[rrd-users] Temperature Monitoring

[Steve Shipway]

OK - point taken - this is what I get for posting quickly.

I was only considering a simple RRA where 1cdp=1dp, so you can do an on-the-fly calculation over a range on a calculated value.  Slow, but it works.

The issue hits when you're working on a rolled-up RRA where 1cdp = x dp where x>1, since then you get

max ( a-b ) = max ( avgX(x) - avgX(b) ) 

and you're stuffed by a greater degree as X increases.  I should have remembered this as it is what messes up my 95th percentile calcs on RRD files that only have MIN/MAX and AVG RRAs as the granularity decreases.

I still favour storing the a and b values for future use, but to get max(a-b) accurately over rolled-up RRAs you'll need to store a-b as well.

Steve

Steve Shipway
University of Auckland ITS
UNIX Systems Design Lead
s.shipway at auckland.ac.nz
Ph: +64 9 373 7599 ext 86487