[rrd-users] beginner questions on multiple series

[Alex van den Bogaerdt]

On Fri, Jul 04, 2008 at 11:28:30AM +0200, SecretCode wrote:
> Thanks Raimund, Alex and Jason (off list)!

I received it on-list and reply on-list.

> I initially want this for my home network, a mere handful of addresses, 
> which shouldn't be a size or processing problem (obviously I can't keep 
> per-minute data for a year).

Why not?  365 days times 24 hours times 60 minutes = 525600 rows.
8 bytes per DS per row, plus some bytes in the header. That's
a small file of just over 8MB per address if it has two DSes and
525600 rows:  365*24*60*8*2=8409600.

Am I missing something, did I screw up the computation or isn't
it as obvious as you state?

> per-minute data for a year). But I'd like to extend it to the office 
> environment which is much larger. But not enterprise-class, no.
> 
> Alex: you mentioned having several buckets in one RRD. Since I'll need 
> to allow for multiple rrds anyway, is there any advantage in having say 
> 3 rrds with 10 addresses in each rather than 30 rrds? (Which would be 
> easier to manage.)

You asked: "... have to be a separate RRD file". The answer is: "no."
But that doesn't mean it isn't the most logical thing to do.

Unless you really need to worry about I/O, memory and/or CPU, just go
for the easy method of having one RRD per address.

Only a careful design could benefit from minimizing overhead by
keeping ip addresses together.  And such a design could backfire, for
instance if you only need to have data from one of the DSes, RRDtool
would need to read all 256*2 DSes to present just that one to you.

Most of the time it will be cheaper to add hardware instead of brains.

This said, a bad design cannot be fixed by adding more hardware, so
do think before you act.

-- 
Alex van den Bogaerdt
http://www.vandenbogaerdt.nl/rrdtool/