The sustained CPU endurance test runs a cpu intensive task in a loop for 24 hours. As maxing out the cpu might be considered abusive, the tests aims to use no more than 50% of each cpu core on the VPS.
The purpose of this type of test is
- first to evaluate the speed to the VPS,
- second to evaluate the stability over time of the cpu available to the VPS
- and third to detect VPS where the CPU is throttled and the performance goes down over time after cpu utilization goes over a threshold the provider considers excessive for a particular plan.
The performance is measured at regular intervals in terms of tasks executed per hour, amount of idle cpu and amount of "cpu steal".
| Provider: | Amazon EC2 |
| Plan: | t3a.medium |
| Number of CPU cores: | 2 |
| Burstable CPU resources: | yes (20% baseline) |
| CPU type: | AMD EPYC 7571 |
| Description | Value | |
|---|---|---|
| Target CPU usage | Overall CPU that the endurance test tries to use across all CPU cores on the VM. | 50% |
| Average iterations per hour | Average number of tasks executed per hour. It represents how much work the VPS was able to achieve per hour. Higher numbers are better. | 1349 |
| Coefficient of variation iterations per hour | This represents the dispersion of the number of tasks per hour. It's the standard deviation divided by the mean number of tasks per hour. It's an indicator on stability of the performance of the VPS. A low standard deviation means the VPS speed stayed close to constant. A high standard deviation means performance varied significantly during the test. | 33.91% |
| Average CPU idle | Average percentage of CPU that was idle during the test. This number should be close to 50% unless the CPU is being throttled like in burstable VPS plans. | 47.5% |
| Min CPU idle | Minimum percentage of CPU idle during the test. | 38.2% |
| Average CPU steal | Average percentage of CPU steal during the test. Lower is better. | 14.93% |
| Max CPU steal | Maximum percentage of CPU steal during the test. | 49.52% |
Number of tasks completed per hour
| Date | Number of iterations per hour | CPU Idle % | CPU Steal % |
|---|---|---|---|
| November 07, 2019 02:00 UTC | 796 | 52.7 | 32.47 |
| November 07, 2019 00:00 UTC | 790 | 46.69 | 39.17 |
| November 06, 2019 22:00 UTC | 759 | 45.64 | 40.56 |
| November 06, 2019 20:00 UTC | 747 | 44.76 | 41.06 |
| November 06, 2019 18:00 UTC | 891 | 45.0 | 35.16 |
| November 06, 2019 16:00 UTC | 1657 | 48.46 | 1.0 |
| November 06, 2019 14:00 UTC | 1633 | 48.63 | 1.15 |
| November 06, 2019 12:00 UTC | 1672 | 48.56 | 0.93 |
| November 06, 2019 10:00 UTC | 1673 | 48.51 | 1.1 |
| November 06, 2019 08:00 UTC | 1709 | 48.01 | 0.9 |
| November 06, 2019 06:00 UTC | 1723 | 47.81 | 1.22 |
| November 06, 2019 04:00 UTC | 1724 | 47.66 | 1.2 |
| November 06, 2019 02:00 UTC | 1711 | 47.5 | 1.24 |
