Application profiling is a critical step in QoS solution design of today's IP networks. To aggregate applications into QoS classes effectively, one has to know the key characteristics of the applications. This is needed to arrive at a QoS solution design that meets the QoS targets for all applications, while allowing for maximum use of the network link capacities. We collected NetFlow data at several locations on a corporate intranet and analyzed the data producing flow, packet and byte application breakdowns. For the most popular and bandwidth-consuming applications, we analyzed the stochastic characteristics like distributions of flow lengths, packet sizes, throughputs, etc. Our study reveals that the measured traffic composition does not vary much on a day-by-day basis, but that it can be very different from location to location. Therefore traffic measurements have to be collected at each congested link in the network, since singular measurements may lead to inaccurate assessments of the live traffic mix, resulting in ineffective QoS solution designs. Also, to account for changes in user behavior, in- or decrease of the user population and changes in the application mix, the process of traffic profiling should be repeated every 3-6 months.
[1]
Azer Bestavros,et al.
Self-similarity in World Wide Web traffic: evidence and possible causes
,
1996,
SIGMETRICS '96.
[2]
Matthew Mathis,et al.
The macroscopic behavior of the TCP congestion avoidance algorithm
,
1997,
CCRV.
[3]
Anja Feldmann,et al.
Efficient policies for carrying Web traffic over flow-switched networks
,
1998,
TNET.
[4]
Daniel P. Heyman,et al.
IP traffic characterization for planning and control
,
1999
.
[5]
Peter B. Danzig,et al.
Characteristics of wide-area TCP/IP conversations
,
1991,
SIGCOMM 1991.