Position Statement for
NSF Workshop on Internet Statistics Measurement and Analysis
Based on my experience as a vendor and now
as a provider, I can suggest several areas
that I think measurement can help out in a
critical manner.
- capacity planning:
What measurement can be used to predict
capacity shortage? Looking at average
load does not help too much because of
packet bursts. Often, one has to go up
to higher bandwidth lines, and/or faster
router cards, and/or larger buffer space,
when the actual average is still well
below the equipment rated capacity.
A provider can't afford to lose business
by losing packets, or by spending too much
money too fast.
However it is very difficult, or expensive
to be able to measure bursts. There may
be an indirect way to project on the burst
behavior which would require coorporation
between vendor and provider. I thought
about this when I was in IBM. I can elaborate
on the details later.
- fault prediction
Perhaps no one can say exactly when a problem
will occur, but a model predicting a probability
would be feasible. At least some idea how likely
a problem can occur will be very useful to a
network provider. I don't even know what mode
to use for this. I know measurement has to be a
part of the methodology. Maybe measurement can
help in finding a right prediction model?
- security
We have heard so much about things like spoofing
attacks, etc. Do people really have hard data
evidence? Are people interested in seeing such
data? Can such data be easily measured?