Re: Linux--Oracle Control File Errors

From: David Sisk <davesisk_at_ipass.net>
Date: 1998/11/16
Message-ID: <WiF32.28$_67.71@news.ipass.net>

Hi:

My best guess is that there is contention for the control file header because another checkpoint is occuring before the first one finished. Checkpoints are triggered by two things during normal operation: Number of bytes written, and elapsed time. Try modifying the following init.ora parms:

log_checkpoint_interval: This is in bytes, so set it larger than your redo log files. (You might also want to make your redo log files larger first. 10-50Mb or even larger is not an unusual setting.) If you set this parm larger than your redo log file size, then you will only get a checkpoint on each redo log switch. Now you're getting a checkpoint every 99K bytes, unless your redo logs are 99K or smaller.
log_checkpoint_timeout: This is in seconds (I believe), so set it such that you get a checkpoint every 30 minutes or hour, etc.

Unless I'm mistaken, the control files are only updated on a checkpoint or DDL operation. Changing these settings should make the errors go away AND speed up your insert performance. Also, take a good look at the PCT_FREE setting on the tables, as this can affect insert performance. (Check the tuning guide.)

Sounds like you're doing a benchmark. 100,000 inserts/minute? Sounds fantastic. (Is it according to TPC-C secs?) Keep my email address and let me know what kind of performance you end up with. I want to convince my employer to switch from NT to Linux, and some of your data may help.

Thanks and regards,
Dave

--
David C. Sisk
http://www.ipass.net/~davesisk/ORACLEonNT.htm


AJG wrote in message <364E5FD7.9045D731_at_noemail.noemail>...


>Hi,

>

>I'm doing some stress testing, experimental tuning specifically for

>INSERTs, and am getting control file errors in the alert log (below), on

>every checkpoint/log switch.

>

>The error does not affect anything in the running instance.  It does not

>affect startup and shutdown either.  It just makes an entry in the alert

>log, with a trace file.

>

>I'm probably pushing the envelope trying to use a 16K block size, but it

>took it OK.  However, it's after I recreated with 16K block size that

>these errors started.

>

>So, I'm just wondering if anyone is seeing this error message also, and

>maybe if there is something for it.

>

>This would be 8.0.5 pre-release for Linux.  I'm running RH 5.1/Linux

>2.0.34 with SMP turned on (2 cpus).

>

>

>Thanks!

>

>Alan.

>

>PS: Stress test results -> 100,000 records/min. is the best yet.

>

>

>

>

>Instance Setup (from alert log):

>

>System parameters with non-default values:

>  processes                = 50

>  shared_pool_size         = 3500000

>  control_files            =

>/opt/oracle/database/a001/ctrl/disk01/control1.ctl

>  db_block_buffers         = 10000

>  db_block_size            = 16384

>  db_writer_processes      = 4

>  log_archive_start        = TRUE

>  log_archive_dest         = /opt/oracle/database/a001/arch/disk01

>  log_archive_format       = T%TS%S.ARC

>  log_buffer               = 5242880

>  log_checkpoint_interval  = 99999

>  db_files                 = 80

>  db_file_multiblock_read_count= 8

>  log_checkpoints_to_alert = TRUE

>  rollback_segments        = rbs01, rbs02, rbs03, rbs04

>  sequence_cache_entries   = 10

>  sequence_cache_hash_buckets= 10

>  global_names             = TRUE

>  db_name                  = a001

>  parallel_max_servers     = 5

>  background_dump_dest     = /opt/oracle/admin/a001/bdump

>  user_dump_dest           = /opt/oracle/admin/a001/udump

>  max_dump_file_size       = 10240

>  core_dump_dest           = /opt/oracle/admin/a001/cdump

>

>

>

>

>Example error from alert log:

>

>Beginning log switch checkpoint up to RBA [0x2f1.2.10], SCN:

>0x0000.00038b45

>Thread 1 advanced to log sequence 753

>  Current log# 1 seq# 753 mem# 0:

>/opt/oracle/database/a001/redoA/disk01/redoA_01.rdo

>  Current log# 1 seq# 753 mem# 1:

>/opt/oracle/database/a001/redoB/disk01/redoB_01.rdo

>Sat Nov 14 20:44:25 1998

>Completed checkpoint up to RBA [0x2f1.2.10], SCN: 0x0000.00038b45

>Sat Nov 14 20:44:52 1998

>Errors in file /opt/oracle/admin/a001/bdump/lgwr_a001_5732.trc:

>ORA-00202: controlfile:

>'/opt/oracle/database/a001/ctrl/disk01/control1.ctl'

>ORA-27044: unable to write the header block of file

>Linux Error: 14: Bad address

>Additional information: 4

>Sat Nov 14 20:44:52 1998

>kccexpd: controlfile resize from 62 to 64 block(s) denied by OS

>krcpwnc: following controlfile record written over:

> RECID #299 Recno 299 Record timestamp

> 11/14/98 15:44:56

> Thread=1 Seq#=299 Link-Recid=298

>  Low

>scn: 0x0000.00003721

> 11/14/98 15:44:55

> Next

>scn: 0x0000.0000372f

>Beginning log switch checkpoint up to RBA [0x2f2.2.10], SCN:

>0x0000.00038c9c

>Thread 1 advanced to log sequence 754

>  Current log# 2 seq# 754 mem# 0:

>/opt/oracle/database/a001/redoA/disk01/redoA_02.rdo

>  Current log# 2 seq# 754 mem# 1:

>/opt/oracle/database/a001/redoB/disk01/redoB_02.rdo

>Sat Nov 14 20:44:57 1998

>Completed checkpoint up to RBA [0x2f2.2.10], SCN: 0x0000.00038c9c

>

>

>TRACE FILE:

>

>Dump file /opt/oracle/admin/a001/bdump/lgwr_a001_5732.trc

>Oracle8 Release 8.0.5.0.0 - Production

>PL/SQL Release 8.0.5.0.0 - Production

>ORACLE_HOME = /opt/oracle/product/8.0.5

>System name:   Linux

>Node name:  darkness.alangano.com

>Release: 2.0.34

>Version: #11 Sun Sep 27 01:32:51 CDT 1998

>Machine: i686

>Instance name: a001

>Redo thread mounted by this instance: 1

>Oracle process number: 5

>Unix process pid: 5732, image: ora_lgwr_a001

>

>*** 1998.11.14.19.34.51.597

>*** SESSION ID:(4.1) 1998.11.14.19.34.51.596

>ORA-00202: controlfile:

>'/opt/oracle/database/a001/ctrl/disk01/control1.ctl'

>ORA-27044: unable to write the header block of file

>Linux Error: 14: Bad address

>Additional information: 4

>ORA-00202: controlfile:

>'/opt/oracle/database/a001/ctrl/disk01/control1.ctl'

>ORA-27044: unable to write the header block of file

>Linux Error: 14: Bad address

>Additional information: 4

>.

>.

>.

>ORA-00202: controlfile:

>'/opt/oracle/database/a001/ctrl/disk01/control1.ctl'

>ORA-27044: unable to write the header block of file

>Linux Error: 14: Bad address

>Additional information: 4

>

>

>

>--

>Alan Gano

>http://home.att.net/~alangano/index.html

>

>

>

Received on Mon Nov 16 1998 - 00:00:00 CST