[Nek5000-users] run over small machine

[nek5000-users at lists.mcs.anl.gov]

Hi neks,

i am simulating a turbulent channel flow over a small machine, 64 cores 
in 4 nodes. So i have tried two different setup, one with much elements 
and N=4 and one with less elements with N=8. With the first setup, the 
run is fine, at high Re i have turbulent flow, but i have restart issue 
and very low performance. If i launch the second run (high order) with 
the same setup of the first run (low order) i have a laminar flow.
Another try done was to run over another machine with more power first 
setup (same elements ) and increasing N since 4 to 7.
I expected a turbulent flow too, instead i have another time a laminar flow.
My decision to decrease elements number and increase N, has been taken 
to following some answers in your mailing list archive to increase 
performance.


So, which could be the causes of this strange behaviors?

Another question is, how can i increase performance for this very small 
machine?
I already wrote i changed number of elements and N, i fixed too time 
step, gmres = 40 in SIZE, Helmotz and Pressure tolerance to 1 E-8 and 1 
E-5 and activated projection (param 94 and 95 equal to 5). There is 
something else that can i do?


Best regards,
Federico.