0) Introduction
When setting up an OpenShift environment for IBM Cloud Pak for Business Automation installation the question can arise, can my environment deal with the needed resource requirements?
An option to get a quick result can be a simple benchmark. With such a benchmark it is also possible to compare different environments. There exist a number of different benchmarks which can be considered for doing such a task. I will introduce here a few benchmarks which I think could be beneficial in a number of situations.
1) Some available benchmarks
The first thing which mattered to me was a quick and easy setup. Besides this you might have specific needs, e.g. you want to test your network speed or your disk performance as part of your OpenShift setup. Therefore we will also have some special benchmarks which focus on specific tuning areas. One also needs to mention that much more benchmarks exist (the ones listed here is a small selection, which can be extended in the future). It is possible to adjust or write complete own test cases.
1.1) K-Bench
A benchmark which can be easily installed and executed is the
K-Bench benchmark. With only 3 commands it is possible to get a result:
- Clone the Git repo
- Run install.sh script (Installation)
- Run run.sh script (Basic test execution)
1.2) Network performance benchmark - ripsaw
When you want to measure your network performance you can consider the
ripsaw benchmark. The installation is described in the linked article. Main focus here is the network performance.
1.3) Disk benchmark tools - etcd performance with fio
You might already know the fio tool for disk performance measurements. In this example we will measure the disk performance for the etcd. A good disk performance is
a necessary requirement for the etcd. What is required to perform the measurement is described in the
Red Hat support article.
2) Examples and comparison of the benchmark results
How does the results look like and what can you learn from the data collected? We will have a look at some examples from our earlier described test procedures.
2.1 Example data for the K-Bench benchmark
With the easy to install K-Bench benchmark performance data can be easily generated. In the following table we will have a look at the Service API performance:
|
System1 |
|
|
|
System2 |
|
|
|
| Activity |
median |
min |
max |
99% |
median |
min |
max |
99% |
| update service latency |
14.306 |
6.547 |
25.634 |
25.634 |
4.107 |
3.458 |
20.141 |
20.141 |
| delete service latency |
55.017 |
31.792 |
73.294 |
73.294 |
4.023 |
2.616 |
7.814 |
7.814 |
| create service latency |
74.069 |
74.069 |
115.575 |
115.575 |
8.389 |
6.714 |
11.684 |
11.684 |
| list service latency |
4.978 |
3.28 |
14.772 |
14.772 |
49.601 |
19.708 |
61.809 |
61.809 |
| get service latency |
4.383 |
3.161 |
6.467 |
6.467 |
57.478 |
16.765 |
90.711 |
90.711 |
You can see two different systems, as you see System1 shows larger latency values than System2 for write intensive activities while System2 shows the reverse for read intensive activities. The benchmark results provide the different statistics information for the executed tests, like min and max values etc.
2.2 Example ripsaw data
With the ripsaw benchmark we showed earlier it is possible to measure the network performance. You can see the network throughput numbers in the below screenshot:
2.3 Example etcd disk performance
As mentioned at the beginning a good disk performance is critical for a proper working of etcd. Therefore we have seen a testcase based on the fio tool. Here we will now see an extract of the result output. The final output shows that the disk performance is sufficient for etcd. Without proper checks it can be difficult to debug issues caused e.g. by slow disk performance.
3) Outlook
For the future we plan to provide some more guidance how to tune the system and use the benchmark results to determine if the environment is sufficient to handle Cloud Pak Business Automation workloads.