On Tue, May 31, 2005 at 08:31:16AM +0800, Vernon Everett wrote:

I just noticed something interesting with disk graphs. Last night, one of my disks hit 99% (for a few minutes) which made me look more closely at the history graphs for the disk in question. The 48 hour graph shows Max to be 99% - No worries there. The 12 day graph shows it to be 95.5% The 48 day only 65.3, and the 576 day, a tiny 39.9% However, every night, since I started monitoring, this disk hits over 95% for a few minutes. (An overnight process uses the space then releases it)

Have I found a bug with Hobbit/RRDTool, or am I doing something silly?

It's not a bug, but it's a common misunderstanding about how RRDtool handles data for the graphs.

The min/max values you see on the graphs are not absolute min/max values. RRDtool doesn't keep any record of these.

What RRDtool does is to compute an average value for each data point, an average over the amount of time each data point represents in the graph. And the min/max values you get are the lowest/highest value of that *average*.

Each graph has 576 data points. For the 48-hour graph this means a data point is the average over 5 minutes - which happens to match the frequency that most of the Hobbit tests run. So the 48-hour graph more or less matches the absolute measurements done (not quite, because the tests can run at intervals that are not exactly 5 minutes). The 12 day graph averages over 30 minutes, the 48 day graph over 2 hours, and the 576 day graph over 24 hours.

So if your disk is 35% full all of the time, except for 5 minutes a day, then the 576-day graph will come out as being completely flat - because the average value *over the 24-hour period* will be constant. And you'll see a max-value of (5*95 + 1435*35)/1440 = 35,2 % (24 hours = 1440 minutes).

Regards, Henrik