Bug #48956 Partition tests are intolerably slow on i5os (factor more than 10)
Submitted: 20 Nov 2009 21:46 Modified: 25 Dec 2009 23:13
Reporter: Joerg Bruehe Email Updates:
Status: No Feedback Impact on me:
None 
Category:MySQL Server: Partitions Severity:S3 (Non-critical)
Version:5.5.0-beta OS:IBM i
Assigned to: CPU Architecture:Any
Triage: Needs Triage: D2 (Serious)

[20 Nov 2009 21:46] Joerg Bruehe
Description:
The builds and test runs on our i5os machines are horribly slow.
For the test suites, this typically causes suite timeouts (the value is set to 18,000 seconds = 5 hours).

Particularly affected are the "partition" tests,
here are just the times of the "parts.partition_alter*" tests
on i5os (both 32 and 64 bit) in comparison to AIX 5.3.
For all tests that passed on i5os, the reported time is more than 10 times as long as on AIX.

Tests which are not listed on i5os fell victim to the suite timeout:

aix5.3-powerpc-64bit:
partition_alter1_1_2_myisam   [ pass ]  32463
partition_alter1_1_myisam     [ pass ]  34182
partition_alter1_2_myisam     [ pass ]  63703
partition_alter2_1_myisam     [ pass ]  148908
partition_alter2_2_myisam     [ pass ]  149033
partition_alter3_myisam       [ pass ]   6855
partition_alter4_myisam       [ pass ]  363207
partition_alter1_1_2_innodb   [ pass ]  171773
partition_alter1_1_innodb     [ pass ]  106300
partition_alter1_2_innodb     [ pass ]  250004
partition_alter2_1_innodb     [ pass ]  415462
partition_alter2_2_innodb     [ pass ]  385979
partition_alter3_innodb       [ pass ]   7353
partition_alter4_innodb       [ pass ]  504883

i5os-power-32bit:
partition_alter1_1_2_myisam   [ pass ]  401119
partition_alter1_1_myisam     [ pass ]  429000
partition_alter1_2_myisam     [ pass ]  792946
partition_alter2_1_myisam     [ fail ]  timeout after 900 seconds
partition_alter2_1_myisam     [ retry-fail ]  timeout after 900 seconds
partition_alter2_2_myisam     [ fail ]  timeout after 900 seconds
partition_alter2_2_myisam     [ retry-fail ]  timeout after 900 seconds
partition_alter3_myisam       [ pass ]  84387
partition_alter4_myisam       [ fail ]  timeout after 900 seconds
partition_alter4_myisam       [ retry-fail ]  timeout after 900 seconds
partition_alter1_1_2_innodb   [ fail ]  timeout after 900 seconds
partition_alter1_1_2_innodb   [ retry-fail ]  timeout after 900 seconds

i5os-power-64bit:
partition_alter1_1_2_myisam   [ pass ]  350052
partition_alter1_1_myisam     [ pass ]  375186
partition_alter1_2_myisam     [ pass ]  688236
partition_alter2_1_myisam     [ fail ]  timeout after 900 seconds
partition_alter2_1_myisam     [ retry-fail ]  timeout after 900 seconds
partition_alter2_2_myisam     [ fail ]  timeout after 900 seconds
partition_alter2_2_myisam     [ retry-fail ]  timeout after 900 seconds
partition_alter3_myisam       [ pass ]  77611
partition_alter4_myisam       [ fail ]  timeout after 900 seconds
partition_alter4_myisam       [ retry-fail ]  timeout after 900 seconds
partition_alter1_1_2_innodb   [ fail ]  timeout after 900 seconds
partition_alter1_1_2_innodb   [ retry-fail ]  timeout after 900 seconds
partition_alter1_1_innodb     [ pass ]  652874
partition_alter1_2_innodb     [ fail ]  timeout after 900 seconds

How to repeat:
Just try it.

Suggested fix:
Not sure.

Judging from the AIX times, the tests seem to be tolerable.
(OTOH, note that Solaris-Sparc also has some issues with these.)

We need to be sure there is nothing in the server code that causes specific problems on i5os.

Maybe there is nothing we can do about this, other than note and document it.
[20 Nov 2009 22:06] Joerg Bruehe
This might be related:
AIX 5.3:
stress.ddl_myisam      [ pass ]  81854

i5os (32 bit):
stress.ddl_myisam      [ fail ]  timeout after 900 seconds
stress.ddl_myisam      [ retry-fail ]  timeout after 900 seconds

i5os (64 bit):
stress.ddl_myisam      [ pass ]  878548
[26 Nov 2009 8:08] Mikael Ronström
See BUG#47135 for a fix of this
[26 Dec 2009 0:00] Bugs System
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".