| |
 |
|
Oracle Tips by Burleson |
SCSI Runs
The SCSI runs required from June 2, 2004 to July
30, 2004 to complete. This 58 day period encompassed seven full query
runs and a single partial run. This was with queries running as close
as possible to 24X7, between equipment failures and connectivity
issues. During the test, the initial 2-disk SCSI array failed during
the second query set run. This required a database rebuild and reload
on the provided ATA array. After the SCSI failure, the entire initial
query run had to be repeated, compared to the SCSI results and the
decision was made at that time to discard the SCSI results and
continue with the ATA array.
The initial
nologgingand
noarchivingruns performed so poorly that
nologging and
noarchiveloggingset points were used during the subsequent tests as well
to allow the tests to complete in a reasonable amount of time. This
is due to the fact that turning on
loggingand
archivelogwould further stress the I/O subsystem resulting in even
poorer performance. The following issues further complicated the
testing:
-
When running the queries against the 100%
SCSI/ATA database, query number one would not complete, and after 30
hours it was halted.
-
When running the queries against the 100%
SCSI/ATA database, query number four would not complete in 30 hours,
and it was halted.
-
When running against the 100% SCSI/ATA database
query number produced the error codes shown below. It then
terminated the session causing all subsequent queries to fail in run
number one, requiring a restart:
Sun Jun 6 11:43:16 2004
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
Sun Jun 6 11:43:17 2004
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
Sun Jun 6 11:43:18 2004
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
Sun Jun 6 11:43:18 2004
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
[oracle@AMD43 bdump]$
[oracle@AMD43 bdump]$ tail -f alert_dss.log
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
Sun Jun 6 11:43:18 2004
Errors in file /home/oracle/admin/dss/udump/dss_ora_615.trc:
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-07445: exception encountered: core dump [kghbigasp()+289] [SIGSEGV]
[Address not mapped to object] [0x427A8] [] []
ORA-00600: internal error code, arguments: [kftts2bz_many_files], [0],
[39218], [], [], [], [], []
-
The queries would run out of temporary
space, undo segment space, and generate the above errors if run
back-to-back in the same user. The script running the queries was
modified to include a new connect statement before each query,
releasing previous undo
segments and temp segments
for reuse. This was not required in the SSD tests. Each run was
actually several smaller runs restarted at the point of failure.
-
During run number two of the initial SCSI test,
the array failed after approximately eight days of 24X7 processing.
This required the rebuild of the database on the new ATA array.
-
During the final runs, a complete power outage
that caused a reassignment of the ATA array address, which resulted
in two days of down time. The data was fully recovered through the
automated recovery processing and the testing resumed.
-
Several days were also lost due to connectivity
issues.
-
Any query that required longer than ~30 hours was
terminated and processing restarted at the next query in the file.
This was required on queries number one, three, 13, 13a, 17, and 18
in various configurations. The resulting time was set to 1440
minutes (24 hours) and the graph axis set to logarithmic to allow
the graphs to show at least some of the shorter query times.
The above book excerpt is from:
Oracle RAC & Tuning with
Solid State Disk
Expert Secrets for High
Performance Clustered Grid Computing
ISBN
0-9761573-5-7
Donald K. Burleson & Mike Ault
http://www.rampant-books.com/book_2005_2_rac_ssd_tuning.htm
|
