23.06.2014 20:41, Nikolay Samofatov wrote:
> Hello, Dmitry!
>
> On 06/23/2014 04:15 AM, Dmitry Yemanov wrote:
>> 21.06.2014 01:52, Nikolay Samofatov wrote:
>>> I attach patch for this functionality to give you an idea of
>>> implementation. It depends on a couple other changes so it doesn't apply
>>> to FB2.5 cleanly.
>> The first thing I'm interested in is the performance impact in OLTP
>> tests similar to TPCC (with lots of short concurrent requests). Do you
>> have any numbers? If the difference is noticable, then I believe the new
>> mode should not become a default one, it should be turned on expicitly.
>
> I didn't run TPCC, but I created some artificial tests to measure overheads 
> of new logic.
>
> Most tests show no statistically significant difference.
> In various tests, the worst I have seen was 2% slowdown when doing count(*) 
> on a fully cached table
> with 2 integers.
> Some tests (namely, execute statement in autonomous transaction in a loop) 
> show statistically
> significant speed up of up to 4%.
>
> I had to rewrite the code a few times, and analyze it with a profiler to 
> achieve this result.
>
> If you can point me to a situation when new code is, say, more than 5% slower 
> than old logic in a
> test (even totally artificial one), I am willing to fix it.

   I expect most performance penalty to happen when there is a lot of active 
concurrent transactions
and our transaction reads a lot of records, produced by active concurrents. I 
mean active at start
of our request, i.e. requests's snapshot contains many entries. And, of course, 
most of the read
records should be in cache.

Regards,
Vlad

------------------------------------------------------------------------------
Open source business process management suite built on Java and Eclipse
Turn processes into business applications with Bonita BPM Community Edition
Quickly connect people, data, and systems into organized workflows
Winner of BOSSIE, CODIE, OW2 and Gartner awards
http://p.sf.net/sfu/Bonitasoft
Firebird-Devel mailing list, web interface at 
https://lists.sourceforge.net/lists/listinfo/firebird-devel

Reply via email to