[rrd-users] Re: FW: Overflow question

Alex van den Bogaerdt alex at ergens.op.het.net
Mon May 31 03:35:37 MEST 2004


On Mon, May 31, 2004 at 09:14:02AM +0800, Hwee Khoon, Neo wrote:

> I have a device that uses 32-bit to indicate the amount of traffic passing
> through it at a particular time. Unfortunately, extremely high traffic
> causes it to overflow, giving me a much lower value that what is expected.

This happens when a counter roll-over occurs and the value you read is higher
than the previous one.  Example:
   this:  100
   next:  200
It could also have been:
   this:  100
   next:  200 (but with a roll-over of 2^32)

   rrdtool calculates (200-100)/(next-this) = 100/300 bytes per second
   in both cases, you will see a gap, not a spike, in your graph.

To work around: poll more often.

> This has result in a series of spike in the rrdgraph during peak hours.

Spikes occur when rrdtool finds a lower counter value than it should:
   this:  200
   next:  100
rrdtool calculates (100-200)/(next-this), finds that it is negative and
decides this can only occur when a roll-over has occured; like this:
   this:  200
   next:  100, but with an implied roll-over of 2^32

   rrdtool calculates (2^32+100-200)/(next-this) = mucho data rate

If the increase from 200 to 100 does NOT occur due to high traffic,
the only other option is that for whatever reason the counter resets
inbetween two polls.  200 -> reset -> add 100 -> 100

To work around:  use DERIVE and make sure to read the manual page for
rrdtool create.

HTH
Alex
-- 
I ask you to respect any "Reply-To" and "Mail-Follow-Up" headers.  If
you reply to me off-list, you'd better tell me you're doing so.  If
you don't, and if I reply to the list, that's your problem, not mine.

--
Unsubscribe mailto:rrd-users-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-users-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-users
WebAdmin    http://www.ee.ethz.ch/~slist/lsg2.cgi



More information about the rrd-users mailing list