All 9 entries tagged Oracle-Admin
No other Warwick Blogs use the tag Oracle-Admin on entries | View entries tagged Oracle-Admin at Technorati | There are no images tagged Oracle-Admin on this blog
May 22, 2007
Writing about web page http://www.netapp.com/library/tr/3369.pdf
NFS mount options:
rw,bg,vers=3, proto=tcp, hard, intr,rsize=32768,wsize=32768,forcedirectio,noac
- Forcedirectio: only use this option with volumns containing data and online log files. (This article page 20 seems to suggest that it is not good idea to forcedirectio on datafiles.
DISK_ASYNCH_IO=True and DB_WRITER-PROCESSES = 2 * No. of CPU Note: AIO is enabled by default for Oracle9i+ on Solaris
DB_FILE_MULTIBLOCK_READ_COUNT = 1-4 (OLTP) or 16-32(DSS)
DB_BLOCK_SIZE = Solaris page size * n
With multiple CPU, use DB_WRITER_PROCESSES instead of DBWR_IO_SLAVES
NFS/CIFS—> Coordinate with Oracle
Fibre Channel or iSCSI—> Coordinate with Server
- SnapManager (rman)
- SnapMirror (OS)
- OS copy
- NDMP command
- Third-party backup tools
- Use Snapshot copies for performing cold or hot backup. Turn off the automatic snapshot scheduler and coordinate snapshot with the state of database
- SnapRestore If using SnapRestore on a volume level, it is recommended to store log,archivelog and control files on a separate and use SnapRestore only on the volume containing datafiles.
- SnapMirror: (between NA devices)
- SnapVault = SnapMirror + archival
SnapVault like a backup coordinator (Between NA and 3rd party?)
- SnapManager for Oracle components: Oracle9i+,Solaris or RH, SnapRestore, FlexClone, Data ONTAP
SnapManager provides GUI to schedules snapshot copies and ensure copies are consistent from an application standpoint and SnapMirror replicates copies.
- DAS: Direct-attached storage
- FAS: fabric-attached storage
- NAS: network Attached Storage : protocol TCP/IP. NAS offers significan advantages over SAN for database deployment
- SAN: storage area network: protocol SCSI and Fibre Channel Protocol (FCP)
- FC: Fibre Channel
- Two Unix PRIMEPWOER 400N servers, running Solaris8 and PRIMECLUSTER 4.0 software , access clustered NetApp FAS systems. 26 Windowns servers in FB SAN configuration. NetApp unified storage solution seamlessly blens NAS and SAN.
- Whether to share oracle_home on filer? Plus side: add more server quickly; downsize: maybe cannot patch oracle one by one.
In our system, maybe I should opt for not sharing. Different servers have different availability requirement. Shared Oracle home require to shutdown all our server during upgrading,patching. If one upgrade fails, you will have tremendous pressure to rectify it.
- How many volumes are optimal? : At least two – one for datafiles; and one for online and archive logs and binaries (if put share oracle home on filer).
- We have several databases. One way is to use one big volume for databases and the datafiles belonging to one database will share one qtree. The other way is use one volume for one single database.
The purpose in doing so is not so that clients can…
As mentioned previously
SnapManager for Oracle is protocol agnostic and thus work seamlessly
from application standpoint
The competing technology simply cannot match
Fibre <- >fabric
the blame has been attributed to it
perceived performance problems counters the traditional view
May 17, 2007
I imported a file containing a big partition table. The file only contains one partition.
After I entered the command, the process just hang there. After a while, database instance crashed due to one disk is full.
It turned out before Oracle imports the data, it firstly try to locate space for the table. Even I only imported one partition, it tried to locate the space for the whole table.
Because the datafile are autoextensible, it grew and grew and eventually fill up the disk.
I tried to export the table with rows=n and import the file, Oracle still tried to expand the space. Weird, I did not remember that happened before.
Silence kills.Having not got time to re-test it, onel caveat learned from it:
- do not turn datafiles autoextensible on when import a file
In another occation, it tookimport process 30 minutes to report that the import user does not exist. What is it doing ?
April 30, 2007
The value of column program in v$session can be “JDBC Thin Client”,”sqlplus..”,”OMS” ..
For a remote JDBC connection, it will display “JDBC Thin Client” correctly. For a local JDBC connection, it displays “null” value.
So for a application connection , a “null” value will probably means JDBC connection
April 10, 2006
I recently upgrade a database server. It is relatively big compared to our other servers, about 80g in total.
I chose to use GUI DBUA mode to upgrade the server and expected it would take about 2 hour. The process, however, managed to run through 3 days. On the 4th day, the X window displayed on my computer was killed due to that some guy upgraded the network switch.
I checked the process and found that the major part upgrading had finished. So I dropped the components not upgraded and every things seemed work fine.
Lessons to learn:
Have to backup services. If the process take longer, bring up backup.
March 31, 2006
Experience of Upgrading Server:
- I tried different methods to upgrade: dbua in GUI mode and slient mode. Either works fine. GUI mode discloses more information about which component to upgrade. This can help you to drop unused componnent before upgrading, for instance, JServer take a great time to upgrading. It is wonderful to drop it.
- The following paragrah is rather confusing.
The SYSAUX tablespace must be created after the new Oracle Database 10g release is started and BEFORE the upgrade scripts are invoked.
February 11, 2006
- One very good point is multiplexing online log and control file even you are using RAID 0+1 (stripe & mirror)
- RAID 0+1 for undo space since it cannot be multiplexed, read/write frequently
- RAID 5 or 0+1 for datafile,
The result of df -h on our serve:
is every entry a paired disk?
If not, how to check, fdisk?
The relationship of controller and disk, 1 to 1 ?
the IO speed decided by controller , I guess ? Then how controller interact with the disks?
Also see: link
0+1 and 1+0 have same performance. 1+0 has better availability.
What we use then ?
RAID 0: no reliability; improve performance
RAID 1: provides the highest performance for redundant storage
RAID 2: No implementation
RAID 3: Not good for database
RAID 4: Performance no better than 5
RAID 5: Parity RAID allows redundancy with less total storage cost. The loss of a single disk will cause read performance to be degraded while the system reads all other disks in the array and recomputes the missing data.
January 20, 2006
Writing about web page http://www.oracle.com/technology/pub/columns/trute_unicode.html
Oracle9i introduced length semantics, which provides the choice of whether to declare strings in terms of characters or bytes. Byte semantics is still the default, and character-length semantics can be declared at column, table, session, and for the entire database. Length semantics makes storage management more intuitive and allows a common database schema to run successfully on different databases with different character sets. This makes migrating applications and databases to Unicode much simpler.
April 15, 2005
One of our Oracle servers managed to crash its host server.
You would assume that if the physical memeory is not enough, oracle will report an error.
But it did not ! It simply use too much memory that eventualy bring down the whole server!
January 26, 2005
1. Delete sys stats if necessary and recreate after upgrade finishing
2. Set pool size to at least 150M
ALTER SYSTEM SET SHARED_POOL_SIZE='150M' SCOPE=spfile;
ALTER SYSTEM SET JAVA_POOL_SIZE='150M' SCOPE=spfile;
If java pool is not used, set it to 0 after finished upgrading
3. Not sure if you just upgrade the client, should you run it ?
Oracle Recovery Manager catalog, enter the following command:
$ rman catalog username/password@alias
RMAN> UPGRADE CATALOG;