X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/33eeb4be26b53ed1561488ce78e392b40ecf74df..a5df3137d0c7edb988478158a8237c0e849c8594:/examples/gras/pmm/pmm.c

diff --git a/examples/gras/pmm/pmm.c b/examples/gras/pmm/pmm.c
index 8dc7235fb6..969fe8500b 100755
--- a/examples/gras/pmm/pmm.c
+++ b/examples/gras/pmm/pmm.c
@@ -10,10 +10,13 @@
 
 #include "gras.h"
 #include "xbt/matrix.h"
-#define PROC_MATRIX_SIZE 2
+#include "amok/peermanagement.h"
+
+#define PROC_MATRIX_SIZE 3
+#define NEIGHBOR_COUNT PROC_MATRIX_SIZE - 1
 #define SLAVE_COUNT (PROC_MATRIX_SIZE*PROC_MATRIX_SIZE)
 
-#define DATA_MATRIX_SIZE 4
+#define DATA_MATRIX_SIZE 18
 const int submatrix_size = DATA_MATRIX_SIZE/PROC_MATRIX_SIZE;
 
 XBT_LOG_NEW_DEFAULT_CATEGORY(pmm,"Parallel Matrix Multiplication");
@@ -27,43 +30,34 @@ GRAS_DEFINE_TYPE(s_result,struct s_result {
 typedef struct s_result result_t;
 
 /* struct to send initial data to slave */
-GRAS_DEFINE_TYPE(s_assignment,struct s_assignment {
+GRAS_DEFINE_TYPE(s_pmm_assignment,struct s_pmm_assignment {
   int linepos;
   int rowpos;
-  xbt_host_t line[PROC_MATRIX_SIZE];
-  xbt_host_t row[PROC_MATRIX_SIZE];
+  xbt_peer_t line[NEIGHBOR_COUNT];
+  xbt_peer_t row[NEIGHBOR_COUNT];
   xbt_matrix_t A GRAS_ANNOTE(subtype,double);
   xbt_matrix_t B GRAS_ANNOTE(subtype,double);
 });
-typedef struct s_assignment s_assignment_t;
+typedef struct s_pmm_assignment s_pmm_assignment_t;
 
 /* register messages which may be sent (common to client and server) */
 static void register_messages(void) {
   gras_datadesc_type_t result_type;
-  gras_datadesc_type_t assignment_type;
+  gras_datadesc_type_t pmm_assignment_type;
 
-  gras_datadesc_set_const("PROC_MATRIX_SIZE",PROC_MATRIX_SIZE);
+  gras_datadesc_set_const("NEIGHBOR_COUNT",NEIGHBOR_COUNT);
   result_type=gras_datadesc_by_symbol(s_result);
-  assignment_type=gras_datadesc_by_symbol(s_assignment);
+  pmm_assignment_type=gras_datadesc_by_symbol(s_pmm_assignment);
 	
   /* receive a final result from slave */
   gras_msgtype_declare("result", result_type);
 
   /* send from master to slave to assign a position and some data */
-  gras_msgtype_declare("assignment", assignment_type);
-
-  /* send from master to slave to ask a final result */
-  gras_msgtype_declare("ask_result", gras_datadesc_by_name("int")); 
+  gras_msgtype_declare("pmm_slave", pmm_assignment_type);
 
-  /* send from master to slave to indicate the begining of step */
-  gras_msgtype_declare("step", gras_datadesc_by_name("int"));
-  /* send from slave to master to indicate the end of the current step */
-  gras_msgtype_declare("step_ack", gras_datadesc_by_name("int"));
-
-  /* send data between slave */
-  gras_msgtype_declare("dataA", gras_datadesc_by_name("double"));
-  /* send data between slave */
-  gras_msgtype_declare("dataB", gras_datadesc_by_name("double"));
+  /* send data between slaves */
+  gras_msgtype_declare("dataA", gras_datadesc_matrix(gras_datadesc_by_name("double"),NULL));
+  gras_msgtype_declare("dataB", gras_datadesc_matrix(gras_datadesc_by_name("double"),NULL));
 }
 
 /* Function prototypes */
@@ -83,87 +77,83 @@ typedef struct {
 } master_data_t;
 
 
-/***  Function Scatter Sequentiel ***/
-
-static void scatter(){
-
-}/* end_of_Scatter */
-
-/***  Function: Scatter // ***/
-
-static void scatter_parl(){
-
-}/* end_of_Scatter // */
-
-/***  Function: multiplication ***/
-
-static void multiplication(){
-
-}/* end_of_multiplication */
-
-/***  Function: gather ***/
-
-static void gather(){
-
-}/* end_of_gather */
-
 int master (int argc,char *argv[]) {
 
-  xbt_ex_t e;
-
-  int i,port,ask_result,step;
+  int i;
 
   xbt_matrix_t A,B,C;
   result_t result;
 
   gras_socket_t from;
 
-  /*  Init the GRAS's infrastructure */
-  gras_init(&argc, argv);
-
-  xbt_host_t grid[SLAVE_COUNT]; /* The slaves */
+  xbt_dynar_t peers; /* group of slaves */
+  xbt_peer_t grid[SLAVE_COUNT]; /* The slaves as an array */
   gras_socket_t socket[SLAVE_COUNT]; /* sockets for brodcast to slaves */
 
-  /*  Initialize Matrices */
-
+  /* Init the GRAS's infrastructure */
+  gras_init(&argc, argv);
+  amok_pm_init();
+  register_messages();
+      
+  /* Initialize data matrices */
   A = xbt_matrix_double_new_id(DATA_MATRIX_SIZE,DATA_MATRIX_SIZE);
   B = xbt_matrix_double_new_seq(DATA_MATRIX_SIZE,DATA_MATRIX_SIZE);
   C = xbt_matrix_double_new_zeros(DATA_MATRIX_SIZE,DATA_MATRIX_SIZE);
 	
-  //xbt_matrix_dump(B,"B:seq",0,xbt_matrix_dump_display_double);
-
-
-  /*  Get arguments and create sockets */
-  port=atoi(argv[1]);
-  //scatter();
-  //scatter_parl();
-  //multiplication();
-  //gather();
-  //display(A);
-  /************************* Init Data Send *********************************/
-  int step_ack;
-  gras_os_sleep(5);
-
-  for( i=1;i<argc && i<=SLAVE_COUNT;i++){
-    grid[i-1]=xbt_host_from_string(argv[i]);
-    socket[i-1]=gras_socket_client(grid[i-1]->name,grid[i-1]->port);
-      
-    INFO2("Connected to %s:%d.",grid[i-1]->name,grid[i-1]->port);
+  /* Create the connexions */
+  xbt_assert0(argc>1, "Usage: master <port>");
+  gras_socket_server(atoi(argv[1]));
+  peers=amok_pm_group_new("pmm");
+   
+  /* friends, we're ready. Come and play */
+  INFO0("Wait for peers for 5 sec");
+  gras_msg_handleall(5);
+  INFO1("Got %ld pals",xbt_dynar_length(peers));
+
+  for (i=0;
+       i<xbt_dynar_length(peers) && i<SLAVE_COUNT;
+       i++) {
+
+    xbt_dynar_get_cpy(peers,i,&grid[i]);
+    socket[i]=gras_socket_client(grid[i]->name,grid[i]->port);
   }
-  /* FIXME: let the surnumerous slave die properly */
+  xbt_assert2(i==SLAVE_COUNT,
+	      "Not enough slaves for this setting (got %d of %d). Change the deployment file",
+	      i,SLAVE_COUNT);
+
+  /* Kill surnumerous slaves */
+  for (i=SLAVE_COUNT; i<xbt_dynar_length(peers); ) {
+    xbt_peer_t h;
+
+    xbt_dynar_remove_at(peers,i,&h);
+    INFO2("Too much slaves. Killing %s:%d", h->name, h->port);
+    amok_pm_kill_hp(h->name,h->port);
+    free(h);
+  }
+
 
+  /* Assign job to slaves */
   int row=0, line=0;
+  INFO0("XXXXXXXXXXXXXXXXXXXXXX begin Multiplication");
   for(i=0 ; i<SLAVE_COUNT; i++){
-    s_assignment_t assignment;
-    int j;
+    s_pmm_assignment_t assignment;
+    int j,k;
 
     assignment.linepos=line; // assigned line
     assignment.rowpos=row;   // assigned row
 
     /* Neiborhood */
-    for (j=0; j<PROC_MATRIX_SIZE; j++) {
-      assignment.row[j] = grid[ j*PROC_MATRIX_SIZE+(row) ] ;
-      assignment.line[j] =  grid[ (line)*PROC_MATRIX_SIZE+j ] ;
+    for (j=0,k=0; j<PROC_MATRIX_SIZE; j++) {
+      if (i != j*PROC_MATRIX_SIZE+(row)) {	    
+	 assignment.row[k] = grid[ j*PROC_MATRIX_SIZE+(row) ] ;
+	 k++;
+      }
+    }
+    for (j=0,k=0; j<PROC_MATRIX_SIZE; j++) {
+      if (i != (line)*PROC_MATRIX_SIZE+j) {	    
+	 assignment.line[k] =  grid[ (line)*PROC_MATRIX_SIZE+j ] ;
+	 k++;
+      }
     }
 
     assignment.A=xbt_matrix_new_sub(A,
@@ -174,67 +164,48 @@ int master (int argc,char *argv[]) {
 				    submatrix_size,submatrix_size,
 				    submatrix_size*line,submatrix_size*row,
 				    NULL);
-    //xbt_matrix_dump(assignment.B,"assignment.B",0,xbt_matrix_dump_display_double);
     row++;
     if (row >= PROC_MATRIX_SIZE) {
       row=0;
       line++;
     }
 		
-    gras_msg_send(socket[i],gras_msgtype_by_name("assignment"),&assignment);
-    //    INFO3("Send assignment to %s : data A= %.3g & data B= %.3g",
-    //	  gras_socket_peer_name(socket[i]),mydata.a,mydata.b);
-
+    gras_msg_send(socket[i],"pmm_slave",&assignment);
+    xbt_matrix_free(assignment.A);
+    xbt_matrix_free(assignment.B);
   }
-  // end assignment
-
-  /******************************* multiplication ********************************/
-  INFO0("XXXXXXXXXXXXXXXXXXXXXX begin Multiplication");
-	
-  for (step=0; step < PROC_MATRIX_SIZE; step++){
-    for (i=0; i< SLAVE_COUNT; i++){
-      TRY {
-	gras_msg_send(socket[i], gras_msgtype_by_name("step"), &step);
-      } CATCH(e) {
-	gras_socket_close(socket[i]);
-	RETHROW0("Unable to send the msg : %s");
-      }
-    }
-    INFO1("XXXXXX Next step (%d)",step);
-
-    /* wait for computing and slave messages exchange */
 
-    i=0;	
-    while  ( i< SLAVE_COUNT) {
-      TRY {
-	gras_msg_wait(1300,gras_msgtype_by_name("step_ack"),&from,&step_ack);
-      } CATCH(e) {
-	RETHROW0("Can't get a Ack step message from slave : %s");
-      }
-      i++;
-      DEBUG3("Got step ack from %s (got %d of %d)",
-	    gras_socket_peer_name(from), i, SLAVE_COUNT);
-    }
-  }
-  /*********************************  gather ***************************************/
+  /* (have a rest while the slave perform the multiplication) */
 
-  ask_result=0;
-  for( i=1;i< argc;i++){
-    gras_msg_send(socket[i],gras_msgtype_by_name("ask_result"),&ask_result);
-    INFO1("Send (Ask Result) message to %s",gras_socket_peer_name(socket[i]));
-  }
-  /* wait for results */
-  for( i=1;i< argc;i++){
-    gras_msg_wait(600,gras_msgtype_by_name("result"),&from,&result);
-    xbt_matrix_copy_values(result.C,C,   submatrix_size,submatrix_size,
+  /* Retrieve the results */
+  for( i=0;i< SLAVE_COUNT;i++){
+    gras_msg_wait(6000,"result",&from,&result);
+    VERB2("%d slaves are done already. Waiting for %d",i+1, SLAVE_COUNT);
+    xbt_matrix_copy_values(C,result.C,   submatrix_size,submatrix_size,
 			   submatrix_size*result.linepos,
 			   submatrix_size*result.rowpos,
 			   0,0,NULL);
+    xbt_matrix_free(result.C);
   }
   /*    end of gather   */
-  INFO0 ("The Result of Multiplication is :");
-  xbt_matrix_dump(C,"C:res",0,xbt_matrix_dump_display_double);
 
+  if (DATA_MATRIX_SIZE < 30) {
+     INFO0 ("The Result of Multiplication is :");
+     xbt_matrix_dump(C,"C:res",0,xbt_matrix_dump_display_double);
+  } else {
+     INFO1("Matrix size too big (%d>30) to be displayed here",DATA_MATRIX_SIZE);
+  }
+
+  amok_pm_group_shutdown ("pmm");   /* Ok, we're out of here */
+
+  for(i=0; i<SLAVE_COUNT; i++) {
+     gras_socket_close(socket[i]);
+  }
+   
+  xbt_matrix_free(A);
+  xbt_matrix_free(B);
+  xbt_matrix_free(C);
+  gras_exit();
   return 0;
 } /* end_of_master */
 
@@ -242,11 +213,14 @@ int master (int argc,char *argv[]) {
  * slave code
  * **********************************************************************/
 
-int slave(int argc,char *argv[]) {
+static int pmm_worker_cb(gras_msg_cb_ctx_t ctx, void *payload) {
+  /* Recover my initialized Data and My Position*/
+  s_pmm_assignment_t assignment = *(s_pmm_assignment_t*)payload;
+  gras_socket_t master = gras_msg_cb_ctx_from(ctx);
 
   xbt_ex_t e; 
 
-  int step,port,l,result_ack=0; 
+  int step,l;
   xbt_matrix_t bA=xbt_matrix_new(submatrix_size,submatrix_size,
 				 sizeof(double),NULL);
   xbt_matrix_t bB=xbt_matrix_new(submatrix_size,submatrix_size,
@@ -258,166 +232,151 @@ int slave(int argc,char *argv[]) {
   
   result_t result;
  
-  gras_socket_t from,sock;  /* to exchange data with my neighbor */
-  gras_socket_t master;     /* for the barrier */
+  gras_socket_t from;  /* to exchange data with my neighbor */
 
   /* sockets for brodcast to other slave */
   gras_socket_t socket_line[PROC_MATRIX_SIZE-1];
   gras_socket_t socket_row[PROC_MATRIX_SIZE-1];
-
-  /* Init the GRAS's infrastructure */
-
-  gras_init(&argc, argv);
-
-  /* Get arguments and create sockets */
-
-  port=atoi(argv[1]);
-  
-  /*  Create my master socket */
-  sock = gras_socket_server(port);
+  memset(socket_line,0,sizeof(socket_line));
+  memset(socket_row,0,sizeof(socket_row));
+   
   int i;
 
-  /*  Register the known messages */
-  register_messages();
+  gras_os_sleep(1); /* wait for my pals */
 
-  /* Recover my initialized Data and My Position*/
-  s_assignment_t assignment;
-  INFO2("Launch %s (port=%d); wait for my enrole message",argv[0],port);
-  TRY {
-    gras_msg_wait(600,gras_msgtype_by_name("assignment"),&master,&assignment);
-  } CATCH(e) {
-    RETHROW0("Can't get my assignment from master : %s");
-  }
   myline  = assignment.linepos;
   myrow   = assignment.rowpos;
   mydataA = assignment.A;
   mydataB = assignment.B;
-  INFO1("mydataB=%p",mydataB);
 
   INFO2("Receive my pos (%d,%d) and assignment",myline,myrow);
 
-  //  xbt_matrix_dump(mydataB,"myB",0,xbt_matrix_dump_display_double);
-
   /* Get my neighborhood from the assignment message (skipping myself) */
-  int j=0;
-  for (i=0,j=0 ; i<PROC_MATRIX_SIZE ; i++){
-    if (strcmp(gras_os_myname(),assignment.line[i]->name)) {
-      socket_line[j]=gras_socket_client(assignment.line[i]->name,
-					assignment.line[i]->port);
-      j++;
-    }
-    xbt_host_free(assignment.line[i]);
+  for (i=0 ; i<PROC_MATRIX_SIZE-1 ; i++){
+    socket_line[i]=gras_socket_client(assignment.line[i]->name,
+				      assignment.line[i]->port);
+    xbt_peer_free(assignment.line[i]);
   }
-  for (i=0,j=0 ; i<PROC_MATRIX_SIZE ; i++){
-    if (strcmp(gras_os_myname(),assignment.row[i]->name)) {
-      socket_row[j]=gras_socket_client(assignment.row[i]->name,
-				       assignment.row[i]->port);
-      j++;
-    }
-    xbt_host_free(assignment.row[i]);    
+  for (i=0 ; i<PROC_MATRIX_SIZE-1 ; i++){
+    socket_row[i]=gras_socket_client(assignment.row[i]->name,
+				     assignment.row[i]->port);
+    xbt_peer_free(assignment.row[i]);    
   }
 
-  
-  do {  //repeat until compute Cb
+  for (step=0; step<PROC_MATRIX_SIZE;step++) {
 	
-    TRY {
-      gras_msg_wait(200,gras_msgtype_by_name("step"),NULL,&step);
-    } CATCH(e) {
-      RETHROW0("Can't get a Next Step message from master : %s");
-    }
-    INFO1("Receive a step message from master: step = %d ",step);
-
     /* a line brodcast */
-    gras_os_sleep(3);  // IL FAUT EXPRIMER LE TEMPS D'ATTENTE EN FONCTION DE "SLAVE_COUNT"
-
     if(myline==step){
-      INFO2("step(%d) = Myline(%d)",step,myline);
-      for (l=0;l < PROC_MATRIX_SIZE-1 ;l++){
-	INFO1("mydataB=%p",mydataB);
-	gras_msg_send(socket_row[l], 
-		      gras_msgtype_by_name("dataB"), 
-		      &mydataB);
-	INFO1("mydataB=%p",mydataB);
-	
-	xbt_matrix_free(bB);
-	INFO1("mydataB=%p",mydataB);
-	xbt_matrix_dump(mydataB,"myB",0,xbt_matrix_dump_display_double);
-	bB = xbt_matrix_new_sub(mydataB,
-				submatrix_size,submatrix_size,
-				0,0,NULL);
+       INFO3("LINE: step(%d) = Myline(%d). Broadcast my data (myport=%d).",
+	     step,myline,gras_os_myport());
+       for (l=0;l < PROC_MATRIX_SIZE-1 ;l++) {
+	  INFO2("LINE:   Send to %s:%d",
+		gras_socket_peer_name(socket_row[l]),
+		gras_socket_peer_port(socket_row[l]));
+	 gras_msg_send(socket_row[l], "dataB", &mydataB);
+       }
+       
 	
-	INFO0("send my data B to my (vertical) neighbors");
-      }
+       xbt_matrix_free(bB);
+       bB = xbt_matrix_new_sub(mydataB,
+			       submatrix_size,submatrix_size,
+			       0,0,NULL);	
     } else {
-      INFO2("step(%d) <> Myline(%d)",step,myline);
       TRY {
-	INFO1("mydataB=%p",mydataB);
 	xbt_matrix_free(bB);
-	INFO1("mydataB=%p",mydataB);
-	gras_msg_wait(600,gras_msgtype_by_name("dataB"),&from,&bB);
-	INFO1("mydataB=%p",mydataB);
+	gras_msg_wait(600,"dataB",&from,&bB);
       } CATCH(e) {
 	RETHROW0("Can't get a data message from line : %s");
       }
-      INFO1("Receive data B from my neighbor: %s",
-	    gras_socket_peer_name(from));
+      INFO4("LINE: step(%d) <> Myline(%d). Receive data from %s:%d",step,myline,
+	    gras_socket_peer_name(from), gras_socket_peer_port(from));
     }
 
     /* a row brodcast */
-    if (myrow==step) {
-      for (l=1;l < PROC_MATRIX_SIZE ;l++){
-	gras_msg_send(socket_line[l-1],gras_msgtype_by_name("dataA"), &mydataA);
-	xbt_matrix_free(bA);
-	bA = xbt_matrix_new_sub(mydataA,
-				submatrix_size,submatrix_size,
-				0,0,NULL);
-	
-	INFO0("send my data A to my (horizontal) neighbors");
-      }
+    if (myrow==step) { 
+       INFO2("ROW: step(%d)=myrow(%d). Broadcast my data",step,myrow);
+       for (l=1;l < PROC_MATRIX_SIZE ; l++) {
+	  INFO2("ROW:   Send to %s:%d",
+		gras_socket_peer_name(socket_line[l-1]),
+		gras_socket_peer_port(socket_line[l-1]));
+	  gras_msg_send(socket_line[l-1],"dataA", &mydataA);
+       }
+       xbt_matrix_free(bA);
+       bA = xbt_matrix_new_sub(mydataA,
+			       submatrix_size,submatrix_size,
+			       0,0,NULL);
     } else {
       TRY {
 	xbt_matrix_free(bA);
-	gras_msg_wait(1200,gras_msgtype_by_name("dataA"), &from,&bA);
+	gras_msg_wait(1200,"dataA", &from,&bA);
       } CATCH(e) {
 	RETHROW0("Can't get a data message from row : %s");
       }
-      INFO1("Receive data A from my neighbor : %s ",
+      INFO3("ROW: step(%d)<>myrow(%d). Receive data from %s",step,myrow,
 	    gras_socket_peer_name(from));
     }
     xbt_matrix_double_addmult(bA,bB,bC);
 
-    /* send a ack msg to master */
-	
-    gras_msg_send(master,gras_msgtype_by_name("step_ack"),&step);
-    
-    INFO1("Send ack to master for to end %d th step",step);
-  	
-  } while (step < PROC_MATRIX_SIZE);
-  /*  wait Message from master to send the result */
+  };
  
+  /* send Result to master */  
   result.C=bC;
   result.linepos=myline;
   result.rowpos=myrow;
- 
-  TRY {
-    gras_msg_wait(600,gras_msgtype_by_name("ask_result"),
-		  &master,&result_ack);
-  } CATCH(e) {
-    RETHROW0("Can't get a data message from line : %s");
-  }
-  /* send Result to master */
+
   TRY {
-    gras_msg_send(master, gras_msgtype_by_name("result"),&result);
+    gras_msg_send(master, "result",&result);
   } CATCH(e) {
-    // gras_socket_close(from);
-    RETHROW0("Failed to send PING to server: %s");
+    RETHROW0("Failed to send answer to server: %s");
   }
   INFO2(">>>>>>>> Result sent to %s:%d <<<<<<<<",
 	gras_socket_peer_name(master),gras_socket_peer_port(master));
   /*  Free the allocated resources, and shut GRAS down */
+
+  xbt_matrix_free(bA);
+  xbt_matrix_free(bB);
+  xbt_matrix_free(bC);
+
+  xbt_matrix_free(mydataA);
+  xbt_matrix_free(mydataB);
   gras_socket_close(master);
   gras_socket_close(from);
+  /* FIXME: some are said to be unknown 
+  for (l=0; l < PROC_MATRIX_SIZE-1; l++) {
+     if (socket_line[l])
+       gras_socket_close(socket_line[l]);
+     if (socket_row[l])
+       gras_socket_close(socket_row[l]); 
+  }*/
+
+  return 0;
+}
+
+int slave(int argc,char *argv[]) {
+  gras_socket_t mysock;
+  gras_socket_t master;
+
+  /* Init the GRAS's infrastructure */
+  gras_init(&argc, argv);
+  amok_pm_init();
+
+  /*  Register the known messages and my callback */
+  register_messages();
+  gras_cb_register("pmm_slave",pmm_worker_cb);
+
+  /* Create the connexions */
+  mysock = gras_socket_server_range(3000,9999,0,0);
+  gras_os_sleep(1); /* let the master get ready */
+  INFO1("Sensor starting (on port %d)",gras_os_myport());
+  master = gras_socket_client_from_string(argv[1]);
+				
+  /* Join and run the group */
+  amok_pm_group_join(master,"pmm");
+  amok_pm_mainloop(600);
+
+  /* housekeeping */
+  gras_socket_close(mysock);
+  //  gras_socket_close(master); Unknown
   gras_exit();
-  INFO0("Done.");
   return 0;
 } /* end_of_slave */