[rrd-users] Collect on one machine, graph on another

Simon Hobson linux at thehobsons.co.uk
Tue Jan 22 09:10:06 CET 2008

>At the moment I have one machine that both collects network traffic
>data and serves up graphs. We're about to get a network upgrade and I
>want to take the opportunity to split the functions - the graphing is
>"useful", the traffic routing is "essential". I've had a couple of
>problems in the past with the graphing consuming all the VM available
>and almost killing the machine, and I have plans for even more
>complex graphs !
>Applying the KISS principal, I was thinking of storing the rrds on
>the gateway, exporting the directory via NFS, and mounting it on the
>web server.
>1) Are there any issues with doing this ? Any locking issues between
>updates and reads ?
>2) Any better suggestions ?

Just to follow up, I've now got this in testing and it does indeed 
appear to work fine. I had several responses off-list - thanks guys. 
Interestingly, performance is not significantly affected - it adds a 
few seconds to drawing each graph, but the calculation/drawing phase 
is significantly longer than the data retrieval phase anyway.

Other suggestions either involve copying the database from one 
machine to the other as required, or using the rrdsrv to pass data 
across. rrdsrv wouldn't work in this case since it means either 
leaving the graph generation on the data collection machine 
(something I am specifically looking to avoid), or having the data 
logged on the graphing machine (which means a lot of extra work to 
cater for network or system downtime (it's due some serious upgrading 
once it's no longer a critical component) on the graphing machine 
without losing data).

More information about the rrd-users mailing list