Salut Cyril,
On Wed, Jan 11, 2012 at 08:32:30PM +0100, Cyril Feraudet wrote:
We use collectd at work (French Bank) to collectd traditional
performances of actually 4275 servers. We use also Collectd to
retrieve
several information about batch, power consuption, etc by using
custom
"types" and "notifications". The easier way is to use unixsock
plugin
to send custom statistics and/or notification. About collect
interval
you can pick them each second if necessary ...
I'm interested in sharing experience with large setups.
I guess you're using the network plugin to send data to some collectd
servers. How many of these do you have? If you're using RRD files,
how
many
per server? Do you use custom RRAs? rrdcached?
In our case, we monitor 2015 clients, sending data to 4 servers with
100k
RRD files each. We're currently evaluating rrdcached performance with
respect
to plain rrdtool. The load average of the 4 servers is between 1 and
2,
depending on the number of clients (between 300 and 600). They
average
between 500 and 1000 I/Os per second. Total network throughput
(collectd-network) is just above 1Mbyte / second.
Hi,
We use rrdfile and database (for the last value with plugin
write_mysql).
About RRA, we keep data at 1 minute interval for 1 year, then each 5
minutes
for 3 years.
Some numbers :
- 5,7 Mo by rrd
- 470000 rrds
- 3To of rrd
- Less than 700 iops
- 4 instances of collectd on the same server
- Rrdcached wrote ~ 7000 datasets/s
- 4MB of bandwidth
Home made write_mysql plugin store the last value for each metrics,
then a batch
perform some calcul to have cpu / groups of servers for example, load,
power
consumption, etc .
Here our home made GUI. I'll be able to share it in GPL soon.
http://feraudet.com/collectd/
Cyril
_______________________________________________
collectd mailing list
collectd@verplant.org
http://mailman.verplant.org/listinfo/collectd
