I posted to the list some time ago about problems scaling Gmetad. I
found that after passing some number of monitored nodes I was showing
nodes failed.
I've finally worked out the issue... my disk was the bottleneck. It
appears that Gmetad serially updates RRDs for a node/cluster. Unable to
tune around this I decided to use tmpfs with a special startup script
and cron job that rsync's the data out of tmpfs every 5 minutes.
Here's my startup script (executed by SMF on Solaris):
#!/usr/bin/bash
## SMF Start method for Gmetad
## -benr
if ( mount | grep "/opt/ganglia/data on swap" >/dev/null )
then
echo "Tmpfs already mounted."
else
echo "Mounting tmpfs..."
chown nobody /opt/ganglia/data
mount -F tmpfs -o size=100m,noxattr swap /opt/ganglia/data
fi
## Now sync the data in if its empty:
if [ -d /opt/ganglia/data/__SummaryInfo__/ ]
then
echo "Cache primed, ready to start."
chown -R nobody /opt/ganglia/data
else
echo "Priming the cache..."
/opt/csw/bin/rsync -at /opt/ganglia/data-disk/ /opt/ganglia/data/
chown -R nobody /opt/ganglia/data
fi
## Finally, start ganglia:
/opt/ganglia/sbin/gmetad
#====================================
The cronjob to sync is simple:
# Sync gmetad ram-disk to physical-disk
#
5,10,15,20,25,30,35,40,45,50,55,0 * * * * /opt/csw/bin/rsync -at
/opt/ganglia/data/ /opt/ganglia/data-disk/
The question I have for the list is... has anyone else run into this
bottleneck? If so, how did you solve the issue? I know several people
have multiple gmetad's out there, perhaps unnecessarily. CPU usage and
disk consumption are low, its just doing a lot of IO. My data/ dir is
only 27MB.
benr.
-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference
Don't miss this year's exciting event. There's still time to save $100.
Use priority code J8TL2D2.
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone
_______________________________________________
Ganglia-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-general
