Subject Re: [firebird-support] Firebird 2.5 on Ubuntu 10.10 Maverick
Author Werner F. Bruhin
Philippe,

Thanks for the links, as you see from below the benchmark stuff in the
other post helped identify the issue, but not found the solution to it yet.

On 05/15/2011 01:29 AM, Philippe Makowski wrote:
> don't use Ubuntu ;)
> kidding (more or less, because it is not my Linux distro first choice)
Yours is? I have to guess Mandriva, see below.
> you can read this :
> http://www.ibphoenix.com/resources/documents/search/doc_26
> and this :
> http://www.ibphoenix.fr/IMG/pdf/fb-linux.pdf
>
> File system and scheduler can have influence on performance
- maybe mention that you assume "Mandriva" as I naively tried "urpmi
libstdc++5" which is obviously not found:-) .

I made changes regarding open files and scheduler, i.e.:
added this to limits.conf:
firebird soft nofile 4096
firebird hard nofile 10240

and made the change in grub to use "elevator=deadline"

Neither of them made any difference.

Did the iibench.py test on Ubuntu which gives:
#rows #seconds #total_seconds cum_ips table_size last_ips #queries
cum_qps last_qps
10000 25.5 25.5 392.0 10000 392.0 2640 103.5 103.5
20000 60.7 86.3 231.9 20000 164.6 7868 91.2 86.1
30000 76.5 162.7 184.4 20000 130.8 13656 83.9 75.7
40000 76.4 239.1 167.3 20000 130.8 19304 80.7 73.9
50000 73.8 312.9 159.8 20000 135.5 24756 79.1 73.9
Done

I can't make it work in the VM, it throws an exception that it can't
pickle something. Tried to run it from Ubuntu against the FB server
running in the VM but can not get it to work either.

Found it, the FireWall on the VM blocked things, here the result against
the the db in the VM:

#rows #seconds #total_seconds cum_ips table_size last_ips #queries
cum_qps last_qps
10000 13.8 13.8 724.1 10000 724.1 1652 119.6 119.6
20000 27.3 41.1 486.5 20000 366.3 4280 104.1 96.3
30000 21.9 63.0 476.2 20000 456.9 6016 95.5 79.3
40000 22.0 85.0 470.4 20000 454.0 7764 91.3 79.4
50000 25.3 110.3 453.4 20000 395.8 9792 88.8 80.3
Done

BTW, I used "--max_rows=50000" but the task does not terminate by
itself, is this as expected? I.e. the "Done" shows and after minutes it
still has not closed I need to "ctrl-c" it and restart the fb server if
I want to run it again, as otherwise a table is in use.
> but it seems strange that you have such different performance results
Yeap, especially seeing the following.

Copied the test db to my external disk which is an eSata drive:
#rows #seconds #total_seconds cum_ips table_size last_ips #queries
cum_qps last_qps
10000 3.5 3.5 2831.0 10000 2831.0 1016 287.6 287.6
20000 4.5 8.0 2487.4 20000 2218.1 1356 168.6 75.4
30000 5.4 13.5 2226.4 20000 1840.3 1732 128.5 69.2
40000 5.3 18.8 2125.7 20000 1871.6 2040 108.4 57.6
50000 4.9 23.7 2106.9 20000 2034.8 2356 99.3 64.3
Done

Obviously this points that the problem I have has to do with the
internal disk.

My fstab contains this:

proc /proc proc
nodev,noexec,nosuid 0 0
UUID=02b1776b-57cc-4934-aa54-02acf90a6716 / ext4
errors=remount-ro 0 1
UUID=25bce7cf-5125-457e-b3e2-bf19f79703d5 none swap
sw 0 0
/dev/sda3 /media/sda3 ntfs
defaults 0 0
/dev/sdb1 /media/MyBookAVI ntfs
defaults 0 0
/dev/sdc1 /media/Kiosk ntfs
defaults 0 0

The external disk I used for the above is /dev/sdb1.

My internal disk is partitioned for dual boot:
/dev/sda1 ntfs
/dev/sda2 ntfs
/dev/sda3 ntfs
/dev/sda4 extended
/dev/sda6 ext4 root
/dev/sda5 linux swap

All of this is on a Sony notebook (i5 core with 4GB) and I didn't want
to depend on the external disk, i.e. be able to travel with just the
notebook. I would therefore prefer if I could tweak the configuration to
have better performance with the internal disk.

Werner




>
> ------------------------------------
>
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>
> Visithttp://www.firebirdsql.org and click the Resources item
> on the main (top) menu. Try Knowledgebase and FAQ links !
>
> Also search the knowledgebases athttp://www.ibphoenix.com
>
> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> Yahoo! Groups Links
>
>
>
>