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
cells
hoc
mod
net
README
balcomp.hoc *
bgrunme
bgsmall.sh
bgsplit.sh
cell_templates.hoc *
clear.hoc *
finit.hoc *
fortmap.hoc *
gidcell.hoc
gidcell.ses *
init.hoc
manage_setup.hoc
metisbal.sh
mosinit_orig.hoc *
onecell.hoc
onecell.ses *
prcellstate.hoc *
prepare.sh
printcon.hoc *
spkplt.hoc *
vclampg.hoc *
vcompclamp.hoc *
vcompsim.hoc *
                            
setuptime = startsw()
{xopen("hoc/defvar.hoc")}

// load balancing implies several runs.
// 0) load balancing just plain off (but will read mcomplex.dat if exists)
// 1) create a mcomplex.dat file (only one cpu needed)
// 2) create single split balance.dat file (any number of cpus can be used)
//  manually: run the balcomp program to create balance.ncpu files
// 3) run with single split load balancing on (ncpu corresponding to files from balcomp)
// add more for whole cell and multisplit load balancing
// 4) create whole cell load balance file
// 5) run with whole cell load balance file
// 6) create multisplit load balance information file
// 7) run using multisplit load balance
default_var("load_balance_phase", 0)

default_var("one_tenth_ncell", 1)
default_var("use_gap", 0)
default_var("use_ectopic", 1)
default_var("use_inject", 1)
default_var("awake", 1)
default_var("default_delay", 0.5)
default_var("ranseedbase", 1)
default_var("use_traubexact", 0) // will be reset to 0 if load balance
default_var("wholecell_prefix", "cxwhole")
default_var("multisplit_prefix", "cx")
default_var("multisplit_nhost", 256)
default_var("msoptfactor", 0.8)

default_var("spike_compress", 5)
default_var("cacheeffic", 1)  // for multisplit it is always 1
default_var("multisend", 0)
default_var("selfevents", 0)

gfac_AMPA = 1
gfac_NMDA = 1
gfac_GABAA = 1

use_p2c_net_connections = 0 // not 0, requires p2c emitted  map and compmap files

{localloadfile("manage_setup.hoc")}

steps_per_ms = 40
dt = .025
secondorder = 2
tstop = 1000

if (0) { pc.runworker()  pc.done() quit() }
if (name_declared("mosinit")) { stop }

//{finitialize(-65) cvode_local(1) cvode.atol(1e-4)} // the finitialze avoids /0 in BREAKPOINT

prun()

endtime = startsw()

if (pc.id == 0) { print "tstop = ", tstop }
if (pc.id == 0) { print "RunTime: ", runtime }

//{localloadfile("prcellstate.hoc")}
proc pcs() {local spgid
	spgid = thishost_gid(103)
	if (spgid >= 0) prcellstate(load_balance_phase, spgid)
	pc.runworker()  pc.done() quit()
}
//pcs()

spike2file()

if (pc.nhost > 5) {cvode_active(1)} // to count equations
{pc.runworker()}

print "Maximum integration interval: ", mindelay()
getstat()
prhist()
print_spike_stat_info()

{pc.done()}

perf2file()
endtime = startsw() - endtime
print "endtime ", endtime

if (!serial) { quit() }