Internet traffic modeling has typically focussed on high
quality measurements such as packet traces, and more
recently flow level measurements. However, these
measurements are rairly available on any but a limited
basis, both in time, and across the network.
SNMP traffic measurements, on the other hand, do not have
the same detail, but are ubiquitously available, and
relatively easy to collect. Further, this is the type of
data that operations typically use to inform decisions, and
so we know that interesting information can be garnered from
SNMP data. To exploit this data we need to overcome the
limitations of SNMP. This talk will describe some of the
basic ways that one can process SNMP data to obtain useful
information, for instance, wavelet based analysis.
We have used these techniques on more than one years worth
of SNMP traffic data covering a large tier 1 ISP with around
1000 routers, and on substantial quantities of broadband
access traffic data. The results give us great insight into
how traffic should be modeled over large scales (in time and
space), the effects of multiplexing traffic, and how smooth
backbone traffic really is.
Copyright. All Rights Reserved.