’  )
,code_sml = lpad(’’, 15, ‘ASDFGH’ )
,code_med = lpad(’’, 150, ‘ASDFGH’ )
,code_lrg = lpad(’’, 1500, ‘ASDFGH’ )
,code_unq = gen_uuid()
where id = :i+1 ;
i = i + 1;
end
elap_ms = datediff(millisecond from t to cast(’now’ as timestamp));
suspend;
end
^


commit^


execute block returns( deleted_rows int, elap_ms int )
as
declare i int = 0;
declare t timestamp;
begin
deleted_rows = rdb$get_context(‘USER_SESSION’, ‘ROWS_TO_HANDLE’);
t = ’now’;
while ( i < deleted_rows ) do
begin
delete from test where id = :i+1 ;
i = i + 1;
end
elap_ms = datediff(millisecond from t to cast(’now’ as timestamp));
suspend;
end
^


set term ;^
commit;