A Hybrid Architecture - Scalable Continuous Media Streaming Systems - page 346

arrival rate = 0.5/sec
arrival rate = 5/sec

a a

1.2

1.1

1

0.3 0.4 0.5 0.6

Proportion of Dynamic Channel

N = 20
N = 30
N = 50
N = 20
N = 30
N = 50

N = 20
N = 30
N = 50
N = 20
N = 30
N = 50

Figure 19.10 Effect of channel partitioning on latency

19.4.2 Channel Partitioning
To investigate the performance impact of different channel allocations, we conducted simu-
lations with the proportion of dynamic multicast channels, denoted by r , ranging from 0.3 to
0.7. The results are plotted in Figure 19.10. Note that we use a normalized latency instead of
actual latency for the y-axis to facilitate comparison. Normalized latency is defined as

w ( r )

(19.18)

min { w ( r ) , ∀ r }

where N dynamic multicast channels.

We simulated three sets of parameters with N

w ( r ) is the latency with r ×

30, and 50 for two arrival rates, namely,
heavy load at 5 requests/second and light load at 0.5 requests/second. Note that normalized
latency obtained from two different values of N cannot be compared directly as the denominator
in equation (19.18) is different.
Surprisingly, the results show that in all cases the latency is minimized by assigning half
of the channels to dynamic multicast and the other half to static multicast. For comparison,
UVoD exhibits a different behavior and requires more channels allocated for static multicast
to minimize latency at high loads as shown in Figure 19.11 for a 50-channel configuration.
UVoD's behavior is explained by the observation that at higher arrival rates, the waiting
time for a free unicast channel increases rapidly near full utilization. Therefore, it is desirable
to allocate more multicast channels to reduce the traffic intensity (arrival rate

= 20 ,

× T R ) routed to