Size Load Generators
The NeoLoad Load Generators code is optimized for multi-core and multi-processor environments. For example, with a machine with four cores, it is vital to make sure the rest of the machine is adequately dimensioned in order to properly use the available CPU capacity. Regarding memory, 1 Gb per core is a good compromise; in this case, 4 Gb of allocated memory is ideal. It is necessary to add to that a 64-bit operating system (to be able to use more than 1.5 Gb of memory), which makes 6 Gb of physical memory. In the case of network adapters, a good compromise is to have one adapter for two cores.
For a four-core machine for example, an appropriate set-up can be 6 Gb physical memory, a 64-bit operating system, 4 Gb memory allocated to the NeoLoad Load Generator and two network adapters.
The number of Virtual Users which a machine can handle depends on many aspects:
- Complexity of the application to test
- Application server properties
- Network connection capacity
- etc.
Starting a ramp-up test of the application with one Load Generator makes it possible to benchmark the Load Generator with that application. In the NeoLoad Monitoring data, thresholds on the CPU and on the memory of the Load Generator machine help identify the maximum number of Virtual Users for the Load Generator.
For more information about ramp-up tests, see Load variation policy. For more information about monitoring thresholds, see Set and configure a counter alert threshold.
Identifying the number of Virtual Users that one Load Generator can handle and the number of Virtual Users for the test, it is easy to calculate the number of Load Generators necessary to run the test. To pinpoint the load limits of a Load Generator, it is necessary to watch its behavior under the ramp-up test load, and specifically:
- CPU
- Memory
- Throughput
- Requests rate
- Virtual Users load
- Tip: By default, NeoLoad sets thresholds when 80% and 90% of the Load Generator CPU and memory are used. When the CPU or memory figures exceed the thresholds for 15 seconds or more, an alert is raised.
The first significant breaking point in the CPU, memory, throughput or hit rate metrics represents the performance limit of the Load Generator. This point must be correlated with the number of Virtual Users generated. Beyond this number of Virtual Users, serious issues may well occur with the Load Generators.
For example, a sizing test is started with one Load Generator. The ramp-up test adds 1 Virtual User every second. The Graphs tab in the Runtime section or the Graphs tab in the Results section shows that the throughput curve falls down dramatically at 3 minutes and 48 seconds while the number of retransmitted segments increases. The network issues lead to a shift with 228 Virtual Users (1 Virtual User per second for 228 seconds). The size limit of the network infrastructure allows 228 Virtual Users. To test the application with a large number of Virtual Users, it is probably necessary to modify the infrastructure.
- Tip: A 20% to 30% security margin out of the sizing limits is recommended for the Load Generators in the real heavy load test.