Optimizing and improving performance
The following settings and configurations maybe optimized in high volume transactions to achieve higher performance.
Variable | Description |
---|---|
STATUS_BEAT_LCR | This determines how often a log message is written showing number of processed LCRs (in LCR count) (use 0 to disable). As of Dbvisit Replicate 2.3.14 the default is 10,000. For high volume configurations consider increasing to higher number. |
OCI_COMMIT_MODE = NOWAIT BATCH | This may improve APPLY processing speed. Only set for non RAC target systems. See Configuration Variable Reference for more information. |
_ENABLE_5_19_PARSE = NO | Disables processing of auditing information. This information is used by CDC/Audit function or by querying the DBRSAPPLY_PKG package on the target database (e.g. in triggers). This information includes source sid, serial#, transaction id, os user, os machine, login name. If this information is not required, then this can be disabled. Please Internal Variables for more information |
MINE
Generally reasons for slow down of MINE are:
- Very large transactions that cause MINE to spill to disk because the transaction is too large to keep in memory (see messages "SPILL_TO_DISK" in the log file). Look at large transactions and increase memory of MINE.
- Slow disks for the PLOGS. Ie MINE cannot write fast enough to the PLOG files. Consider moving the PLOGS to fast disks separate from the Oracle redo and archive logs.
- Many "Create Table as Selects" - CTAS. These are slow operations for MINE because Oracle does not write the whole table to the redo and so MINE has to extract the whole table with a normal select statement to replicate the data. If these are used as temporary tables, it is recommend to filter out CTAS
- Exotic datatypes such as LOBS etc. These require extra processing to replicate. If these data types are not needed, it is recommend to exclude these columns from the replication.
Improving performance
Following settings/tips can be applied to improve Replicate performance. Please test them first on a test system before implementing in a production environment.
Process | Type | Description | Setting |
---|---|---|---|
MINE | No CDC/AUDIT | Disable maintaining of session information. This information is needed only if CDC/Audit is in use to populate auxiliary fields such as source user login, program, module, action and so on. If _MINE_PRINT_OPEN_TRANSACTIONS is used, this feature populates the sid/serial info. | _ENABLE_5_19_PARSE = NO |
No trace files | Disable keeping of past debug messages. So the trace files will not contain this information. This is only recommended in a tested environment. | _DISABLE_DBGPRINTF = YES | |
No redo log checksums | Disable checking of redolog checksums. This is recommended only if
| _MINE_CHECK_ORACHECKSUM = NO | |
No error logging to database | Do not log errors to DBRSCOMMON_ERRORLOG. Note that this decreases the table size and not performance. | _LOG_ERRORS_TO_DATABASE = NO | |
Less frequent updates of table names in list progress | This setting only updates table names in list progress when a log switch occurs. Note that new tables are added on the fly even with this setting. The setting can confuse if there is a lot of DDL, table renames etc. Then list progress may show stale information until next log switch happens. | _MINE_UPDATE_TABLIST_ON_SWITCH_ONLY = YES | |
No compression of kept PLOGs/redologs | Disables compression of obsolete PLOGs (or redologs - that's with FETCHER only). This means less logs will be available for support if needed. | DELETE_OBSOLETE_PLOGS_GZIP = NO DELETE_OBSOLETE_RLOGS_GZIP = NO | |
Move PLOGS to fast disks | Consider moving the PLOGS to fast disks separate from the Oracle redo and archive logs. | ||
APPLY | No trace files | See MINE | |
No error logging to database | See MINE | ||
Commit Nowait | Change default commit settings. Use only if target is not RAC, otherwise data loss can occur. | OCI_COMMIT_MODE = NOWAIT BATCH | |
No audit info, no triggers | if DBRSAPPLY_PKG is used at all, APPLY can stop populating it's variables. It's used for:
| _SET_APPPLY_PKG_AUDIT NO (on login) _SET_APPPLY_PKG_CONFLICT NO (on conflict) _SET_APPPLY_PKG_ALWAYS NO (on each change; NO is the default) | |
No compression of kept PLOGs | Disables compression of obsolete PLOGs. This means less PLOGs will be available for support if needed. | DELETE_OBSOLETE_PLOGS_GZIP = NO | |
Disabling Deletion of obsolete PLOGs and RLOGS | Disabling deletion of Obsolete PLOGS (or Redologs with FETCHER only) . This means the replication does not have to go through the plog maintenance ,which can speed up the replication. If this parameter is set to NO, The PLOGS and RLOGS maintenance has to be done manually or by using a separate method | set *._DELETE_OBSOLETE_PLOGS_ENABLED = NO |
Examples:
This shows an example of how to change the settings to further improve the performance:
Please add these settings to your MINE ddc file on the source server after the line ########## Any manual settings should go below this line ########## as instructed for them to take effect. Note a restart is also needed to take effect.
memory_set _ENABLE_5_19_PARSE NO -- this disables auxiliary data like action, module, osprog, osproc etc. available on apply with CDC/Audit or using the DBRSAPPLY_PKG package memory_set _DISABLE_DBGPRINTF YES -- this disables most of the contents of trace files memory_set _MINE_CHECK_ORACHECKSUM NO -- disable checking of redo log checksums memory_set _LOG_ERRORS_TO_DATABASE NO -- disables logging of errors to DBRSCOMMON_ERRORLOG table memory_set _MINE_UPDATE_TABLIST_ON_SWITCH_ONLY YES -- updates table names in list progress only on log switch; so if you rename a table, you won't see the change until a log switch
Please add these settings to your APPLY ddc file on the target server after the line ########## Any manual settings should go below this line ########## as instructed for them to take effect. Note a restart is also needed to take effect.
memory_set _DISABLE_DBGPRINTF YES memory_set _LOG_ERRORS_TO_DATABASE NO
Further performance improvements
Performance profiling statistics can be collected by Dbvisit Replicate and these can be sent to Dbvisit for further analysis to determine if further optimisations can be made for the specific environment. Please see Dbvisit Replicate Profile Performance Statistics for more information on how to collect profiling statistics.