Partial -prof gc results (churn and time missing for some parameters)

[Bernd Eckenfels]

Hello Aleksey,

thanks for the analysis. Sounds about right, I do need to be reminded
constantly that JMH is not using a constant number of executions. Thats
another one of those false asumptions :)  (and the fact that the buffer
is actually not allocated per execution, which is of course good in
this case).

I did had some longer running attempts but they all got canceled, I
should have waited. :)

Gruss
Bernd


 Am
Thu, 30 Apr 2015 12:23:19 +0300 schrieb Aleksey Shipilev
<aleksey.shipilev at oracle.com>:
> I would guess that nothing cheesy is going on. On small datasets, you
> produce the same 400+ bytes per benchmark op, but every microsecond,
> which adds up to 400+ Mb/(sec*thread) GC pressure. On large datasets,
> you produce the same 400 bytes each 10 milliseconds, and thus the GC
> pressure is four orders of magnitude smaller.