Re: Long Checkpoints
"Parker, J" wrote:
>
> How can you tell if you have high LRU contention? (by looking at bufwaits?)
No, I mean spin locks as shown by onstat -g spi |grep -i lru.
My own dbstats script calculates a contention factor in much the same
way that Art calculates the buffer waits, as a percentage of activity.
Here's an extract:
-------------------------------------------------------------------------
# Calculate LRU contention
SPI=`onstat -g spi|grep -i lru |awk 'BEGIN {w=0} {w=w+$1} END {print
w}'`
# Calculate MCF (Marks Contention Factor :)
if [ $BUF -eq 0 ]
then
LC=0
else
LC=`echo "scale=2; $SPI * 100 / $BUF" |bc`
fi
if [ $LC -gt 1 ]
then
W="$MARKER1"
else
W="$MARKER0"
fi
echo "LRU Contention: ${SPI} $LC% $W"
-------------------------------------------------------------------------
Where BUF = pagreads + bufwrits. The MARKER1 is just an arrow indicating
a problem. In this case I look for greater than 1% MCF. I find that
multiple CPU database servers always have some contention, so that's why
I accept 1%. But this is a very crude calculation that you may need to
work on. It works for me. ;-)
Cheers,
--
Mark.
+----------------------------------------------------------+-----------+
| Mark D. Stock mailto:mdstock@mydas.freeserve.co.uk |//////// /|
| http://www.informix.com http://www.informixhandbook.com |///// / //|
| http://www.iiug.org +-----------------------------------+//// / ///|
| |What year 2000 bug? year 2000 bug? |/// / ////|
| |year 2000 bug? year 2000 bug? year |// / /////|
| |2000 bug? year 2000 bug? year 1900 |/ ////////|
+----------------------+-----------------------------------+-----------+