"Charles Hooper" <hooperc2000_at_yahoo.com> wrote in message
news:1175908606.190716.85950_at_w1g2000hsg.googlegroups.com...
> On Apr 6, 7:15 pm, "Gerry Sinkiewicz" <sinki..._at_snet.net> wrote:
>> "Mladen Gogala" <mgogala.SPAM-ME...._at_verizon.net> wrote in message
>>
>> news:pan.2007.04.05.20.01.31_at_verizon.net...
>>
>> > On Thu, 05 Apr 2007 05:42:39 -0700, Charles Hooper wrote:
>>
>> >> log buffer is a little less than 100MB
>>
>> > Allow me to express my gratitude in the name of Kingston Technology,
>> > world's leading RAM manufacturer.
>>
>> > --
>> >http://www.mladen-gogala.com
>>
>> You know I am looking for an opinion on log buffer size.
>> I have heard of a software vendor that suggested a 500MB log buffer,
>> yikes.
>>
>> Personally I use a 2MB redo log buffer and about 20 redo-log groups at
>> 50MB
>> each for a busy OLTP oriented
>> instance.
>
> The recommendations that I have seen in a couple performance tuning
> books and a couple articles suggests that a log buffer of 0.5MB to 1MB
> is typically sufficient to good, and much larger than 1MB is a waste
> of memory. On 10g, even if you specify a 1MB log buffer, Oracle may
> actually allocate nearly 16MB for the log buffer. Frequent commits in
> the database (in any session's transaction) mean that only a small
> portion of the log buffer will be used if the log buffer is large -
> the OP's system was committing 1.3 times per second and sending an
> average of 1,703 bytes at a time to the redo logs, so most of the
> 100MB allocated for the log buffer was wasted. Data in the log buffer
> only stays in the log buffer a short amount of time (3 seconds or
> less) before it is flushed to disk.
>
> In my database, a 50MB redo log file is too small, but it may be fine
> in your database. Ideally, the redo logs should switch roughly every
> 20 to 30 minutes during times of typical to heavy activity. If Oracle
> is switching log files every 1-3 minutes, then you might want to
> reconsider the size of the redo log files (each switch causes a
> checkpoint). During batch processing runs, my database will burn
> through a 512MB log file in a couple minutes; since the batch
> processes only occur a couple times a day, I force log file switches
> at least once an hour. Take a look at the frequency of log switches
> in your database, and use that to determine the most appropriate size
> for the redo logs.
>
> Charles Hooper
> PC Support Specialist
> K&M Machine-Fabricating, Inc.
>
Thanks, I do have a couple of instances that do archive logs frequently at
times.
I will re-evaluate those redo-log groups and make them larger.
Received on Sat Apr 07 2007 - 08:33:17 CDT
