Github user vlsi commented on the issue:
https://github.com/apache/jmeter/pull/231
> Philippe: It is just that as per your confirmation, one of the aims of
this new component was:
> Vladimir: "The very basic requirement is to ensure you end up exactly 50
iterations **per hour**.
> Philippe: That's why I tested such things, so I think it needs to be
mitigated a bit.
Please, be careful: I suggested 50 per **hour**, and you are testing 50 per
**second**. That is 3600 times higher rate than I suggested. I don't mean the
timer works for 50/hour rate only, but 50/second is a bit different thing.
>Now regarding "Repeatable test profile", for 2 runs, I don't see "similar
test profile"
Let us consider "10 second test with 1 per second on average".
Current implementation writes "o.a.j.t.ConstantPoissonProcessGenerator:
Generated ..." log record with expected timestamps of the samples.
```
2017-12-23 21:39:01,449 INFO o.a.j.t.ConstantPoissonProcessGenerator:
Generated 11 timings (ExponentialTimer 10 required, rate 1.0, duration 10,
exact lim 20000, i11) in 0 ms, restar
t was issued 12 times. First 15 events will be fired at:
0.11401676510964737 (+0.11401676510964737), 0.6130748250774025
(+0.49905805996775515), 3.342639426533685 (+2.7295646014562824
), 3.377172823174598 (+0.03453339664091315), 5.112423358789579
(+1.735250535614981), 5.1755972505795125 (+0.06317389178993338),
5.514713410628992 (+0.3391161600494792), 7.94457420086467
(+2.4298607902356784), 8.509235493336268 (+0.5646612924715981), 8.5888989163364
(+0.07966342300013096), 10.107204382018718 (+1.5183054656823192)
```
A bit more readable: 0.1, 0.6, 3.3, 3.4, 5.1, 5.2, 5.5, 7.9, 8.5, 8.6, 10.1
seconds. Note: there's exactly 10 samples in between 0..10 sec.
Let me restart the test:
```
2017-12-23 21:40:19,752 INFO o.a.j.t.ConstantPoissonProcessGenerator:
Generated 11 timings (ExponentialTimer 10 required, rate 1.0, duration 10,
exact lim 20000, i11) in 0 ms, restart was issued 12 times. First 15 events
will be fired at: 0.11401676510964737 (+0.11401676510964737),
0.6130748250774025 (+0.49905805996775515), 3.342639426533685
(+2.7295646014562824
), 3.377172823174598 (+0.03453339664091315), 5.112423358789579
(+1.735250535614981), 5.1755972505795125 (+0.06317389178993338),
5.514713410628992 (+0.3391161600494792), 7.94457420086467
(+2.4298607902356784), 8.509235493336268 (+0.5646612924715981), 8.5888989163364
(+0.07966342300013096), 10.107204382018718 (+1.5183054656823192)
```
Note: the timestamps are exactly the same (the log timestamp is different,
so it is a different execution).
This is super-useful when testing low-rate things (e.g. 100 per hour) as it
enables to have the same load profile, thus there's less moving parts when
comparing different runs. Otherwise it is hard to tell if the difference in
"CPU usage" profile is caused by different sample sequence or different system
under test behavior.
Note: in order for "repeatable test profile" to work, you need to set
non-zero "randomSeed" (see
https://github.com/apache/jmeter/pull/231/files#diff-76a5e8f118fb5e533f52ea80fd10dd0bR58
)
Unfortunately, JMeter is unable to initialize "test element property with a
random-generated value". Test elements are created with constant values only,
so I had to resort to a 0 being "random". I agree this is more like "easier for
developer" rather than "easier for JMeter user", however I'm not sure what is
JMeter way of doing that config.
>Finally regarding Batched Departures, I am not sure to understand how it
is supposed to influence load and if it is supposed to work correctly with
First part in GUI.
The idea is "batched departures" keep the load level (total number of
samples per hour), however the samples are scheduled to come in pairs, triples,
etc.
For instance, with "batch size=2, throughput=10, duration=10" it produces
samples as follows: 0.52, 0.52, 0.67, 0.67, 3.4, 3.4, 9.37, 9.37, 9.43, 9.43.
Does the above make sense?
PS. That is why I was saying [the documentation is
missing](https://github.com/apache/jmeter/pull/231#issuecomment-342008122)
---
