Traffic Distribution

You might expect that as traffic streams aggregate together (that is, multiple traffic
streams from multiple users), the traffic streams would smooth. However, we have just
said this doesn’t happen in the handset, so why should it happen when we aggregate
traffic together from lots of handsets? Specifically:
 The merging of traffic streams does not necessarily result in traffic smoothing.
 Bursty data streams aggregated together may produce even more bursty data
streams.
For some years, we have known that data is different from voice. In August 1996,
Jack Scanlon, a well-known and respected Motorola executive, announced at an analyst’s
briefing: “Network design tools today are voice-related, not data—data is different.”
He was absolutely right (and he was talking specifically about wireless
networks). However, the equivalent lightbulb moment today is the realization and
recognition that multimedia is different from data. Networks do not always behave as
expected. Behavior is not consistent with traditional queuing theory. The traffic properties
are different, and the traffic distribution is different.
Traffic distribution is a function of file size and session length. Traffic properties are
a function of error and delay sensitivity (including sensitivity to delay variability). If a
user has a big file to send and is going to be online a long time and the traffic being sent
is delay- and error-sensitive, then by definition, the user will be bandwidth-hungry
and will require a disproportionate allocation of radio and network bandwidth quantity
and quality.
For a group of users, you can then assume that a small but significant percentage of
users will have a disproportionate amount of RF power and network bandwidth allocated
to them. This is known as heavy tailed traffic distribution. Traffic distribution and
traffic properties directly influence radio bandwidth and network bandwidth provisioning
(because bursty bandwidth remains bursty into the network core).

Traffic properties and session properties are interrelated. Session persistency determines
traffic distribution; session complexity determines traffic property. Session
properties are determined by application software.
The job of application layer software is to increase session length and session complexity.
Session length and session complexity together produce heavy tailed traffic
distribution, which in turn drives the requirement for radio and network bandwidth
quantity and quality. This is an important point. If someone asks you how much radio
bandwidth you need, you can only answer the question if you have some idea of what
the future traffic mix will be in terms of offered traffic distribution and required traffic
properties—ditto with network bandwidth.