/* * @HEADER * * *********************************************************************** * * Zoltan Toolkit for Load-balancing, Partitioning, Ordering and Coloring * Copyright 2012 Sandia Corporation * * Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation, * the U.S. Government retains certain rights in this software. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * 3. Neither the name of the Corporation nor the names of the * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Questions? Contact Karen Devine kddevin@sandia.gov * Erik Boman egboman@sandia.gov * * *********************************************************************** * * @HEADER */ /************************************************************** * Basic example of using Zoltan to partition a graph. ***************************************************************/ #include #include #include #include #include "zoltan.h" /* Name of file containing graph to be partitioned */ static char *global_fname="graph.txt"; /* Structure to hold graph data ZOLTAN_ID_TYPE is defined when Zoltan is compiled. It's size can be obtained at runtime by a library call. (See zoltan_types.h). */ typedef struct{ int numMyVertices; /* total vertices in in my partition */ int numAllNbors; /* total number of neighbors of my vertices */ ZOLTAN_ID_TYPE *vertexGID; /* global ID of each of my vertices */ int *nborIndex; /* nborIndex[i] is location of start of neighbors for vertex i */ ZOLTAN_ID_TYPE *nborGID; /* nborGIDs[nborIndex[i]] is first neighbor of vertex i */ int *nborProc; /* process owning each nbor in nborGID */ } GRAPH_DATA; /* Application defined query functions */ static int get_number_of_vertices(void *data, int *ierr); static void get_vertex_list(void *data, int sizeGID, int sizeLID, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int wgt_dim, float *obj_wgts, int *ierr); static void get_num_edges_list(void *data, int sizeGID, int sizeLID, int num_obj, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int *numEdges, int *ierr); static void get_edge_list(void *data, int sizeGID, int sizeLID, int num_obj, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int *num_edges, ZOLTAN_ID_PTR nborGID, int *nborProc, int wgt_dim, float *ewgts, int *ierr); /* Functions to read graph in from file, distribute it, view it, handle errors */ static int get_next_line(FILE *fp, char *buf, int bufsize); static int get_line_ints(char *buf, int bufsize, int *vals); static void input_file_error(int numProcs, int tag, int startProc); static void showGraphPartitions(int myProc, int numIDs, ZOLTAN_ID_TYPE *GIDs, int *parts, int nparts); static void read_input_file(int myRank, int numProcs, char *fname, GRAPH_DATA *myData); static unsigned int simple_hash(unsigned int *key, unsigned int n); int main(int argc, char *argv[]) { int i, rc; float ver; struct Zoltan_Struct *zz; int changes, numGidEntries, numLidEntries, numImport, numExport; int myRank, numProcs; ZOLTAN_ID_PTR importGlobalGids, importLocalGids, exportGlobalGids, exportLocalGids; int *importProcs, *importToPart, *exportProcs, *exportToPart; int *parts = NULL; FILE *fp; GRAPH_DATA myGraph; /****************************************************************** ** Initialize MPI and Zoltan ******************************************************************/ MPI_Init(&argc, &argv); MPI_Comm_rank(MPI_COMM_WORLD, &myRank); MPI_Comm_size(MPI_COMM_WORLD, &numProcs); rc = Zoltan_Initialize(argc, argv, &ver); if (rc != ZOLTAN_OK){ printf("sorry...\n"); MPI_Finalize(); exit(0); } /****************************************************************** ** Read graph from input file and distribute it ******************************************************************/ fp = fopen(global_fname, "r"); if (!fp){ if (myRank == 0) fprintf(stderr,"ERROR: Can not open %s\n",global_fname); MPI_Finalize(); exit(1); } fclose(fp); read_input_file(myRank, numProcs, global_fname, &myGraph); /****************************************************************** ** Create a Zoltan library structure for this instance of load ** balancing. Set the parameters and query functions that will ** govern the library's calculation. See the Zoltan User's ** Guide for the definition of these and many other parameters. ******************************************************************/ zz = Zoltan_Create(MPI_COMM_WORLD); /* General parameters */ Zoltan_Set_Param(zz, "DEBUG_LEVEL", "0"); Zoltan_Set_Param(zz, "LB_METHOD", "GRAPH"); Zoltan_Set_Param(zz, "LB_APPROACH", "PARTITION"); Zoltan_Set_Param(zz, "NUM_GID_ENTRIES", "1"); Zoltan_Set_Param(zz, "NUM_LID_ENTRIES", "1"); Zoltan_Set_Param(zz, "RETURN_LISTS", "ALL"); /* Graph parameters */ Zoltan_Set_Param(zz, "CHECK_GRAPH", "2"); Zoltan_Set_Param(zz, "PHG_EDGE_SIZE_THRESHOLD", ".35"); /* 0-remove all, 1-remove none */ /* Query functions - defined in simpleQueries.h */ Zoltan_Set_Num_Obj_Fn(zz, get_number_of_vertices, &myGraph); Zoltan_Set_Obj_List_Fn(zz, get_vertex_list, &myGraph); Zoltan_Set_Num_Edges_Multi_Fn(zz, get_num_edges_list, &myGraph); Zoltan_Set_Edge_List_Multi_Fn(zz, get_edge_list, &myGraph); /****************************************************************** ** Zoltan can now partition the simple graph. ** In this simple example, we assume the number of partitions is ** equal to the number of processes. Process rank 0 will own ** partition 0, process rank 1 will own partition 1, and so on. ******************************************************************/ rc = Zoltan_LB_Partition(zz, /* input (all remaining fields are output) */ &changes, /* 1 if partitioning was changed, 0 otherwise */ &numGidEntries, /* Number of integers used for a global ID */ &numLidEntries, /* Number of integers used for a local ID */ &numImport, /* Number of vertices to be sent to me */ &importGlobalGids, /* Global IDs of vertices to be sent to me */ &importLocalGids, /* Local IDs of vertices to be sent to me */ &importProcs, /* Process rank for source of each incoming vertex */ &importToPart, /* New partition for each incoming vertex */ &numExport, /* Number of vertices I must send to other processes*/ &exportGlobalGids, /* Global IDs of the vertices I must send */ &exportLocalGids, /* Local IDs of the vertices I must send */ &exportProcs, /* Process to which I send each of the vertices */ &exportToPart); /* Partition to which each vertex will belong */ if (rc != ZOLTAN_OK){ printf("sorry...\n"); MPI_Finalize(); Zoltan_Destroy(&zz); exit(0); } /****************************************************************** ** Visualize the graph partitioning before and after calling Zoltan. ******************************************************************/ parts = (int *)malloc(sizeof(int) * myGraph.numMyVertices); for (i=0; i < myGraph.numMyVertices; i++){ parts[i] = myRank; } if (myRank== 0){ printf("\nGraph partition before calling Zoltan\n"); } showGraphPartitions(myRank, myGraph.numMyVertices, myGraph.vertexGID, parts, numProcs); for (i=0; i < numExport; i++){ parts[exportLocalGids[i]] = exportToPart[i]; } if (myRank == 0){ printf("Graph partition after calling Zoltan\n"); } showGraphPartitions(myRank, myGraph.numMyVertices, myGraph.vertexGID, parts, numProcs); if (parts) free(parts); /****************************************************************** ** Free the arrays allocated by Zoltan_LB_Partition, and free ** the storage allocated for the Zoltan structure. ******************************************************************/ Zoltan_LB_Free_Part(&importGlobalGids, &importLocalGids, &importProcs, &importToPart); Zoltan_LB_Free_Part(&exportGlobalGids, &exportLocalGids, &exportProcs, &exportToPart); Zoltan_Destroy(&zz); /********************** ** all done *********** **********************/ MPI_Finalize(); if (myGraph.numMyVertices > 0){ free(myGraph.vertexGID); free(myGraph.nborIndex); if (myGraph.numAllNbors > 0){ free(myGraph.nborGID); free(myGraph.nborProc); } } return 0; } /* Application defined query functions */ static int get_number_of_vertices(void *data, int *ierr) { GRAPH_DATA *graph = (GRAPH_DATA *)data; *ierr = ZOLTAN_OK; return graph->numMyVertices; } static void get_vertex_list(void *data, int sizeGID, int sizeLID, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int wgt_dim, float *obj_wgts, int *ierr) { int i; GRAPH_DATA *graph = (GRAPH_DATA *)data; *ierr = ZOLTAN_OK; /* In this example, return the IDs of our vertices, but no weights. * Zoltan will assume equally weighted vertices. */ for (i=0; inumMyVertices; i++){ globalID[i] = graph->vertexGID[i]; localID[i] = i; } } static void get_num_edges_list(void *data, int sizeGID, int sizeLID, int num_obj, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int *numEdges, int *ierr) { int i, idx; GRAPH_DATA *graph = (GRAPH_DATA *)data; if ( (sizeGID != 1) || (sizeLID != 1) || (num_obj != graph->numMyVertices)){ *ierr = ZOLTAN_FATAL; return; } for (i=0; i < num_obj ; i++){ idx = localID[i]; numEdges[i] = graph->nborIndex[idx+1] - graph->nborIndex[idx]; } *ierr = ZOLTAN_OK; return; } static void get_edge_list(void *data, int sizeGID, int sizeLID, int num_obj, ZOLTAN_ID_PTR globalID, ZOLTAN_ID_PTR localID, int *num_edges, ZOLTAN_ID_PTR nborGID, int *nborProc, int wgt_dim, float *ewgts, int *ierr) { int i, j, from, to; int *nextProc; ZOLTAN_ID_TYPE *nextNbor; GRAPH_DATA *graph = (GRAPH_DATA *)data; *ierr = ZOLTAN_OK; if ( (sizeGID != 1) || (sizeLID != 1) || (num_obj != graph->numMyVertices)|| (wgt_dim != 0)){ *ierr = ZOLTAN_FATAL; return; } nextNbor = nborGID; nextProc = nborProc; for (i=0; i < num_obj; i++){ /* * In this example, we are not setting edge weights. Zoltan will * set each edge to weight 1.0. */ to = graph->nborIndex[localID[i]+1]; from = graph->nborIndex[localID[i]]; if ((to - from) != num_edges[i]){ *ierr = ZOLTAN_FATAL; return; } for (j=from; j < to; j++){ *nextNbor++ = graph->nborGID[j]; *nextProc++ = graph->nborProc[j]; } } return; } /* Function to find next line of information in input file */ static int get_next_line(FILE *fp, char *buf, int bufsize) { int i, cval, len; char *c; while (1){ c = fgets(buf, bufsize, fp); if (c == NULL) return 0; /* end of file */ len = strlen(c); for (i=0, c=buf; i < len; i++, c++){ cval = (int)*c; if (isspace(cval) == 0) break; } if (i == len) continue; /* blank line */ if (*c == '#') continue; /* comment */ if (c != buf){ strcpy(buf, c); } break; } return strlen(buf); /* number of characters */ } /* Function to return the list of non-negative integers in a line */ static int get_line_ints(char *buf, int bufsize, int *vals) { char *c = buf; int count=0; while (1){ if ( (c-buf) >= bufsize) break; while (!(isdigit(*c))){ if ((c - buf) >= bufsize) break; c++; } if ( (c-buf) >= bufsize) break; vals[count++] = atoi(c); while (isdigit(*c)){ if ((c - buf) >= bufsize) break; c++; } if ( (c-buf) >= bufsize) break; } return count; } /* Proc 0 notifies others of error and exits */ static void input_file_error(int numProcs, int tag, int startProc) { int i, val[2]; val[0] = -1; /* error flag */ fprintf(stderr,"ERROR in input file.\n"); for (i=startProc; i < numProcs; i++){ /* these procs have posted a receive for "tag" expecting counts */ MPI_Send(val, 2, MPI_INT, i, tag, MPI_COMM_WORLD); } for (i=1; i < startProc; i++){ /* these procs are done and waiting for ok-to-go */ MPI_Send(val, 1, MPI_INT, i, 0, MPI_COMM_WORLD); } MPI_Finalize(); exit(1); } /* Draw the partition assignments of the objects */ static void showGraphPartitions(int myProc, int numIDs, ZOLTAN_ID_TYPE *GIDs, int *parts, int nparts) { int partAssign[25], allPartAssign[25]; int i, j, part, cuts, prevPart=-1; float imbal, localImbal, sum; int *partCount; memset(partAssign, 0, sizeof(int) * 25); for (i=0; i < numIDs; i++){ partAssign[GIDs[i]-1] = parts[i]; } MPI_Reduce(partAssign, allPartAssign, 25, MPI_INT, MPI_MAX, 0, MPI_COMM_WORLD); if (myProc == 0){ partCount = (int *)calloc(sizeof(int), nparts); cuts = 0; for (i=20; i >= 0; i-=5){ for (j=0; j < 5; j++){ part = allPartAssign[i + j]; partCount[part]++; if (j > 0){ if (part == prevPart){ printf("-----%d",part); } else{ printf("--x--%d",part); cuts++; prevPart = part; } } else{ printf("%d",part); prevPart = part; } } printf("\n"); if (i > 0){ for (j=0; j < 5; j++){ if (allPartAssign[i+j] != allPartAssign[i+j-5]){ printf("x "); cuts++; } else{ printf("| "); } } printf("\n"); } } printf("\n"); for (sum=0, i=0; i < nparts; i++){ sum += partCount[i]; } imbal = 0; for (i=0; i < nparts; i++){ /* An imbalance measure. 1.0 is perfect balance, larger is worse */ localImbal = (nparts * partCount[i]) / sum; if (localImbal > imbal) imbal = localImbal; } printf("Object imbalance (1.0 perfect, larger numbers are worse): %f\n",imbal); printf("Total number of edge cuts: %d\n\n",cuts); if (nparts) free(partCount); } } /* * Read the graph in the input file and distribute the vertices. */ void read_input_file(int myRank, int numProcs, char *fname, GRAPH_DATA *graph) { char buf[512]; int bufsize; int numGlobalVertices, numGlobalNeighbors; int num, nnbors, ack=0; int vGID; int i, j, procID; int vals[128], send_count[2]; int *idx; unsigned int id; FILE *fp; MPI_Status status; int ack_tag = 5, count_tag = 10, id_tag = 15; GRAPH_DATA *send_graph; if (myRank == 0){ bufsize = 512; fp = fopen(fname, "r"); /* Get the number of vertices */ num = get_next_line(fp, buf, bufsize); if (num == 0) input_file_error(numProcs, count_tag, 1); num = sscanf(buf, "%d", &numGlobalVertices); if (num != 1) input_file_error(numProcs, count_tag, 1); /* Get the number of vertex neighbors */ num = get_next_line(fp, buf, bufsize); if (num == 0) input_file_error(numProcs, count_tag, 1); num = sscanf(buf, "%d", &numGlobalNeighbors); if (num != 1) input_file_error(numProcs, count_tag, 1); /* Allocate arrays to read in entire graph */ graph->vertexGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * numGlobalVertices); graph->nborIndex = (int *)malloc(sizeof(int) * (numGlobalVertices + 1)); graph->nborGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * numGlobalNeighbors); graph->nborProc = (int *)malloc(sizeof(int) * numGlobalNeighbors); graph->nborIndex[0] = 0; for (i=0; i < numGlobalVertices; i++){ num = get_next_line(fp, buf, bufsize); if (num == 0) input_file_error(numProcs, count_tag, 1); num = get_line_ints(buf, strlen(buf), vals); if (num < 2) input_file_error(numProcs, count_tag, 1); vGID = vals[0]; nnbors = vals[1]; if (num < (nnbors + 2)) input_file_error(numProcs, count_tag, 1); graph->vertexGID[i] = (ZOLTAN_ID_TYPE)vGID; for (j=0; j < nnbors; j++){ graph->nborGID[graph->nborIndex[i] + j] = (ZOLTAN_ID_TYPE)vals[2 + j]; } graph->nborIndex[i+1] = graph->nborIndex[i] + nnbors; } fclose(fp); /* Assign each vertex to a process using a hash function */ for (i=0; i nborGID[i]; graph->nborProc[i] = simple_hash(&id, numProcs); } /* Create a sub graph for each process */ send_graph = (GRAPH_DATA *)calloc(sizeof(GRAPH_DATA) , numProcs); for (i=0; i < numGlobalVertices; i++){ id = (unsigned int)graph->vertexGID[i]; procID = simple_hash(&id, numProcs); send_graph[procID].numMyVertices++; } for (i=0; i < numProcs; i++){ num = send_graph[i].numMyVertices; send_graph[i].vertexGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * num); send_graph[i].nborIndex = (int *)calloc(sizeof(int) , (num + 1)); } idx = (int *)calloc(sizeof(int), numProcs); for (i=0; i < numGlobalVertices; i++){ id = (unsigned int)graph->vertexGID[i]; nnbors = graph->nborIndex[i+1] - graph->nborIndex[i]; procID = simple_hash(&id, numProcs); j = idx[procID]; send_graph[procID].vertexGID[j] = (ZOLTAN_ID_TYPE)id; send_graph[procID].nborIndex[j+1] = send_graph[procID].nborIndex[j] + nnbors; idx[procID] = j+1; } for (i=0; i < numProcs; i++){ num = send_graph[i].nborIndex[send_graph[i].numMyVertices]; send_graph[i].nborGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * num); send_graph[i].nborProc= (int *)malloc(sizeof(int) * num); send_graph[i].numAllNbors = num; } memset(idx, 0, sizeof(int) * numProcs); for (i=0; i < numGlobalVertices; i++){ id = (unsigned int)graph->vertexGID[i]; nnbors = graph->nborIndex[i+1] - graph->nborIndex[i]; procID = simple_hash(&id, numProcs); j = idx[procID]; if (nnbors > 0){ memcpy(send_graph[procID].nborGID + j, graph->nborGID + graph->nborIndex[i], nnbors * sizeof(ZOLTAN_ID_TYPE)); memcpy(send_graph[procID].nborProc + j, graph->nborProc + graph->nborIndex[i], nnbors * sizeof(int)); idx[procID] = j + nnbors; } } free(idx); /* Process zero sub-graph */ free(graph->vertexGID); free(graph->nborIndex); free(graph->nborGID); free(graph->nborProc); *graph = send_graph[0]; /* Send other processes their subgraph */ for (i=1; i < numProcs; i++){ send_count[0] = send_graph[i].numMyVertices; send_count[1] = send_graph[i].numAllNbors; MPI_Send(send_count, 2, MPI_INT, i, count_tag, MPI_COMM_WORLD); MPI_Recv(&ack, 1, MPI_INT, i, ack_tag, MPI_COMM_WORLD, &status); if (send_count[0] > 0){ MPI_Send(send_graph[i].vertexGID, send_count[0], ZOLTAN_ID_MPI_TYPE, i, id_tag, MPI_COMM_WORLD); free(send_graph[i].vertexGID); MPI_Send(send_graph[i].nborIndex, send_count[0] + 1, MPI_INT, i, id_tag + 1, MPI_COMM_WORLD); free(send_graph[i].nborIndex); if (send_count[1] > 0){ MPI_Send(send_graph[i].nborGID, send_count[1], ZOLTAN_ID_MPI_TYPE, i, id_tag + 2, MPI_COMM_WORLD); free(send_graph[i].nborGID); MPI_Send(send_graph[i].nborProc, send_count[1], MPI_INT, i, id_tag + 3, MPI_COMM_WORLD); free(send_graph[i].nborProc); } } } free(send_graph); /* signal all procs it is OK to go on */ ack = 0; for (i=1; i < numProcs; i++){ MPI_Send(&ack, 1, MPI_INT, i, 0, MPI_COMM_WORLD); } } else{ MPI_Recv(send_count, 2, MPI_INT, 0, count_tag, MPI_COMM_WORLD, &status); if (send_count[0] < 0){ MPI_Finalize(); exit(1); } ack = 0; graph->numMyVertices = send_count[0]; graph->numAllNbors = send_count[1]; if (send_count[0] > 0){ graph->vertexGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * send_count[0]); graph->nborIndex = (int *)malloc(sizeof(int) * (send_count[0] + 1)); if (send_count[1] > 0){ graph->nborGID = (ZOLTAN_ID_TYPE *)malloc(sizeof(ZOLTAN_ID_TYPE) * send_count[1]); graph->nborProc = (int *)malloc(sizeof(int) * send_count[1]); } } MPI_Send(&ack, 1, MPI_INT, 0, ack_tag, MPI_COMM_WORLD); if (send_count[0] > 0){ MPI_Recv(graph->vertexGID,send_count[0],ZOLTAN_ID_MPI_TYPE, 0, id_tag, MPI_COMM_WORLD, &status); MPI_Recv(graph->nborIndex,send_count[0] + 1, MPI_INT, 0, id_tag + 1, MPI_COMM_WORLD, &status); if (send_count[1] > 0){ MPI_Recv(graph->nborGID,send_count[1], ZOLTAN_ID_MPI_TYPE, 0, id_tag + 2, MPI_COMM_WORLD, &status); MPI_Recv(graph->nborProc,send_count[1], MPI_INT, 0, id_tag + 3, MPI_COMM_WORLD, &status); } } /* ok to go on? */ MPI_Recv(&ack, 1, MPI_INT, 0, 0, MPI_COMM_WORLD, &status); if (ack < 0){ MPI_Finalize(); exit(1); } } } unsigned int simple_hash(unsigned int *key, unsigned int n) { unsigned int h, rest, *p, bytes, num_bytes; char *byteptr; num_bytes = (unsigned int) sizeof(int); /* First hash the int-sized portions of the key */ h = 0; for (p = (unsigned int *)key, bytes=num_bytes; bytes >= (unsigned int) sizeof(int); bytes-=sizeof(int), p++){ h = (h*2654435761U) ^ (*p); } /* Then take care of the remaining bytes, if any */ rest = 0; for (byteptr = (char *)p; bytes > 0; bytes--, byteptr++){ rest = (rest<<8) | (*byteptr); } /* Merge the two parts */ if (rest) h = (h*2654435761U) ^ rest; /* Return h mod n */ return (h%n); }