Printable View
In my INIT.ORA file the parameter "log_small_entry_max_size " is not there.
Does this mean the value is not set? If NO then how can I find out the default value of this parameter. I know its OS dependent, but how will know the exact value. My server is 8i and is running on an NT machine.
The reason why I want to know is I am trying to tune the redo-log buffer for latch contention.
Please help.
it´s obsolete in 8i
Thanks for the info Pando. My Latch Hit Ratio is falling below 90%. How do I then tune it. Kindly advise.
Tune strategy depend from type of application(s).
If Ur application has big flow of short trancactions (1-5 db blocks per transaction) avg frequency of commit startments is high.
In this situation lgwr should execute a lot of short copy operations form db cache to log cache and a lot of write operations from log cache to redo log file.
and u should tune # of latch for "redo copy" and "redo entries".
If Ur application has flow of long trancactions (when # of db blocks ~> 50% of size of log cache) then lgwr should execute long "writes".
and u should tune # of latch for "redo writing".
Shestakov: Thanks, but that info doesn't help me much.
My Latch Hit Ratio is falling below 90%. Can someone here give me the step-by-step procedure to tune. Please advise.
As far the theory goes, it says:
To resolve redo latch contention, increase LOG_SIMULTANEOUS_COPIES and decrease LOG_SMALL_ENTRY_MAX_SIZE.
But if these parameters are obselete in 8i, how do we tune? Please advise. Hope someone here is knowledgable about this. Eagerly waiting for the respose.
This isn't absolutly true.Quote:
Originally posted by dbafreak
But if these parameters are obselete in 8i, how do we tune? Please advise. Hope someone here is knowledgable about this. Eagerly waiting for the respose.
1) LOG_SIMULTANEOUS_COPIES now has name
_LOG_SIMULTANEOUS_COPIES - this is undocumented parameter
and u can increase it.
2) LOG_SMALL_ENTRY_MAX_SIZE is obselete.
-------------------------------------------------
Then u can try use asyncronous io operations on log cache, using parameters:
_LGWR_ASYNC_IO ( set to TRUE)
and
_LGWR_IO_SLAVES (set to 1)
both undocumented.
Thanks Shestakov. That was useful information. However, I believe these parameters are not recommended by Oracle to use. Oracle says it maybe harmful to use these as they are undocumented. What maybe the work around?
About _LOG_SIMULTANEOUS_COPIES:
Oracle say, that this parameter set = CPU_COUNT * 2 automaticaly
(little note: CPU_COUNT from init.ora, but not from equipment of a real box)
If u have >2 CPU, u can increase value of this parameter up to CPU_COUNT * 2 + 2
without any perfomance degradation.
About _LGWR_ASYNC_IO and _LGWR_IO_SLAVES:
1) U should check, is ur OS support async io operation, if yes then this is way to
increase activity of disk operations in lgwr.
2) U should set _LGWR_IO_SLAVES = 1 !!! if u going to change _LGWR_ASYNC to TRUE.