Import SCAs to a DI results in "checkpoint not complete" in alert log

[Former Member]

Hello everybody,<br>

<br>

I am setting up a new shared NW 7.11 DI system here. Right now I am doing some test imports.<br>

<br>

The sap system ist running on a SLES 10 SP2 server with oracle 10.2.0.4 as DB. 4 CPUs 16GB RAM <br>

<br>

Oh yes, maybe I should say that I am not a true oracle expert, I know a thing or aonther but...<br>

<br>

When doing the import in CMS, naturally, a lot of data is writen to the DB. So I checked the alert log when importing, to see how the redo log groups and archiving is getting along with the import.<br>

<br>

I see quit some "checkpoint not complete" messages in the alert log. Which usualy is not a good thing when happening very often.<br>

<br>

So I did what usually is recommended when this happens. I increased the number of redo log groups from 3 to 6 to 10 and made some test with increasing the number of dbwr processes. <br>

<br>

After sepnding 2 days testing and reading documentation. I only can say that all the measures I have taken did not make any real difference at all. The import took round about 5 minutes with 2 dbwr and 6 redo log groups doing 12 logswitches and and it had the same result with 4 dbwr process and 10 redo log groups.<br>

<br>

5 minute is an acceptable time for this import, but I don#t understnd the log entries.<br>

<br>

I am note sure if this alert log entries make any difference at all. i realsised the first time i saw them that oracle say a checkpoint is not complete and, sometime in the same secound, it makes the checkpoing and frees a redo log group. Looks like this:<br>

<br>

Thu Feb 11 12:20:21 2010<br>

Thread 1 cannot allocate new log, sequence 462<br>

Checkpoint not complete<br>

Current log# 3 seq# 461 mem# 0: /oracle/C11/origlogA/log_g13m1.dbf<br>

Current log# 3 seq# 461 mem# 1: /oracle/C11I/mirrlogA/log_g13m2.dbf<br>

Thu Feb 11 12:20:23 2010<br>

Completed checkpoint up to RBA [0x1c5.2.10], SCN: 5932669<br>

Thu Feb 11 12:20:23 2010<br>

Beginning log switch checkpoint up to RBA [0x1ce.2.10], SCN: 5949907<br>

Thu Feb 11 12:20:23 2010<br>

Thread 1 advanced to log sequence 462 (LGWR switch)<br>

Current log# 7 seq# 462 mem# 0: /oracle/C11I/origlogA/log_g17m1.dbf<br>

Current log# 7 seq# 462 mem# 1: /oracle/C11/mirrlogA/log_g17m2.dbf<br>

<br>

So I understand why I don't see a big performance impact when increasing dbwr and log groups, becuase the checkpint not complete wait time is quit short. But I do not really understand why oracel is not freeing the log groups sooner. it seems not to be a problem after all when it is absolutely required. <br>

<br>

I read about the new feature that oracle can adjust the chckpoint intraval to meet FAST_START_MTTR_TARGET but this is not set in my case..<br>

<br>

Thu Feb 11 10:23:36 2010<br>

MTTR advisory is disabled because FAST_START_MTTR_TARGET is not set<br>

<br>

Can any of you experts out there give me some advice or better an explaination what this entrys mean or are they benign and can be ignored after all? I almost guess so because in Enterprise Manger nothing is shown in alert log warnings etc.<br>

<br>

Looking forward hearing fron you<br>

Jan<br>

Edited by: Jan Benkert on Feb 11, 2010 1:55 PM

[Former Member]

Thanks for your replys,

I have already read the notes. but I didn't find anything good there.

I am not sure what kind of data is imported. It is a source code import out of a SCA file in NWDIs CMS.

I increased my redolog groups to 15 now. and it sems this 15 groups can handle the load with out checkpont not complete messages.

Can anyone tell me if there is a down side on haveing too much redolog groups?

It would be very kind if anyone can tell me what is better. more groups or bigger redolog members.?

Thanks for your help

Jan

[Former Member]

I would rather increase the size instead of the number of online redo log files.

Increasing the number of log files will not reduce the number of log switches needed.

But increasing the size will.

[former_member204746]

checkpoint not complete is not a major error. consider it as a warning. If this only happens once in a while, then, no need to panic. if it happens often, then resize or add members.

[Former Member]

Thanks for the answer.

I switched to 200MB redolog files. and it seems to work just fine.

[Former Member]

I am not exactly sure what kind of data you import, but in case of a significant amount of LOB data you can set CACHE=NO for the LOB segment. Then all changes will be done directly on disk and the DBWRs don't need to be very busy. If you can restart the load when something goes wrong you can even set LOGGING=NO for the LOB segment, so that no redo log data is created.

[former_member204746]

you have 2 possible solutions:

1. increase size of redo logs

2. increase number of redologs

also check that mirrored redo logs are set on different physical disks.

Read SAP notes:

79341 Checkpoint not complete

1068186 "checkpoint not complete" message increasingly in alert file