IDS tuning
Dejan Stojcevski asked how to enlarge IDS buffers for a better read/write ratio without lengthening checkpoints. Anup Agrawal suggested tuning LRU_MAX_DIRTY and LRU_MIN_DIRTY to lower values. geoff.poole described a cron script on 7.31 forcing LRU writes (onmode -B) then onmode -c when dirty buffers are low. Head A Lettuce found checkpoints did not block queries on 9.4, unlike 7.3.
Auto-generated by Claude from the posts below — may be imperfect; read the full thread.
Suppose I have an unlimited resourcess in my box (unlimited memory, processors ...). What I would like to do is increase the buffers for IDS in order to achieve better buffer read/write ratio. But at the same time I would like if not to lower checkpoint timings then at least to keep them at the present timings. What should I tune/change in IDS configuration?What is a best approach to achieve this? Dejan.
Dejan,
If you want to increase the buffers and would like to
keep the same checkpoint timing. try tuning these
parameters. what is read/write Êched you are getting
now?
LRU_MAX_DIRTY 10 # LRU percent dirtybegin cleaning limit 2
LRU_MIN_DIRTY 5 # LRU percent dirtyend cleaning limit 1
Thanks
Anup
--- "DEJAN STOJC...."
<dejan.stojcevski@cosmofon.com.mk> wrote:
> Suppose I have an unlimited resourcess in my box
> (unlimited memory, processors ...).
> What I would like to do is increase the buffers for
> IDS in order to achieve better buffer read/write
> ratio. But at the same time I would like if not to
> lower checkpoint timings then at least to keep them
> at the present timings. What should I tune/change in
> IDS configuration?What is a best approach to achieve
> this?
> Dejan.
>
>
__________________________________
Yahoo! Mail - PC Magazine Editors' Choice 2005
http://mail.yahoo.com
We have
such an interesting problem on 2 applications, for the larger
servers we've started forcing LRU-writing and then force-checkpoints
(onmode -B and onmode -c). This is a very wide subject and particular to
versions, I've started to evolve a checkpoint script run every 4mins from
cron with a ONCONFIG check point duration of 5mins. IDS 7.31.FD7. This
prevents long checkpoints which I think you were indicating in your email,
but the idea is evolving .. looking to have different checkpoint stagegies
for different periods during the day. IDS 9 and 10 are a different kettle
of fish!
output() { echo "$(date +'%Y-%m-%d %H:%M:%S') $*" ; }
onmode -Bsleep 15
onstat -R > $D/onstat.R.$DATETIME
MAXDIRTY=$(cat $D/onstat.R.$DATETIME|\\\\
sed 's/%//'|\\\\
awk '($2=="m") { if ($3>max) max=$3 } END { printf "%d\\
",
max*10 }')
output $(grep "hash buckets" $D/onstat.R.$DATETIME|awk '{ print
"Dirty",$1,"Total",$5 }')
((PERCENT=$MAXDIRTY/10))output "MaxDirty $PERCENT %"
if [ $MAXDIRTY -gt 10 ]
then
#===============================================================
# If run onmode -c now might cause extended checkpoint duration
#===============================================================
output "WARNING: no onmode -c run, since maxdirty > 1%"
exit 1
fi
#================================
# Raise require-checkpoint flag
#================================
START=$(date +'%H %M %S')
onmode -c
output "CheckpointDuration " $(echo "$(date +'%H %M %S') $START"|\\\\
awk '{ diff=($1*3600+$2*60+$3)-($4*3600+$5*60+$6)
if (diff<0) diff=diff+3600*24
hours=int(diff/3600)
mins=int((diff-hours*3600)/60)
secs=int(diff-hours*3600-min*60)
printf " d: d: d\\
",hours,mins,secs }')
Geoff Poole
Verizon Florida USA
"DEJAN STOJC...."
<dejan.stojcevski
@cosmofon.com.mk> To
Sent by: ids@iiug.org
forum.subscriber@ cc
iiug.org
Subject
IDS tuning [5742]
09/13/05 03:18 AM
Suppose I have an unlimited resourcess in my box (unlimited memory,
processors ...).
What I would like to do is increase the buffers for IDS in order to achieve
better buffer read/write ratio. But at the same time I would like if not to
lower checkpoint timings then at least to keep them at the present timings.
What should I tune/change in IDS configuration?What is a best approach to
achieve this?
Dejan.
On informix 9.4 my testing showed that checkpoints don't matter. I ran a 1 second query in the middle of a long checkpoint and it still only took 1 second. But this is not true on informix 7.3 (it made me wait). ----- Original Message ----- From: "DEJAN STOJC...." <dejan.stojcevski@cosmofon.com.mk> To: <ids@iiug.org> Sent: Tuesday, September 13, 2005 2:18 AM Subject: IDS tuning [5742] > Suppose I have an unlimited resourcess in my box (unlimited memory, > processors ...). > What I would like to do is increase the buffers for IDS in order to > achieve better buffer read/write ratio. But at the same time I would like > if not to lower checkpoint timings then at least to keep them at the > present timings. What should I tune/change in IDS configuration?What is a > best approach to achieve this? > Dejan.