[gmx-users] Parallelization of nose-hoover thermostat.

[Mehmet Suezen]

Erik Lindahl wrote:
> You mean you get different parallellization performance depending on the
> thermostat you use?
> 
> That sounds a little strange; you will get slightly worse performance if
> you use constraints
> together with Parinello-Rahman coupling, since the constraints have to
> be iterated in this case,
> but an NVT ensemble shouldn't slow things down as far as I know... What
> kind of results do you get?
> 
> Cheers,
> 
> Erik

Specifically speaking, I run same configuration in both ensemble (nve
and nvt hoover). 
it is strange because in 4 proc runs, nve runs is about 4 times faster
then nvt run.
Most probably I'm missing someting in parameter files or something else
in .top, that may affect performance this much. Or I really spoiled to
code after my addition of pair potential, but 
I didn't touch any parallelization related routines. Or there might be a
compilation procedure was not well gone. I'll extent my test tho.

Down Below is the portion of .mdp file

Cheers,

Mehmet


-------
; Neighbor Searching
nstlist=0        ; frequency of constructing list (zero means once to
all-pairs)
ns_type=Simple   ; construction method
pbc=no           ; periodicity in the system
rlist=0          ; cutoff distance in the short-range neighborlist (nm)

; Electrostatics and VdW
;   non-pbc  with nstlist and rlist zero
;infinite range all particle contributions will be taken in to account
coulombtype=Cut-Off         ; direct calculation by some cut-off
rcoulomb=0                  ; all interactions
vdwtype=Cut-Off             ; vdw cut_off
rvdw=0                      ; no pair pot. cut-off (nm)
DispCorr=no                 ; dispersion correction

; Temperature Coupling
tc_grps=System          ; groups to be coupled
tau_t=1.0               ; time-constant for coupling
ref_t=2773              ; reference temperature
----