Subject R: [firebird-support] How to improve Firebird 2.5.3 Disk I/O on Windows server 2012 R2
Author Costantino Molinari

To Karol:

            Original page size was 16K, than I changed page size to 4K just as  my NTFS cluster size and time incredibly doubled !!

            Really confused, but maybe it has to do with the 64K stripe size of RAID 10….


To Set:

TERM SET^ suggestion worked fine.


Time of execution is 11.0 seconds average (tried 3 different executions, every time with empty table)


Don’t know what is the difference between my system (2 x 12Core Xeon Proc./32GB DDR3 ram/6x1.2TB 15K SAS RAID 10/1GB Controller Cache) and yours (16.5 secs), but mine took less time as expected.

Maybe not too less as I have whised, but now I’m going to change some settings in firebird.conf and retry your simple but very useful test.


Thanks a lot


I’ll let everyone know if I have news





Da: []
Inviato: lunedì 29 settembre 2014 13:48
Oggetto: Re: [firebird-support] How to improve Firebird 2.5.3 Disk I/O on Windows server 2012 R2



>Obviously the logic of my application is not the best, but it needs to pass

>through a special interface, that transforms classic cobol read/write
>routines, in sql statements. This brings to the need to have a single
>commit after a single insert command of a single record. I know this is not
>good, but knowing this and knowing that I cannot change this, I only want to
>know if I can get the best for my Firebird Installation of 2.5.3 SuperServer
>onto Windows Server 2012 R2 x64.

OK, commit after every single insert will slow things down considerably and may be the reason for your time trouble (it is similar to having trouble with a car being slower than a bike, and the reason being that you have an additional requirement to start and stop every 5 meters). I'm not certain (because I've never had a similar requirement and know very little about CommitRetaining), but maybe changing from Commit to CommitRetaining might make it possible for a prepared statement to survive the CommitRetaining (?) and then only do a proper Commit occationally. It could at least be worth a try if feasible.

>I'm trying several settings, some others have suggested in this thread.
>About your test (thanks for that), I have created the table, than copied and
>pasted the SQL loop, but I get this error:
>Preparing statement: execute block returns (i integer) as
>declare variable i2 integer
>Error: *** IBPP::SQLException ***
>Context: Statement::Prepare( execute block returns (i integer) as
>declare variable i2 integer )
>Unexpected end of command - line 2, column 21

This is probably due to IBPP believing the ; to be the end of the statement. Try to 'SET TERM ^^ ;' before execute block, change the final end of the execute block to end^^ and then do 'SET TERM ; ^^' at the end.

Sorry for not being able to help more,