The table below shows PASOE load testing on several different Amazon AWS machine sizes. This should be used as an example and a baseline for sizing your application. However, since your application will
NOT be the same as the one in these tests, you should size your application using the
PASOE - Sizing Your Machine (ATM Benchmark) article.
Testing application
This test used a modified version of the ATM test. The ATM test has been around for over 15 years at Progress. The test:
- Emulates an Automatic Teller Machine.
- Updates account, bank, teller and deposit tables to an OpenEdge database.
- Is completely state-free using the WEB transport.
- Runs PASOE on an isolated system. The database and test driver will run on two other machines.
- Used Amazon AWS machines in the same region.
The testing was done using JMeter from a windows machine. PASOE resided on different size Linux machines and the database was on a separate Linux machine. The database was scaled large enough that it and the machine it was running on never experienced a bottleneck (memory, CPU, and record contention were not issues).
The parameters
The test turnaround time was kept to around 1/10 of a second. Testing showed if there was a bottleneck in CPU or memory (the elapsed turnaround time increased dramatically). Your application may have a longer turnaround time, so start with minimal load, get a baseline of best turnaround time and increase load until you see a marked increase in time.
| Amazon AWS Machine (Linux) | c3.large | c3.xlarge | c3.2xlarge | c3.4xlarge |
|---|
Virtual CPUs (CVPU) Memory (RAM) | 2 3.75 Gb | 4 7.5 Gb | 8 15 Gb | 16 30 Gb |
| | | | | |
| Concurrent Clients | 50 | 125 | 250 | 500 |
| | | | | |
| Load Average | 1.42 | 2.95 | 9.11 | 16.5 |
| | | | | |
| Total Responses | 14,174 | 35,650 | 71,371 | 139,948 |
| | | | | |
| Average Elapsed Time | 113 ms | 103 ms | 99 ms | 95 ms |
CPU seemed to be the other large indicator of performance degradation. When the load started to equal the number of CPUs then the elapsed turnaround time increased immediately.