Cell splitting in neural networks extends strong scaling (Hines et al. 2008)

 Download zip file   Auto-launch 
Help downloading and running models
Accession:97917
Neuron tree topology equations can be split into two subtrees and solved on different processors with no change in accuracy, stability, or computational effort; communication costs involve only sending and receiving two double precision values by each subtree at each time step. Application of the cell splitting method to two published network models exhibits good runtime scaling on twice as many processors as could be effectively used with whole-cell balancing.
Reference:
1 . Hines ML, Eichner H, Schürmann F (2008) Neuron splitting in compute-bound parallel network simulations enables runtime scaling with twice as many processors. J Comput Neurosci 25:203-10 [PubMed]
Citations  Citation Browser
Model Information (Click on a link to find other models with that property)
Model Type: Realistic Network;
Brain Region(s)/Organism: Generic;
Cell Type(s):
Channel(s):
Gap Junctions:
Receptor(s):
Gene(s):
Transmitter(s):
Simulation Environment: NEURON;
Model Concept(s): Methods;
Implementer(s): Hines, Michael [Michael.Hines at Yale.edu];
/
splitcell
nrntraub
mod
alphasyndiffeq.mod
alphasynkin.mod *
alphasynkint.mod *
ampa.mod
ar.mod *
cad.mod *
cal.mod *
cat.mod *
cat_a.mod *
gabaa.mod
iclamp_const.mod *
k2.mod *
ka.mod *
ka_ib.mod *
kahp.mod *
kahp_deeppyr.mod *
kahp_slower.mod *
kc.mod *
kc_fast.mod *
kdr.mod *
kdr_fs.mod *
km.mod *
naf.mod *
naf_tcr.mod
naf2.mod *
nap.mod *
napf.mod *
napf_spinstell.mod *
napf_tcr.mod *
par_ggap.mod *
pulsesyn.mod *
rampsyn.mod *
rand.mod *
ri.mod
traub_nmda.mod
                            
COMMENT
Four helpful hints:

1) before calling scale_connection_coef, one must call some NEURON
function (such as ri(x)) that forces calculation of all the connection
coefficients for all the sections.

2) if any diam or L is changed, then one must re-call the
scale_connection_coef procedure again for all compartments AFTER
re-forcing the normal calculation of them via a call to, e.g. ri(x).

3) note that ri(0.5) gives the resistance in mega ohms between 0.5
location and the 0 end and ri(1) gives the resistance in mega ohms
between the 0.5 location and the 1 end.

4) Call with a section access'ed.  Call below with (1,factor) to
change the axial resistance of (a parent's) x=0.5 to x=1 part and call
with (0.5, factor) to change the axial resistance for (a child's) x=0
to x=0.5 part.  Note: factor = current_ri_value/desired__ri_value.

ENDCOMMENT

NEURON { SUFFIX nothing }

VERBATIM
#ifdef NRN_MECHANISM_DATA_IS_SOA
#define get_nnode(sec) _nrn_mechanism_get_nnode(sec)
#define get_node(sec, idx) _nrn_mechanism_get_node(sec, idx)
#else
#define get_nnode(sec) sec->nnode
#define get_node(sec, idx) sec->pnode[idx]
const char* secname();
#endif
ENDVERBATIM

PROCEDURE scale_connection_coef(x, factor) {
VERBATIM {
	Section* sec;
	Node* nd;
#if defined(t)
	NrnThread* _nt = nrn_threads;
#endif
	sec = chk_access();
	if (_lx <= 0. || _lx > 1.) {
		hoc_execerror("out of range, must be 0 < x <= 1", (char*)0);
	}
	/*printf("scale_connection_coefs %s(%g) %d\n", secname(sec), _lx, get_nnode(sec));*/
	/* assumes 0 end of child connected to parent */
	if (_lx == 1.) {
		nd = get_node(sec, get_nnode(sec) - 1);
	}else{
		nd = get_node(sec, (int) (_lx*(double)(get_nnode(sec) - 1)));
	}
	/*printf("%g %g\n", NODEA(nd), NODEB(nd));*/
	NODEA(nd) *= _lfactor;
	NODEB(nd) *= _lfactor;
}
ENDVERBATIM
}