FilteredMIS.cpp

Go to the documentation of this file.

/****************************************************************/
/* Parallel Combinatorial BLAS Library (for Graph Computations) */
/* version 1.6 -------------------------------------------------*/
/* date: 6/15/2017 ---------------------------------------------*/
/* authors: Ariful Azad, Aydin Buluc  --------------------------*/
/****************************************************************/
/*
 Copyright (c) 2010-2017, The Regents of the University of California
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
 in the Software without restriction, including without limitation the rights
 to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 copies of the Software, and to permit persons to whom the Software is
 furnished to do so, subject to the following conditions:
 
 The above copyright notice and this permission notice shall be included in
 all copies or substantial portions of the Software.
 
 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
 THE SOFTWARE.
 */
 
#define DETERMINISTIC
#include "CombBLAS/CombBLAS.h"
#include <mpi.h>
#include <sys/time.h>
#include <iostream>
#include <functional>
#include <algorithm>
#include <vector>
#include <string>
#include <sstream>
#ifdef THREADED
    #ifndef _OPENMP
    #define _OPENMP
    #endif
    #include <omp.h>
#endif
 
 
double cblas_alltoalltime;
double cblas_allgathertime;
int cblas_splits;
 
#include "TwitterEdge.h"
 
#define EDGEFACTOR 5 // For MIS
#define ITERS 16
#define PERCENTS 4  // testing with 4 different percentiles
 
using namespace std;
using namespace combblas;
 
 
template <typename PARMAT>
void Symmetricize(PARMAT & A)
{
    // boolean addition is practically a "logical or"
    // therefore this doesn't destruct any links
    PARMAT AT = A;
    AT.Transpose();
    A += AT;
}
 
 
struct DetSymmetricize: public std::binary_function<TwitterEdge, TwitterEdge, TwitterEdge>
{
    // have to deterministically choose between one of the two values.
    // cannot just add them because that will change the distribution (small values are unlikely to survive)
    TwitterEdge operator()(const TwitterEdge & g, const TwitterEdge & t)
    {
        TwitterEdge toret = g;
 
        if(((g.latest + t.latest) & 1) == 1)
        {
            toret.latest = std::min(g.latest, t.latest);
        }
        else
        {
            toret.latest = std::max(g.latest, t.latest);
        }
        return toret;
    }
};
 
typedef SpParMat < int64_t, TwitterEdge, SpDCCols<int64_t, TwitterEdge > > PSpMat_Twitter;
typedef SpParMat < int64_t, bool, SpDCCols<int64_t, bool > > PSpMat_Bool;
 
void SymmetricizeRands(PSpMat_Twitter & A)
{
    PSpMat_Twitter AT = A;
    AT.Transpose();
    // SpParMat<IU,RETT,RETDER> EWiseApply (const SpParMat<IU,NU1,UDERA> & A,
    // const SpParMat<IU,NU2,UDERB> & B, _BinaryOperation __binary_op, bool notB, const NU2& defaultBVal)
    // Default B value is irrelevant since the structures of the matrices are
    A = EWiseApply<TwitterEdge, SpDCCols<int64_t, TwitterEdge > >(A, AT, DetSymmetricize(), false, TwitterEdge());
}
 
#ifdef DETERMINISTIC
MTRand GlobalMT(1);
#else
MTRand GlobalMT;    // generate random numbers with Mersenne Twister 
#endif
 
 
struct Twitter_obj_randomizer : public std::unary_function<TwitterEdge, TwitterEdge>
{
  const TwitterEdge operator()(const TwitterEdge & x) const
  {
    short mycount = 1;
    bool myfollow = 0;
    time_t mylatest = static_cast<int64_t>(GlobalMT.rand() * 10000);    // random.randrange(0,10000)
 
    return TwitterEdge(mycount, myfollow, mylatest);
  }
};
 
struct Twitter_materialize: public std::binary_function<TwitterEdge, time_t, bool>
{
    bool operator()(const TwitterEdge & x, time_t sincedate) const
    {
        if(x.isRetwitter() && x.LastTweetBy(sincedate))
            return false;   // false if the edge is going to be kept
        else
            return true;    // true if the edge is to be pruned
    }
};
 
//  def rand( verc ):
//  import random
//  return random.random()
struct randGen : public std::unary_function<double, double>
{
    const double operator()(const double & ignore)
    {
        return GlobalMT.rand();
    }
};
 
 
int main(int argc, char* argv[])
{
    int nprocs, myrank; 
#ifdef _OPENMP
    int cblas_splits = omp_get_max_threads();
        int provided, flag, claimed;
        MPI_Init_thread(&argc, &argv, MPI_THREAD_FUNNELED, &provided );
        MPI_Is_thread_main( &flag );
        if (!flag)
            SpParHelper::Print("This thread called init_thread but Is_thread_main gave false\n");
        MPI_Query_thread( &claimed );
        if (claimed != provided)
            SpParHelper::Print("Query thread gave different thread level than requested\n");
#else
    MPI_Init(&argc, &argv);
    int cblas_splits = 1;   
#endif
 
    MPI_Comm_size(MPI_COMM_WORLD,&nprocs);
    MPI_Comm_rank(MPI_COMM_WORLD,&myrank);
    if(argc < 2)
    {
        if(myrank == 0)
        {
            cout << "Usage: ./FilteredMIS <Scale>" << endl;
        }
        MPI_Finalize();
        return -1;
    }
    {
        // Declare objects
        PSpMat_Twitter A;
        shared_ptr<CommGrid> fullWorld;
        fullWorld.reset( new CommGrid(MPI_COMM_WORLD, 0, 0) );
        FullyDistVec<int64_t, int64_t> indegrees(fullWorld);    // in-degrees of vertices (including multi-edges and self-loops)
        FullyDistVec<int64_t, int64_t> oudegrees(fullWorld);    // out-degrees of vertices (including multi-edges and self-loops)
        FullyDistVec<int64_t, int64_t> degrees(fullWorld);  // combined degrees of vertices (including multi-edges and self-loops)
        PSpMat_Bool * ABool;
 
        SpParHelper::Print("Using synthetic data, which we ALWAYS permute for load balance\n");
        SpParHelper::Print("We only balance the original input, we don't repermute after each filter change\n");
        SpParHelper::Print("BFS is run on UNDIRECTED graph, hence hitting CCs, and TEPS is bidirectional\n");
 
        double initiator[4] = {.25, .25, .25, .25}; // creating erdos-renyi
        double t01 = MPI_Wtime();
        DistEdgeList<int64_t> * DEL = new DistEdgeList<int64_t>();
 
        unsigned scale = static_cast<unsigned>(atoi(argv[1]));
        ostringstream outs;
        outs << "Forcing scale to : " << scale << endl;
        SpParHelper::Print(outs.str());
 
        // parameters: (double initiator[4], int log_numverts, int edgefactor, bool scramble, bool packed)
        DEL->GenGraph500Data(initiator, scale, EDGEFACTOR, true, true );    // generate packed edges
        SpParHelper::Print("Generated renamed edge lists\n");
 
        ABool = new PSpMat_Bool(*DEL, false);
        int64_t removed  = ABool->RemoveLoops();
        ostringstream loopinfo;
        loopinfo << "Converted to Boolean and removed " << removed << " loops" << endl;
        SpParHelper::Print(loopinfo.str());
        ABool->PrintInfo();
        delete DEL; // free memory
        A = PSpMat_Twitter(*ABool); // any upcasting generates the default object
 
        double t02 = MPI_Wtime();
        ostringstream tinfo;
        tinfo << "Generation took " << t02-t01 << " seconds" << endl;
        SpParHelper::Print(tinfo.str());
 
        // indegrees is sum along rows because A is loaded as "tranposed", similarly oudegrees is sum along columns
        ABool->PrintInfo();
        ABool->Reduce(oudegrees, Column, plus<int64_t>(), static_cast<int64_t>(0));
        ABool->Reduce(indegrees, Row, plus<int64_t>(), static_cast<int64_t>(0));
 
        // indegrees_filt and oudegrees_filt is used for the real data
        FullyDistVec<int64_t, int64_t> indegrees_filt(fullWorld);
        FullyDistVec<int64_t, int64_t> oudegrees_filt(fullWorld);
 
        typedef FullyDistVec<int64_t, int64_t> IntVec;  // used for the synthetic data (symmetricized before randomization)
            FullyDistVec<int64_t, int64_t> degrees_filt[4] = {IntVec(fullWorld), IntVec(fullWorld), IntVec(fullWorld), IntVec(fullWorld)};  
        int64_t keep[PERCENTS] = {100, 1000, 2500, 10000};  // ratio of edges kept in range (0, 10000)
 
        degrees = indegrees;
        degrees.EWiseApply(oudegrees, plus<int64_t>());
        SpParHelper::Print("All degrees calculated\n");
        delete ABool;
 
        float balance = A.LoadImbalance();
        ostringstream outlb;
        outlb << "Load balance: " << balance << endl;
        SpParHelper::Print(outlb.str());
 
        // We symmetricize before we apply the random generator
        // Otherwise += will naturally add the random numbers together
        // hence will create artificially high-permeable filters
        Symmetricize(A);    // A += A';
        SpParHelper::Print("Symmetricized\n");
 
        A.Apply(Twitter_obj_randomizer());
        A.PrintInfo();
        SymmetricizeRands(A);
        SpParHelper::Print("Symmetricized Rands\n");
        A.PrintInfo();
 
 
        FullyDistVec<int64_t, int64_t> * nonisov = new FullyDistVec<int64_t, int64_t>(degrees.FindInds(bind2nd(greater<int64_t>(), 0)));
        SpParHelper::Print("Found (and permuted) non-isolated vertices\n");
        nonisov->RandPerm();    // so that A(v,v) is load-balanced (both memory and time wise)
        A(*nonisov, *nonisov, true);    // in-place permute to save memory
        SpParHelper::Print("Dropped isolated vertices from input\n");
 
        indegrees = indegrees(*nonisov);    // fix the degrees arrays too
        oudegrees = oudegrees(*nonisov);
        degrees = degrees(*nonisov);
        delete nonisov;
 
        for (int i=0; i < PERCENTS; i++)
        {
            PSpMat_Twitter B = A;
            B.Prune(bind2nd(Twitter_materialize(), keep[i]));
            PSpMat_Bool BBool = B;
            BBool.PrintInfo();
            float balance = B.LoadImbalance();
            ostringstream outs;
            outs << "Load balance of " << static_cast<float>(keep[i])/100 << "% filtered case: " << balance << endl;
            SpParHelper::Print(outs.str());
 
            // degrees_filt[i] is by-default generated as permuted
            BBool.Reduce(degrees_filt[i], Column, plus<int64_t>(), static_cast<int64_t>(0));  // Column=Row since BBool is symmetric
        }
 
        float balance_former = A.LoadImbalance();
        ostringstream outs_former;
        outs_former << "Load balance: " << balance_former << endl;
        SpParHelper::Print(outs_former.str());
 
        for(int trials =0; trials < PERCENTS; trials++)
        {
            cblas_allgathertime = 0;
            cblas_alltoalltime = 0;
            double MISVS[ITERS]; // numbers of vertices in each MIS
            double TIMES[ITERS];
 
            LatestRetwitterMIS::sincedate = keep[trials];
            LatestRetwitterSelect2nd::sincedate = keep[trials];
            ostringstream outs;
            outs << "Initializing since date (only once) to " << LatestRetwitterMIS::sincedate << endl;
            SpParHelper::Print(outs.str());
 
            for(int sruns = 0; sruns < ITERS; ++sruns)
            {
                double t1 = MPI_Wtime();
                int64_t nvert = A.getncol();
 
                //# the final result set. S[i] exists and is 1 if vertex i is in the MIS
                //S = Vec(nvert, sparse=True)
                FullyDistSpVec<int64_t, uint8_t> S ( A.getcommgrid(), nvert);
 
                //# the candidate set. initially all vertices are candidates.
                //# this vector doubles as 'r', the random value vector.
                //# i.e. if C[i] exists, then i is a candidate. The value C[i] is i's r for this iteration.
                //C = Vec.ones(nvert, sparse=True)
                //FullyDistSpVec's length is not the same as its nnz
                //Since FullyDistSpVec::Apply only affects nonzeros, nnz should be forced to glen
                // FullyDistVec ( shared_ptr<CommGrid> grid, IT globallen, NT initval);
                FullyDistVec<int64_t, double> * denseC = new FullyDistVec<int64_t, double>( A.getcommgrid(), nvert, 1.0);
                FullyDistSpVec<int64_t, double> C ( *denseC);
                delete denseC;
 
                FullyDistSpVec<int64_t, double> min_neighbor_r ( A.getcommgrid(), nvert);
                FullyDistSpVec<int64_t, uint8_t> new_S_members ( A.getcommgrid(), nvert);
                FullyDistSpVec<int64_t, uint8_t> new_S_neighbors ( A.getcommgrid(), nvert);
 
                while (C.getnnz() > 0)
                {
                    //# label each vertex in C with a random value
                    C.Apply(randGen());
 
                    //# find the smallest random value among a vertex's neighbors
                    //# In other words: min_neighbor_r[i] = min(C[j] for all neighbors j of vertex i)
                    //min_neighbor_r = Gmatrix.SpMV(C, sr(myMin,select2nd)) # could use "min" directly
                    SpMV<LatestRetwitterMIS>(A, C, min_neighbor_r, false);  // min_neighbor_r empty OK?
                    #ifdef PRINTITERS
                    min_neighbor_r.PrintInfo("Neighbors");
                    #endif
 
                    #ifdef DEBUG
                    min_neighbor_r.DebugPrint();
                    C.DebugPrint();
                    #endif
 
                    //# The vertices to be added to S this iteration are those whose random value is
                    //# smaller than those of all its neighbors:
                    //# new_S_members[i] exists if C[i] < min_neighbor_r[i]
                    //# If C[i] exists and min_neighbor_r[i] doesn't, still a value is returned with bin_op(NULL,C[i])
                    //new_S_members = min_neighbor_r.eWiseApply(C, return1, doOp=is2ndSmaller, allowANulls=True, allowBNulls=False, inPlace=False, ANull=2)
                    new_S_members = EWiseApply<uint8_t>(min_neighbor_r, C, return1_uint8(), is2ndSmaller(), true, false, (double) 2.0,  (double) 2.0, true);
                    #ifdef PRINTITERS
                    new_S_members.PrintInfo("New members of the MIS");
                    #endif
 
                    #ifdef DEBUG
                    new_S_members.DebugPrint();
                    #endif
 
                    //# new_S_members are no longer candidates, so remove them from C
                    //C.eWiseApply(new_S_members, return1, allowANulls=False, allowIntersect=False, allowBNulls=True, inPlace=True)
                    C = EWiseApply<double>(C, new_S_members, return1_uint8(), return1_uint8(), false, true, (double) 0.0, (uint8_t) 0, false);
                    #ifdef PRINTITERS
                    C.PrintInfo("Entries to be removed from the Candidates set");
                    #endif
 
                    //# find neighbors of new_S_members
                    //new_S_neighbors = Gmatrix.SpMV(new_S_members, sr(select2nd,select2nd))
                    SpMV<LatestRetwitterSelect2nd>(A, new_S_members, new_S_neighbors, false);
 
                    //# remove neighbors of new_S_members from C, because they cannot be part of the MIS anymore
                    //# If C[i] exists and new_S_neighbors[i] doesn't, still a value is returned with bin_op(NULL,C[i])
                    //C.eWiseApply(new_S_neighbors, return1, allowANulls=False, allowIntersect=False, allowBNulls=True, inPlace=True)
                    C = EWiseApply<double>(C, new_S_neighbors, return1_uint8(), return1_uint8(), false, true, (double) 0.0, (uint8_t) 0, false);
                    #ifdef PRINTITERS
                    C.PrintInfo("Candidates set after neighbors of MIS removed");
                    #endif
 
                    //# add new_S_members to S
                    //S.eWiseApply(new_S_members, return1, allowANulls=True, allowBNulls=True, inPlace=True)
                    S = EWiseApply<uint8_t>(S, new_S_members, return1_uint8(), return1_uint8(), true, true, (uint8_t) 1, (uint8_t) 1, true);
                    S.PrintInfo("The current MIS:");
                }
                double t2 = MPI_Wtime();
 
                ostringstream ositr;
                ositr << "MIS has " << S.getnnz() << " vertices" << endl;
                SpParHelper::Print(ositr.str());
 
                ostringstream ositr2;
                ositr << "MIS time: " << t2-t1 << " seconds" << endl;
                SpParHelper::Print(ositr.str());
 
                TIMES[sruns] = t2-t1;
                MISVS[sruns] = S.getnnz();
            } // end for(int sruns = 0; sruns < ITERS; ++sruns)
 
            ostringstream os;
            os << "Per iteration communication times: " << endl;
            os << "AllGatherv: " << cblas_allgathertime / ITERS << endl;
            os << "AlltoAllv: " << cblas_alltoalltime / ITERS << endl;
 
            sort(MISVS, MISVS+ITERS);
            os << "--------------------------" << endl;
            os << "Min MIS vertices: " << MISVS[0] << endl;
            os << "Median MIS vertices: " << (MISVS[(ITERS/2)-1] + MISVS[ITERS/2])/2 << endl;
            os << "Max MIS vertices: " << MISVS[ITERS-1] << endl;
            double mean = accumulate( MISVS, MISVS+ITERS, 0.0 )/ ITERS;
            vector<double> zero_mean(ITERS);    // find distances to the mean
            transform(MISVS, MISVS+ITERS, zero_mean.begin(), bind2nd( minus<double>(), mean ));
            // self inner-product is sum of sum of squares
            double deviation = inner_product( zero_mean.begin(),zero_mean.end(), zero_mean.begin(), 0.0 );
            deviation = sqrt( deviation / (ITERS-1) );
            os << "Mean MIS vertices: " << mean << endl;
            os << "STDDEV MIS vertices: " << deviation << endl;
            os << "--------------------------" << endl;
 
            sort(TIMES,TIMES+ITERS);
            os << "Filter keeps " << static_cast<double>(keep[trials])/100.0 << " percentage of edges" << endl;
            os << "Min time: " << TIMES[0] << " seconds" << endl;
            os << "Median time: " << (TIMES[(ITERS/2)-1] + TIMES[ITERS/2])/2 << " seconds" << endl;
            os << "Max time: " << TIMES[ITERS-1] << " seconds" << endl;
            mean = accumulate( TIMES, TIMES+ITERS, 0.0 )/ ITERS;
            transform(TIMES, TIMES+ITERS, zero_mean.begin(), bind2nd( minus<double>(), mean ));
            deviation = inner_product( zero_mean.begin(),zero_mean.end(), zero_mean.begin(), 0.0 );
            deviation = sqrt( deviation / (ITERS-1) );
            os << "Mean time: " << mean << " seconds" << endl;
            os << "STDDEV time: " << deviation << " seconds" << endl;
            os << "--------------------------" << endl;
            SpParHelper::Print(os.str());
        }   // end for(int trials =0; trials < PERCENTS; trials++)
    }
    MPI_Finalize();
    return 0;
}