Re: redo copy "sleep rate" is this a problem?

The log buffer size is measured in bytes. Log writes are padded to whatever Oracle believes the device block size is - which is usually 512 or 1024 bytes, but I have seen 256 bytes on one platform. You can determine the size for your system by calculating:

    (redo size + redo wastage) / redo blocks written.

The answer will be about 20 bytes short of an 'obvious' size (i.e. power of 2).

For a web-based OLTP system I would be inclined to assume that your log buffer size was too small. I would have set it to somewhere between 32K and 128K as a starting point. Your comment about long service waits sounds like a mis-understanding of how LGWR uses the log buffer - though for systems with very large transactions it is possible to make the log buffer too big (which in that context you sometimes see at 10Mb+).

The problem with your interpretation of the script is that (as with so many useful monitoring scripts) you have to know which rows to ignore.

The redo copy latches are almost always taken in 'immediate mode' because if the acquisition fails, there's always another latch to try for. For redo copy latches, compare the 'immediate_gets' figure with the 'gets' - you should find that the former is vastly larger than the latter, and if it is then you can more or less ignore the sleep rate stat on it.

--

Jonathan Lewis
Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk

Greg Stark wrote in message
<87d7t48fvu.fsf_at_HSE-Montreal-ppp19485.qc.sympatico.ca>...
>
>"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> writes:
>
>> If you've got the script you might also have
>> the book. Page 29 points out that the sleep
>> rate on the redo copy latch is always high
>> since redo copy is rarely taken in 'willing to wait'
>> mode.
>
>I didn't but I found a store that had it so I do now.
>
>So what's a normal value for this latch? Because our redo log buffer is
indeed
>rather small (8192, my docs say it's in blocks but that doesn't sound
>possible, it must be bytes, right?) I want to raise it but I'm worried it
will
>cause long service waits as LGWR cleans up lots of redo logs. This is an
OLTP
>DB, a web site in fact, so having a 1-2 second pause for all connections
could
>really wreak havoc.
>
>--
>greg
Received on Sun Nov 21 1999 - 03:53:49 CST