run yade in parallel

Asked by Mahdeyeh

Hi everyone
I want to run polyhedra particles in this computer:

CPU (Cores):16
2.3 Hz
Hard disc (HDD): 80 GB
Ram: 32 GB
ubuntu : 16.04
yade 1.20.0

number of bodies : 1511

for increasing speed in iterations per seconds and anything else, how many number of cores on the machine is better to use ? -j8 or -j12 or -j14 or -j16?

Thanks,
Mahdeyeh

Question information

Language:
English Edit question
Status:
Solved
For:
Yade Edit question
Assignee:
No assignee Edit question
Solved by:
Jan Stránský
Solved:
Last query:
Last reply:
Revision history for this message
Jan Stránský (honzik) said :
#1

Hello,
the best option is to test and decide yourself according to your specific simulation.
Cheers
Jan

Revision history for this message
Mahdeyeh (mahdiye.sky) said :
#2

Than you Jan
I think so.

I ran the codes with : yade j-8 code.py , but it shows this error:

"ERROR /build/yade-KKgSmd/yade-1.20.0/pkg/common/InsertionSortCollider.cpp:65 insertionSortParallel: Parallel insertion sort needs verletDist>0"

I think it because of my particles are polyhedra not sphere, yes? so is this error is important? How I can solve it?

Here is my code:

from yade import export,polyhedra_utils
import os
from yade import plot
import math
from yade import utils
import pylab
import matplotlib; matplotlib.rc('axes',grid=True)
from matplotlib import pyplot
import numpy as np
from numpy import *
from yade import export as expt

# Input data:

RawVer=np.genfromtxt('ring.txt',names=True,dtype=None)
# ListVer is list of all the vertices of ring`s polygons
Ver=()
ListVer=[]
for b in RawVer:
    if b[0]=='*LWPOLYLINE':
        ListVer.append(Ver)
        Ver=()
        continue
    Cordn=b[0]
    Cordn=np.fromstring(Cordn, sep=',')
    Cordn=tuple(Cordn.tolist())
    Cordn1=Cordn+(0.1,) # add z vertex to coordinates
    Cordn2=Cordn+(0.2,) # add z vertex to coordinates
    if not Cordn1 in Ver:
        Ver=Ver+(Cordn1,Cordn2)
ListVer.append(Ver)

RawVer1=np.genfromtxt('boundary.txt',names=True,dtype=None)
# ListVer1 is list of all the vertices of boundary`s polygons
Ver1=()
ListVer1=[]
for b in RawVer1:
    if b[0]=='*LWPOLYLINE':
        ListVer1.append(Ver1)
        Ver1=()
        continue
    Cordn=b[0]
    Cordn=np.fromstring(Cordn, sep=',')
    Cordn=tuple(Cordn.tolist())
    Cordn1=Cordn+(-2,) # add z vertex to coordinates
    Cordn2=Cordn+(2,) # add z vertex to coordinates
    if not Cordn1 in Ver1:
        Ver1=Ver1+(Cordn1,Cordn2)
ListVer1.append(Ver1)

RawVer2=np.genfromtxt('waste.txt',names=True,dtype=None)
# ListVer2 is list of all the vertices of caved waste rock`s polygons
Ver2=()
ListVer2=[]
for b in RawVer2:
    if b[0]=='*LWPOLYLINE':
        ListVer2.append(Ver2)
        Ver2=()
        continue
    Cordn=b[0]
    Cordn=np.fromstring(Cordn, sep=',')
    Cordn=tuple(Cordn.tolist())
    Cordn1=Cordn+(0.1,) # add z vertex to coordinates
    Cordn2=Cordn+(0.2,) # add z vertex to coordinates
    if not Cordn1 in Ver2:
        Ver2=Ver2+(Cordn1,Cordn2)
ListVer2.append(Ver2)

# Materials type:

Dolomite = PolyhedraMat()
Dolomite.density = 2870 #kg/m^3
Dolomite.young = 24.36e9 #Pa
Dolomite.poisson = 0.2
Dolomite.frictionAngle = radians(55.12) #rad

Shale = PolyhedraMat()
Shale.density = 2750 #kg/m^3
Shale.young = 6e9 #Pa
Shale.poisson = 0.23
Shale.frictionAngle = radians(42) #rad

for ii in ListVer:
    O.bodies.append(polyhedra_utils.polyhedra(Dolomite,v=ii,fixed=False, color=(0.1,0.5,0.2), mask=3))

for iii in ListVer1:
    O.bodies.append(polyhedra_utils.polyhedra(Dolomite,v=iii,fixed=True, color=(1,0,1), mask=4))

for iiii in ListVer2:
    O.bodies.append(polyhedra_utils.polyhedra(Shale,v=iiii,fixed=False, color=(0.9,0.81,0.45), mask=5))

O.bodies.erase(340) # delete wall under ring: id: 340

# block z translation and block x, y, z rotations
for b in O.bodies:
    if b.mask is 3:
        b.state.blockedDOFs='zXY'
for b in O.bodies:
    if b.mask is 5:
        b.state.blockedDOFs='zXY'

# function for calming particles
def calm():
    for c in O.bodies:
        c.state.vel=Vector3(0,0,0)
        c.state.angVel=Vector3(0,0,0)

# returns a value that can be useful for evaluating the stability of the packing. It is defined as (mean force on particles)/(mean contact force), so that it tends to 0 in a stable packing.
def checkUnbalanced():
    print "iter %d , unbalanced forces %f"%(O.iter, utils.unbalancedForce()) # %[(keyname)][flags][width][.precision]typecode : String Formatting
    iter00=O.iter

    Unbalanced=open("Unbalanced iter Unbalanced forces.txt","a")
    Unbalanced.write(repr(iter00)+' '+repr(utils.unbalancedForce())+' '+"\n")
    Unbalanced.close()

# Engines:

#os.mkdir(O.tags['id']) # Created separate directory and put all files of VTK in it, till external files be seperate because of large simulation.

O.engines=[
   ForceResetter(),
   InsertionSortCollider([Bo1_Polyhedra_Aabb(),]), # We can set collider's verletDist to a fraction of the polyhedra minimum size, since it determines how much is each body enlarged to avoid collision detection at every step.
   InteractionLoop(
      [Ig2_Polyhedra_Polyhedra_PolyhedraGeom(),],
      [Ip2_PolyhedraMat_PolyhedraMat_PolyhedraPhys()], # collision "physics"
      [Law2_PolyhedraGeom_PolyhedraPhys_Volumetric()] # contact law -- apply forces
   ),
   NewtonIntegrator(gravity=(0,-9.81,0),damping=0.2),
]

utils.calm()

# the model has to calm, because there are some overlaps in particles.
O.engines=O.engines+[PyRunner(iterPeriod=20,command='calm()',label="calmRunner")] # because we need to calm only on the first few steps in our model.
O.engines=O.engines+[PyRunner(command='checkUnbalanced()',iterPeriod=10000,label="checker")] # call our function defined above 1000 cycles

O.dt=10e-6

O.saveTmp('model1')
O.save('model1.bz2')

# first run of model
O.run(50000,True)

O.save('model2.bz2')

# Outputs:

clrOre=[0.1,0.5,0.2]
clrWaste=[0.9,0.81,0.45]

def positions():
    for b in O.bodies:
        if b.shape.color==clrOre:
            time1=O.time

            Positions=open("Positions time x y z.txt","a")
            Positions.write(repr(time1)+' '+repr(b.state.pos[0])+' '+repr(b.state.pos[1])+' '+repr(b.state.pos[2])+' '"\n")
            Positions.close()

def oreAmount():
    clrOre=[0.1,0.5,0.2]
    M_o=0
    for b in O.bodies:
        if b.shape.color==clrOre:
                M_o+=b.state.mass
    return M_o

os.mkdir(O.tags['id']) # Created separate directory and put all files of VTK in it, till external files be seperate because of large simulation.
O.engines=O.engines+[PyRunner(iterPeriod=20000,command='VTKview()',label="VTKview")] # every 20000 cycles we can see picture of model in Paraview
PolVtkData = expt.VTKExporter(O.tags['id']+'/'+'polData') # save animation and images of simulation in VTK format (paraview) in defined cycle
def VTKview():
    PolVtkData.exportPolyhedra()

m_o = 0.00005 # mass of ore from the block
m_w = 0.00005 # mass of waste

R = 0 # R total ore utilization

D = 0
D_m = 0

E = 0 # E extraction

M_o = oreAmount() # M_o total ore mass

totalEMass = 1.25 * M_o # Convergence criteria for the models are set as 125% extraction of the ring mass.

def excavation():
        m_o = 0.00005
        m_w = 0.00005
        for b in O.bodies:
            if b.state.pos[0]>(-2.4) and b.state.pos[0]<1 and b.state.pos[1]>0 and b.state.pos[1]<1: # -2.4<x<1 because we don`t know about repose angle and distance for bucket yet

                if b.shape.color==clrOre:
                    m_o += b.state.mass #total ore mass extracted
                    print 'ore'
                if b.shape.color==clrWaste:
                    m_w += b.state.mass #total waste mass
                    print 'waste'

        print 'm_o' , m_o , 'm_w' , m_w
        iter1=O.iter

        Mass=open("Mass iter massore masswaste.txt","a")
        Mass.write(repr(iter1)+' '+repr(m_o)+' '+repr(m_w)+' '"\n")
        Mass.close()

        E = ( m_o + m_w) / M_o * 100 ##extraction definition

        D = (m_w / (m_w + m_o)) * 100 ##ore dilution

        D_m = (1- ( m_o / (m_w + m_o))) * 100 ##metal dilution

        R = (m_o / M_o) * 100 ## Total ore recovery
        print 'E' , E , 'D' , D , 'D_m' , D_m , 'R' , R
        iter2=O.iter

        DataAnalysis=open("DataAnalysis iter E D DM R.txt","a")
        DataAnalysis.write(repr(iter2)+' '+repr(E)+' '+repr(D)+' '+repr(D_m)+ ' '+repr(R)+"\n")
        DataAnalysis.close()

# Extra Engines:

O.engines=O.engines+[PyRunner(command='positions()',realPeriod=0.5,label="flow")] # call our function defined above every 60 seconds

O.engines=O.engines+[PyRunner(iterPeriod=200,command='excavation()',label="excavate")] # call our function defined above every 200 cycles

polyhedra_utils.SizeRatio() # save sizes of polyhedra into file
export.textPolyhedra('poly.txt') # Save polyhedra into a text file, id, number of vertices and number of surfaces, particleVelocity and ... .

Revision history for this message
Mahdeyeh (mahdiye.sky) said :
#3

I add : InsertionSortCollider(..., verletDist=.05), for above error and it does not show but now when I run model it says this:

"WARN /build/yade-KKgSmd/yade-1.20.0/pkg/common/InsertionSortCollider.cpp:80 insertionSortParallel: Parallel insertion: only 2 thread(s) used. The number of bodies is probably too small for allowing more threads, or the geometry is flat. The contact detection should succeed but not all available threads are used."

what should I do? is that decrease speed run?

Revision history for this message
Best Jan Stránský (honzik) said :
#4

> number of bodies : 1511
> ... The number of bodies is probably too small for allowing more threads

> is that decrease speed run?

no, rather it is not increasing speed.
If you are interested, have a look at the original code [1]. The warning depends on cvunkVerlet variable, which depends on verletDist. So you can try to play with verletDist.

Jan

[1] https://gitlab.com/yade-dev/trunk/blob/1.20/pkg/common/InsertionSortCollider.cpp#L80

Revision history for this message
Mahdeyeh (mahdiye.sky) said :
#5

Thanks Jan Stránský, that solved my question.