Capacity Planning - The Practice of Cloud System Administration - page 372

18.2.2 Knowing Your Capacity Limits
The capacity limit of any resource is the point at which performance starts to degrade rap-
idly or become unpredictable, as shown in Figure 18.1 . Capacity limits should be determ-
ined by load testing. Load testing is normally performed in an isolated lab environment. It
can be performed by synthetically generating traffic or by replaying production traffic.
Figure 18.1: Response time starts to degrade beyond a certain percent utilization.
For each primary and secondary resource, you need to know the capacity limit. To gen-
erate such data for each resource in isolation, your lab setup should be equipped with lots
ofexcesscapacityforallbutonelow-levelresource.Thatresourcewillthenbethelimiting
factor for the service, and response times can be graphed against the percentage utilization
of that resource and separately against the absolute amount of that resource that is avail-
able. For completeness, it is best to repeat this test several times, each time scaling up the
whole environment uniformly. This approach will enable you to determine if the limit is
more closely related to percentage utilization or to quantity of remaining resources.