diff --git a/exercise3/wave_1d_parallel.c b/exercise3/wave_1d_parallel.c
index d81c084..275a7bf 100644
--- a/exercise3/wave_1d_parallel.c
+++ b/exercise3/wave_1d_parallel.c
@@ -53,6 +53,8 @@ real_t
 // Save the present time step in a numbered file under 'data/'.
 void domain_save(int_t step) {
   // BEGIN: T8
+  if (comm_rank != ROOT)
+    return;
   char filename[256];
   sprintf(filename, "data/%.5ld.dat", step);
   FILE *out = fopen(filename, "wb");
@@ -67,12 +69,13 @@ void domain_save(int_t step) {
 // and set the time step.
 void domain_initialize(void) {
   // BEGIN: T3
-  buffers[0] = malloc((N + 2) * sizeof(real_t));
-  buffers[1] = malloc((N + 2) * sizeof(real_t));
-  buffers[2] = malloc((N + 2) * sizeof(real_t));
+  const int_t bufsize = (comm_rank == ROOT ? N : n + 2);
+  buffers[0] = malloc(bufsize * sizeof(real_t));
+  buffers[1] = malloc(bufsize * sizeof(real_t));
+  buffers[2] = malloc(bufsize * sizeof(real_t));
 
-  for (int_t i = 0; i < N; i++) {
-    U_prv(i) = U(i) = cos(2 * M_PI * i / (real_t)N);
+  for (int_t i = 0; i < n; i++) {
+    U_prv(i) = U(i) = cos(2 * M_PI * (i + offset) / (real_t)N);
   }
   // END: T3
 
@@ -99,7 +102,7 @@ void move_buffer_window(void) {
 // Derive step t+1 from steps t and t-1.
 void time_step(void) {
   // BEGIN: T4
-  for (int_t i = 0; i < N; i++) {
+  for (int_t i = 0; i < n; i++) {
     U_nxt(i) = -U_prv(i) + 2.0 * U(i) + (dt * dt * c * c) / (dx * dx) * (U(i - 1) + U(i + 1) - 2.0 * U(i));
   }
   // END: T4
@@ -109,8 +112,12 @@ void time_step(void) {
 // Neumann (reflective) boundary condition.
 void boundary_condition(void) {
   // BEGIN: T6
-  U(-1) = U(1);
-  U(N) = U(N - 2);
+  if (comm_rank == ROOT) {
+    U(-1) = U(1);
+  }
+  if (comm_rank == comm_size - 1) {
+    U(n) = U(n - 2);
+  }
   // END: T6
 }
 
@@ -118,7 +125,16 @@ void boundary_condition(void) {
 // Communicate the border between processes.
 void border_exchange(void) {
   // BEGIN: T5
-  ;
+  int left = (comm_size + comm_rank - 1) % comm_size;
+  int right = (comm_rank + 1) % comm_size;
+
+  MPI_Sendrecv(&U(n - 1), 1, MPI_DOUBLE, right, 0,
+               &U(-1), 1, MPI_DOUBLE, left, 0,
+               MPI_COMM_WORLD, MPI_STATUS_IGNORE);
+
+  MPI_Sendrecv(&U(0), 1, MPI_DOUBLE, left, 1,
+               &U(n), 1, MPI_DOUBLE, right, 1,
+               MPI_COMM_WORLD, MPI_STATUS_IGNORE);
   // END: T5
 }
 
@@ -127,7 +143,16 @@ void border_exchange(void) {
 // to root and assemble it in the root buffer
 void send_data_to_root() {
   // BEGIN: T7
-  ;
+  if (comm_rank != ROOT) {
+    MPI_Send(&U(0), n, MPI_DOUBLE, ROOT, 0, MPI_COMM_WORLD);
+  } else {
+    int offset = n;
+    for (int i = 1; i < comm_size; i++) {
+      int proc_n = N / comm_size + (i < N % comm_size);
+      MPI_Recv(&U(offset), proc_n, MPI_DOUBLE, i, 0, MPI_COMM_WORLD, MPI_STATUS_IGNORE);
+      offset += proc_n;
+    }
+  }
   // END: T7
 }
 
@@ -170,7 +195,17 @@ int main(int argc, char **argv) {
   // TASK: T2
   // Time your code
   // BEGIN: T2
+  if (comm_rank == ROOT)
+    gettimeofday(&t_start, NULL);
   simulate();
+  MPI_Barrier(MPI_COMM_WORLD);
+
+  if (comm_rank == ROOT) {
+    gettimeofday(&t_end, NULL);
+
+    printf("Total elapsed time: %lf seconds\n",
+           WALLTIME(t_end) - WALLTIME(t_start));
+  }
   // END: T2
 
   domain_finalize();