[rrd-users] Details of consolidation functions?

[Simon Hobson]

Jim Eshelman wrote:

>I've read the documentation and Alex's tutorial trying to understand 
>exactly how the AVERAGE, MIN, and MAX consolidation functions work, 
>but I'd like to know more.  Let's say that I have created the 
>following RRD to track uptime of routers:
>
>    '--step=1',  # one second step
>          'DS:snmp_up:GAUGE:3600:0:1',
>          'DS:icmp_up:GAUGE:3600:0:1',
>          'RRA:MIN:0.9999:10:10',       # 10 steps => 10 secs.per 
>consolidated data point, 10 cdp's => 100 secs. 
>          'RRA:MIN:0.9999:30:240',      # 30 secs., 240 half-minute 
>points, =>  120 mins. of data
>          'RRA:MIN:0.9999:300:432',     # 300 secs. per cdp (5 
>mins.), 432/12 = 36 hours.
>          'RRA:MIN:0.9999:1800:336',    # 30 mins. per cdp, 336/2 = 
>168 hours = 1 week
>          'RRA:MIN:0.9999:7200:432',    # 180 mins. cdp, one month data
>          'RRA:MIN:0.9999:86400:425',  # one year data
>
>All the input PDP's (i.e. values entered) are either 1 or 0 (up or 
>down).  As defined above any update value of 0 (down) will cause the 
>active row in each rra to show 0, right?

Wrong I think in the general case. The data is normalised before it 
is consolidated, so you could easily end up with a value of 0.129 or 
whatever unless your updates are all on an exact multiple of the 
step. It is only because your step size is one second that will 
preclude any normalisation.

>By changing the consolidation function to AVERAGE it would show the 
>AVERAGE of all the 1's and 0's entered during that row's active 
>period, is that so?

Yes

>(actually the AVERAGE of all the interpolated data points in 
>the row, not just the ones entered. But with an xff of 0.9999 they 
>should be nearly the same.)  Am I correct in concluding that these 
>rra definitions should be changed to AVERAGE to give a better 
>picture of uptime with these settings?

Dunno, we don't know what YOUR requirements are ! If you want a 
measure of uptime then probably - ie so you can say the router has 
been up for 99% of the time. On the other hand, if you want to 
clearly see any times when the router has been down, then min might 
be a better choice - any period where there has been downtime would 
then show as a clear dip to zero, rather than a dip from 1 to 0.995.

You could always store both.