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
net
durand.hoc *
groucho.hoc
groucho_gapbld.hoc *
groucho_gapbld_mix.hoc *
network_specification_interface.hoc
serial_or_par_wrapper.hoc *
synaptic_compmap_construct.hoc *
synaptic_map_construct.hoc *
                            
// groucho_gapbld.hoc
/*
*****************************this is one big comment ***************************
from            SUBROUTINE GROUCHO_gapbld (thisno, numcells, numgj,
     &       gjtable, allowedcomps, num_allowedcomps, display)
c       Construct a gap-junction network for groucho.f
$1 thisno double
$2 numcells = number of cells in population, e.g. number of tufted IB cells
$3 numgj = total number of gj to be formed in this population
// this matrix is returned: gjtable = table of gj's: each row is a gj.  
     Entries are: cell A,
c    compartment on cell A; cell B, compartment on cell B
$o4 c allowedcomps = a list of compartments where gj allowed to form
$5 num_allowedcomps = number of compartments in a cell on which a gj 
c    might form.
$6 display is an integer flag.  If display = 1, print gjtable

        INTEGER thisno, numcells, numgj, gjtable(numgj,4),
     &    num_allowedcomps, allowedcomps(num_allowedcomps)
        INTEGER i,j,k,l,m,n,o,p, ictr /0/
c ictr keeps track of how many gj have been "built"
        INTEGER display

        double precision seed, x(2), y(2)

Note: this function is for gap junctions that form between a cells that are
members of a population of a single cell type
*****************************this is one big comment ***************************
*/
objref gjtable, x, y, allowedcomps
obfunc groucho_gapbld() {localobj used
// see above note for arguments $1,$2,$3,$o4,$5
// print "arrived"
	thisno = $1
	numcells = $2
	numgj = $3
	allowedcomps = $o4
	num_allowedcomps = $5
	display = $6

	seed = new Vector()
	seed.append(137.e0)

	objref gjtable
	gjtable = new Matrix(numgj+1, 4+1) // fortran notation indicies start at 1

	ictr = 0
	k = 2
	not_unique = 0 // make global so not local in loops
        used = new Matrix(numcells+1, numcells+1, 2) // sparse

// 2
// print "starting loop"
	while (ictr < numgj) {
 //         print "ictr = ",ictr
	  not_unique = 1 // 1 is true, 0 is false
	    while (not_unique) {
		x = durand (seed, k, x)
// This defines a candidate cell pair
		y = durand (seed, k, y)
// This defines a candidate pair of compartments

		i = int ( x.x[0] * numcells ) + 1
		j = int ( x.x[1] * numcells ) + 1
//		print "i,j: ",i,", ",j
// no longer required		if (i.eq.0) i = 1
// no longer required		if (i.gt.numcells) i = numcells
// no longer required		if (j.eq.0) j = 1
// no longer required		if (j.gt.numcells) j = numcells

// Is the unordered cell pair (i,j) in the list so far?
// not necessary to be this efficient  if (ictr.eq.0) goto 1

 		not_unique = 0
 if (0) {
		for eL = 1, ictr {
//		  print "compare i,j with eL = ",eL, " : ",gjtable.x(eL,1),", ",gjtable.x(eL,3)
		  if ((gjtable.x(eL,1) == i) && (gjtable.x(eL,3) == j)) { not_unique = 1 }
 		  if ((gjtable.x(eL,1) == j) && (gjtable.x(eL,3) == i)) { not_unique = 1 }
		}
//		print " at end of loop not_unique = ",not_unique
  }else{
                if (used.getval(i, j) || used.getval(j, i)){
                        not_unique = 1
                }else{
                        used.x[i][j] = 1
                }
		if (one_tenth_ncell) {
			not_unique = 0
		}
  }
	    } // while replaces if (not_unique.eq.1) goto 2

// Proceed with construction
// 1
	  ictr = ictr + 1
          m = int ( y.x[0] * num_allowedcomps ) + 1
          n = int ( y.x[1] * num_allowedcomps ) + 1
//	print "assigning quantities: ", i, ", ", j, ", ", allowedcomps.x[m], ", ",allowedcomps.x[n]

         gjtable.x(ictr,1) = i
         gjtable.x(ictr,3) = j
         gjtable.x(ictr,2) = allowedcomps.x (m)
         gjtable.x(ictr,4) = allowedcomps.x (n)
	}
//            if (ictr.lt.numgj) goto 2

// Possibly print out gjtable when done.
       if ((display == 1) && (thisno == 0)) {
        print " GJTABLE "
        for i = 1, numgj {
        printf("%6d %6d %6d %6d",gjtable.x(i,1), gjtable.x(i,2), \
                gjtable.x(i,3), gjtable.x(i,4))
        }
       }
	return gjtable
}