The Persistency Metric

In a complex session we want and need to have continuous activity throughout the
session. We also want to increase the length of the session—the persistency metric.
Figure 14.6 shows an initial channel allocation at (a). Then successive channel additions (b), (c), and (d) as the session progresses and session complexity/session amplitude
increases. Each channel is variable rate. Each channel addition implies a step function
increase in billable value.
We can buffer to clip the amplitude peaks in the information rate, but this destroys
some of the properties of the offered traffic. If we wish to preserve the properties of the
offered traffic, we need to match the information envelope to the physical layer. For
example, (b), (c), and (d) in Figure 14.6 represent additional OVSF code allocations,
each of which are variable rate (session-based static and dynamic rate matching).
Because we are not using buffering, we can describe this as a conversational complex
content exchange. This is expensive bandwidth requiring radio and network resources
to be matched in near real time (every 10 ms) to the information rate. Because the
cost of delivery is high, we would hope that we could obtain a commensurate tariff
premium—billing by session length and session amplitude value (session complexity).
Consider also the impact of buffering.
 If we buffer the session (that is, downgrade the session to be streamed or interactive),
then we are introducing delay and delay variability. We have reduced
the value of the session. This may or may not be important to the use or the
user’s application.
 If we do not buffer the session, we have to overprovision radio and network
bandwidth resources, that is, provide sufficient throughput to support the
offered traffic at all times throughout the session. The use of dynamic matching,
however, minimizes the bandwidth overhead (though it does imply substantial
signaling overhead).
Even when we buffer, we will still end up using the same amount of transmission
bandwidth but extended over time. Another way of looking at this is to consider that
by overprovisioning delivery bandwidth, we can minimize or completely avoid the
use of memory bandwidth.