MySQL Bugs: #22354: Atomics and TPC-B testing show performance drops between 5.0.24 vs 5.1.12

Bug #22354	Atomics and TPC-B testing show performance drops between 5.0.24 vs 5.1.12
Submitted:	14 Sep 2006 15:54	Modified:	4 Dec 2008 19:18
Reporter:	Jonathan Miller	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Cluster: Cluster (NDB) storage engine	Severity:	S5 (Performance)
Version:	5.1.12	OS:	Linux (Linux (AMD64))
Assigned to:	Jonathan Miller	CPU Architecture:	Any

Description:
I have run the following tests against both 5.1.12 and 5.0.24 three times and took the average to create the charts.

Atomics - 1,000,000 iterations for each test.
TPC-B - 10 Minute Runs with the following number of processes (12, 24, 36, 48, 60)

The test show that we have about a 27% drop in 5.1.12 compared to 5.0.24. 

I am assuming that this has to do with changes in the MySQLD.

How to repeat:
[ndbd default]
NoOfReplicas: 2
IndexMemory: 800M
DataMemory: 2500M
BackupMemory: 64M
MaxNoOfConcurrentScans: 500
DataDir: /space/run
NoOfFragmentLogFiles: 150

[MGM DEFAULT]
PortNumber: 14000
DataDir= /space/run/

[ndb_mgmd]
Id=1
HostName= n15

[ndbd]
Id= 2
HostName= n15

[ndbd]
Id= 3
HostName= n14

[mysqld]
Id= 4

[mysqld]
Id= 5

[api]
Id= 6

[mysqld]
Id= 7

[mysqld]
Id= 8

[mysqld]
Id= 9

[mysqld]
Id= 10

[api]
Id= 11

[api]
Id= 12

TPCB:
4 seperate hosts running MySQLD
Load the database
perl load_tpcb.pl --sock
Start the slave process on 3 of the hosts
perl tpcb_driver.pl -sl
Start the master on the last host
perl tpcb_driver.pl -ma -nu 4 -u root -sock nt -5 -tm 3

ATOMICS:
On a singe MySQLD host:
./AtomicsTest --host n13 --user root --test ALL --iter 1000000 --pass test

Charts and files are attached to this bug

What was your mysqld setting for ndb-use-exact count resp. ndb-index-stat 
(i.e also upload you all you my.cnf's, or 1 if they are identical)

Can you try with some different combinations?
  (for one of the testcases above)

Also could you try innodb and/or myisam 5.0 vs 5.1

I just last week ran a micro benchmark(ndbapi) of that actually showed
  that 5.1 was faster than 5.0 (mail sent to dev-ndb)
  (unless using varsize which was slightly slower than fixed size)

Does test use varsize ?

/Jonas

Jonas,

The test you requested except for the Innodb and MyISAM have been attached.
/jeb

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".

No feedback was provided for this bug for over a month, so it is
being suspended automatically. If you are able to provide the
information that was originally requested, please do so and change
the status of the bug back to "Open".

Too much time has passed. ACRT will pickup any drops moving forward.