Accuracy of throughput shaping timer vs constant throughput timer?

[David]

Is the throughput shaping timer best used with certain types of thread
groups or within certain scopes?

Or is the throughput shaping timer calculating throughput differently
than the constant throughput timer?

I did a test using both timer types and constant throughput timer had
more accurate control/approximation of the target throughput to
generate. The shaping time couldn't produce the desired throughput
whether using more or less threads than the constant throughput timer.
It was off by a half in my tentative testing.

I used standard thread group, and put timers in global test plan
scope. Tried with 200 threads, 100 threads, 5 threads. Used 1 second
ramp up per thread (e.g. 200 sec for 200 threads) and also tried 1
second ramp up total regardless of thread count.

Set target throughput to 5 and 10 requests/sec (300 and 600 requests/
min for the constant throughput timer), being 5 & 10 RPS for the
throughput shaping timer. I set intervals of 1 RPS to 5 RPS for 60 sec
then 5 RPS to 10 RPS for 60 sec, then 10 RPS to 10 RPS for 60 sec,
then 10 RPS to 7 RPS for 120 sec. For throughput shaping timer, I
didn't really care about the shape of throughput, jsut wanted to check
that the target throughput was achieved. Didn't seem that way for me.

Test used HTTP request samplers hitting some web service APIs. That
return response that's less than a second. 300 ms to 1 sec.

[Andrey Pohilko]

Yes, Shaper use different throughput calculation algorithm, it has its drawbacks.