> I have one problem.  When the server application is restarted, the
> counters are all zeroed.  This causes the RRD to assume that the
> counter has rolled over at some specific word size and compute the
> average throughput based on that assumption.  Since the server never
> gets anywhere close to filling the word size of an unsigned long int,
> that means the RRD's assumptions are way, way off.  This shows up on
> the graph as a huge ugly spike which is wrong and makes it look like
> my server's rate control is faulty.
> 
> When I create the .rrd file, I use 
> 
>    "DS:readbytes:COUNTER:600:0:1250000",
> 
> for each client because the server is connected to a 10bT ethernet,
> and 1250000 Bytes per second is the maximum theoretical throughput.
> This limits the spikes to 10Mbps, but since the actual data is far
> closer to 1Mbps, the graphs still look horrible.  MRTG never had this
> problem.
> 
> Is there a work around?
> Dave

To be exact, you are using a workaround right now.

The proper way to handle this is to log an "U"nknown value.
Assume you take a sample and the counter is 12345,
next time you take a sample the counter is reset and it is 123.
What do you know ? Nothing, apart from the reset. You don't know at
which point you reset. Was it at 12356 ? 12367 ?

Your front end should notice the reset and insert the U value.

Regards,
Alex

--
* To unsubscribe from the rrd-users mailing list, send a message with the
  subject: unsubscribe to rrd-users-request@list.ee.ethz.ch