What we measured
Virtual learners log in and then follow a course just like real learners: dashboard, "My courses", opening a module, then a presence signal every 5 seconds and progress saved every 10 seconds. Their number increases in stages, from 10 to 1,200, each stage lasting several minutes.
The test was run on a single production server (8 cores), during the day, alongside the normal activity of the other platforms hosted on it. The figures refer to learners active at the same moment, not to enrolled learners: a platform usually has far more enrolled learners than users logged in simultaneously.
The results in charts
How to read the results
- Median response time: half of the requests respond faster. It stays around 40 milliseconds up to 800 learners, the same as with 10: nobody notices any slowdown.
- The slowest 5%: they stay under 400 milliseconds up to 800 learners. The isolated peak at 600 learners corresponds to a moment of one-off load on the server, which cleared at the next stage.
- From 1,000 learners, the processor approaches saturation: response times increase, but the platform keeps responding. Even at 1,200 learners, only 0.03% of requests failed.
- The database is never the bottleneck: a dozen connections at most, with no waiting. It is the processor that sets the capacity.
Detail by stage
| Simultaneous learners | Median | Slowest 5% | Requests / s | Processor |
|---|---|---|---|---|
| 10 | 35 ms | 108 ms | — | — |
| 50 | 39 ms | 94 ms | — | — |
| 100 | 39 ms | 80 ms | — | — |
| 200 | 36 ms | 71 ms | — | — |
| 300 | 32 ms | 76 ms | — | — |
| 400 | 31 ms | 109 ms | 128 | 34% |
| 600 | 39 ms | 1,300 ms (peak) | 193 | 46% |
| 800 | 41 ms | 395 ms | 265 | 64% |
| 1,000 | 161 ms | 652 ms | 314 | 78% |
| 1,200 | 276 ms | 897 ms | 360 | 89% |
No errors up to 400 learners, out of 64,042 requests over two runs; a single failed request out of 27,463 at the 600 and 800 stages.
A lighter course player
During a course, more than 80% of requests come from the player: the presence signal and the saving of progress. We have made them lighter: the signal drops from 12.5 to 3.4 milliseconds, saving from 13.5 to 5.8. For 200 learners, the processor used drops from 93% to 51%, that is, 45% less. The SCORM tracking recorded is strictly identical, checked in the database on around ten cases.
These improvements are included in the next update of your platform.
What this means for your sessions
- Whole intakes at the same time: several hundred learners can follow their module at the same moment without any slowdown.
- Smooth session openings: logging in remains the most demanding step for the server. For several hundred learners at a fixed time, allow a few minutes for the opening rather than a start to the exact second.
- Beyond about 1,000 simultaneous active learners, plan for a more powerful server or several servers: the hosting simulator helps you estimate the configuration.