Watch, Follow, &
Connect with Us

Please visit our new home
community.embarcadero.com.

Public Report
Report From: InterBase/Server/Performance    [ Add a report in this area ]  
Report #:  117734   Status: Need Feedback
IB 2009 is more 40% faster than IB XE3 for reading data from cache
Project:  InterBase Build #:  11.0.2.541
Version:    11.0 Submitted By:   Stephane DORIGO
Report Type:  Issue Date Reported:  8/9/2013 3:38:36 AM
Severity:    Serious / Highly visible problem Last Updated: 8/14/2013 6:27:00 PM
Platform:    32 Bit Internal Tracking #:  
Resolution: Need More Info (Resolution Comments) Resolved in Build: : None
Duplicate of:  None
Voting and Rating
Overall Rating: No Ratings Yet
0.00 out of 5
Total Votes: None
Description
Reading performance test between IB 2009, IB XE and IB XE3 :
------------------------------------------------------------
On a server (Intel Xeon dual CPU 2,5 Ghz, 4 cores per CPU, 16 Go RAM, HDD SAS 15K) with Windows 2003 serveur 32 bits.

Execution of the same SELECT query on 3 latest version of Interbase instance (IB_2009, IB_XE 10.0.1.595, IB_XE3 11.0.2.541) in exactly the same conditions.

  - The 3 instances generate same plan for the query.

  - First time we lunch the query (data essentialy readings from disk to cache) :
      . IB 2009 = ~1'23.
      . IB XE   = ~1'34.
      . IB XE3  = ~1'41.

  - Next times we lunch the query (all data is readed from cache) we find that IB 2009 (ODS 13.2) is more 40% faster than IB XE and IB XE3 (ODS 15.0) !
      . IB 2009 = ~20s.
      . IB XE   = ~37s.
      . IB XE3  = ~35s.

HOW EXPLAIN THIS ?


(In writing and stored procs execution Interbase XE3 is widely the best and Interbase 2009 is the worse).
Steps to Reproduce:
On a Windows server :
  - Install 3 lastest version of the different Interbase instances (2009, XE, XE3).
  - Use the same client for all instances. (Interbase XE3 gds32.dll in our test)
  - No other process than Interbase instance must take CPU time when your query is running.
  - Only 1 user (the query luncher) is connected on the database when your query is running.
  - Same page size for all database (4096 Ko in our test).
  - Same Interbase cache size for all database (50000 pages in our test).
  - Same IBConfig file for all instances.
  - Convert the same database for the 3 different instances.
  - Backup/Restore the 3 databases to start "clean".
  - Preferably, put your 3 databases on a different disk than the OS.
Workarounds
None.
Attachment
None
Comments

Tomohiro Takahashi at 8/11/2013 6:10:12 PM -
Is this report same as QC#117674 ? If yes, please do not post same reports.

Stephane DORIGO at 8/13/2013 2:29:44 AM -
Yes is the same a little bit corrected.
I have reposted because it does not appeared when I have validated. I realized after that you have to validate it to appear in the report list. Sorry...
So you can delete report QC#117674 and keep only QC#117734.

Sriram Balasubramanian at 8/14/2013 10:14:39 AM -
Stephane,
Do you have a database backup file and SELECT query that you can share with us. This would definitely help us analyze your report better.

If you need to provide the database and SELECT query in private, please let me know (Sriram.Balasubramanian@embarcadero.com). You may want to copy Quinn Wildman (InterBase Support), Quinn.Wildman@embarcadero.com, also so he can send you a private FTP location to upload your files.

Best wishes,
Sriram
InterBase Team Lead

Stephane DORIGO at 9/2/2013 8:12:27 AM -
DB file uploaded on the EFT location you have transmited by mail.

Stephane DORIGO at 8/30/2013 12:11:45 PM -
Done.
I sent you an email with a link to our ftp server to download test files.

Tomohiro Takahashi at 8/13/2013 7:00:28 PM -
ok, I deleted QC#117674 as Sysop..

Server Response from: ETNACODE01