[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [IPFIX] Fwd: [RFC 3917]Overload & Reliability requirements of Metering process

Sujay,


IMO it is safer to call the period as an outage period.
But then if the reset period is short, and the collector has
sufficient data points it may very well model and interpolate the missing data.


You mean it will guess. That wouldn't be appropriate for many applications.



And if it does this it may as well choose
to ignore the partial flows residing during the reset period, even if
the switch attempts to send it.
My point is that it might be difficult for the collector to know which flows are partial, since those flows may have started before or during the outage and may be exported during or some time after the outage.
So, the implication is that the collecting process builds a model, receives notification of a monitoring outage, decides that certain flows are lower than expected, and adjusts them to fit the model.
If you propose sending such an outage notification, I'd be very interested to see a proposal for how the CP should model the EP - ideally backed up by some experimental evidence. 

Cheers,
P.

--
Paul Aitken
Cisco Systems Ltd, Edinburgh, Scotland.