Spike Rates at Peak
This section describes the spike rates at peak.
For example, in a spike case, 20 percent of users attempt to begin the call model (as described in Table 8) within 10 minutes from the start of day. Their focus is primarily on tasks 1 through 4, assuming that tasks 5 through 12 will even out over time as subscribers perform other operations.
Oracle transactions = 3
NDB hits = 7
RADIUS transactions = 8, TTLS =1, FA = 1, and accountings = 2
200,000 sessions over 10 minutes = 20,000 sessions per minute, or 167 sessions per second
This means you need to do 501 Oracle transactions, 1169 NDB hits, 167 TTLS, 167 FA, plus 334 RADIUS accountings per second.
This example is within the recommended base configuration guidelines for SBRC, but you can easily see that spike rates of shorter duration (for example, a GGSN reconnecting all its users), a larger subscriber base, or a higher than expected percentage of users attempting to connect at the same time (for example, call traffic generally spikes after notable public events) could easily push the rate even higher. Establishing a realistic call model and planning beforehand as to what limits you to consider the worst-case peak load for your use case will give you the flexibility you need when establishing and maintaining a reliable and performing solution.