|Subject||ODP: Re: [firebird-support] About firebird performance|
-------- Oryginalna wiadomość --------
Od: "Alexey Kovyazin ak@... [firebird-support]" <email@example.com>
Data: 05.09.2015 19:06 (GMT+01:00)
Temat: Re: [firebird-support] About firebird performance
I discovered something else quite interesting:
I realized both IB and FB dbs have pagebuffers set to 100.000. I changed FB to 0 and now it works like I expected!!..Same query now completes on FB on 50 sec. It seems this is the problem although I don't understand why.
100 000 is too high for Firebird SuperServer, it should be set to 10000.
You can use optimized Firebird configuration files
Nevertheless, I'll try on a physical machine as you and Carlos suggest and see how it behaves.
2015-09-05 15:47 GMT+02:00 Alexey Kovyazin ak@... [firebird-support] <firstname.lastname@example.org>:
Thanks for statistics, it clarifies situation.
There is 304880 data pages, each is 4kb, so Firebird need to read 1.1Gb of data to count records.
Count uses NATURAL scan, so it's a simple set of disk reads, CPU is not involved.
Regular SATA3 disk should read 30-70Mb per second (I assume database is not fragmented, since you told it was freshly restored), it gives us estimation from 15 to 30 seconds.
So, I think there is a disk IO problem.
Try it on another computer (better not virtual).
He again Alexey,
As I suspected, my problem is not caused by GC:
gstat over the table reported this:
TABLE1 (503)Primary pointer page: 9374, Index root page: 9375Average record length: 317.63, total records: 3261775Average version length: 0.00, total versions: 0, max versions: 0Data pages: 304880, data page slots: 304880, average fill: 90%
Also, as you suggested ran the query twice a got exaclty the same times: 51minutes to complete!!!
I must have something obvious which is killing firebird performance, but cannot guess it.
Any help would be appretiated.
2015-09-05 11:59 GMT+02:00 Alexey Kovyazin ak@... [firebird-support] <email@example.com>:
So you finally decided to try Firebird - that's good.
I think that you are facing garbage collection problem - if there are a lot of record versions in the TABLE1 which are not interested to any transaction, your query will force garbage collection, and it can take some time.
How to check? run SELECT count(*) twice and compare execution times, and other query statistics (reads/writes).
Also, run gstat -a -r <database> and check information for TABLE1 - VERSIONS and MAX VERSIONS, if there are big numbers there, problem is certainly related with record versions, with the initial cause in wrong transactions management - i.e., long-running write transactions or forced rollbacks.
I'm doing some basic tests with Firebird and I'm facing something I
cannot understand (I'm a newbie to Firebird which has worked with
IB7.5 for years, so I apologize if I ask something obvious):
- I have a linux server with firebird 2.5.4 Superserver: multi-core (4), 3GB RAM
- I placed a "huge" database: 1000 tables, 9GB
- When I issue a query like this "select count(*) from TABLE1" it
takes very long time to complete (>30 min).
- TABLE 1 has 3 million records
As I told you, I'm quite used to IB7.5, but I guess this is not normal
at all...probably I'm missing something but I couldn't find any
document which gives me an idea of what I'm doing wrong.
With same hardware an IB7.5 same query took 40 sec to complete (first
time issued, no cache).
Could anyone help?
Planatec Software S.L. ** <http://www.planatec.es>
telf: +34 964 340 560 ** fax: +34 961 130 921
--Planatec Software S.L. ** <http://www.planatec.es>
--Planatec Software S.L. ** <http://www.planatec.es>telf: +34 964 340 560 ** fax: +34 961 130 921