| Cluster size
|| Call Rate
|| Event Rate
|| TCAP message Rate
| One node
|| 22,500 events/s
|| [One node individual results]
| Two nodes
|| 45,000 events/s
|| [Two node individual results]
- were evenly distributed between all available cluster nodes
- had originating and terminating treatment
- were monitored for the entire duration on the originating leg
- lasted 60s.
50% of calls involved a map query.
Types of graphs
The individual test results include four types of graphs:
||Total throughput (calls per second) of a cluster over the course of the test run. The "ramp-up" period at the start of each run is required because the load generators are Java programs, which need time for the JVM to perform its compilation and optimisation of Java code. (If we did not use a ramp-up period, the load generators would be overloaded initially, which would skew the results.)|
||Distribution of response-time changes during the course of the test run.|
|Response Time Distribution
||Overall distribution of response times from the entire test run.|
||Percentage of CPU used on Rhino servers, as measured by vmstat.|
Calculating number of dialogs
To calculate the number of dialogs (for example, using 5,000 calls per second):
- All calls involve originating and terminating treatment (in our example, this means 10,000 dialogs).
- 50% of calls involve a query to the HLR (2,500 dialogs).
... for a total of 12,500 dialogs per second.