The new metric for video QoS, specified in Equation 1,
also depends on the resource consumption (RC), as illustrated
in Section 3.2.
We consider four factors in determining the resource overhead, i.e.,
CPU utilization (cpu), network bandwidth utilization ( 
),
buffer utilization (Buf) and
storage bandwidth overhead ( 
).
Benefit functions for each parameter can be used to indicate
which resources may be more critical based on the design of the system.
An interesting observation is that the resource cost must consider the current set of available resources. As resources get used up to service requests, they become more critical and resource cost becomes more dependent on the fraction of the available resources utilized. When resource availability is low, admission control mechanisms must allocate resources to the high priority tasks.
We characterize the resource consumption of a server 
due to a request
by a ratio that captures the utilization of resources, similar
to the load-factor measure in [27].
is proportional to:
where 
is the resource needed by request and 
is the
resource available on server .
Different requests have different resource requirements and the RC factor
may vary from one request to another.
Lower the RC value, greater is the capacity of the system to service
additional requests.
Hence, the resource cost component of the QoS metric is essentially the cost
of the bottleneck resource, since this constraining resource measures the
degree of QoS delivered.
