Hi ! I am playing with those parameterson FB 3.0 Beta 2 and wish to show the results...
The test is very simple, on the client app I loop 50000 times inserting into a very simple table: Create Table Foo(bar integer), no indices, no constraints, nothing. The app uses a prepared query and commit at every single row (I know it's not efficient, but I am testing just the effects of the above parameters) The environment: Windows 8 6 GB RAM Firebird: 3.0.0.31896 Beta 2 (runing as CS) Page size: 4KB Page Buffers: 256 And get the following results (time in seconds), wich I think is expected: +-----------------------+--------------------------------------+ | | MaxUnflushedWrites | |MaxUnflushedWriteTime |------+--------+------+--------+------+ | | -1 | 100 | 10k | 100k | 1 | +-----------------------+------+--------+------+--------+------+ | -1 | 21 | 62 | 28 | 26 | | +-----------------------+------+--------+------+--------+------+ | 5 | 26 | 62 | 28 | 27 | 1820 | +-----------------------+------+--------+------+--------+------+ | 500 | 24 | 62 | 27 | 27 | | +-----------------------+------+--------+------+--------+------+ | 5000 | 26 | 61 | 26 | 26 | | +-----------------------+------+--------+------+--------+------+ The same test with Forced Writes = ON runs in 61 seconds My doubt is with MaxUnflushedWrites = 1, I expected it to be similar to Forced Wirtes ON, but it runs in 1820 seconds ! (30 times slower). I suppose that even with forced writes ON (and carefull write) a group of pages are written at once, and thus the file system could order the writes to achieve a better performance and with MaxUnflushedWrites = 1 each page are written with out reordering. I hope someone could explain me if that is the case and if not why there is a huge diference. TIA Alexandre Benson Smith THOR Software ------------------------------------------------------------------------------ Firebird-Devel mailing list, web interface at https://lists.sourceforge.net/lists/listinfo/firebird-devel
