add collective overhead v2: constant plus per byte

src/network-workloads/model-net-mpi-replay.c

@@ -74,7 +74,8 @@ int num_traces_of_job[5];
 tw_stime soft_delay_mpi = 1250;
 tw_stime nic_delay = 250;
 tw_stime copy_per_byte_eager = 0.55;
-tw_stime col_overhead_per_byte = 1.5;
+tw_stime col_overhead_per_byte = 0.8;
+tw_stime col_soft_overhead = 6000;
 char file_name_of_job[5][8192];
 
 struct codes_jobmap_ctx *jobmap_ctx;
@@ -1507,7 +1508,7 @@ static void codes_exec_mpi_send(nw_state* s,
 
     tw_stime collective_overhead = 0;
     if(s->is_collective)
-        collective_overhead = col_overhead_per_byte * mpi_op->u.send.num_bytes;            
+        collective_overhead = col_soft_overhead + col_overhead_per_byte * mpi_op->u.send.num_bytes;            
    
     if(mpi_op->u.send.num_bytes < EAGER_THRESHOLD)
     {
@@ -2537,6 +2538,8 @@ const tw_optdef app_opt [] =
     TWOPT_UINT("preserve_wait_ordering", preserve_wait_ordering, "only enable when getting unmatched send/recv errors in optimistic mode (turning on slows down simulation)"),
     TWOPT_UINT("debug_cols", debug_cols, "completion time of collective operations (currently MPI_AllReduce)"),
     TWOPT_UINT("enable_col_overhead", enable_col_overhead, "adding overhead to collective operations (currently MPI_AllReduce)"),
+    TWOPT_STIME("col_overhead_per_byte",col_overhead_per_byte, "collective call overhead per byte"),
+    TWOPT_STIME("col_soft_overhead",col_soft_overhead, "collective call constant overhead"),
     TWOPT_UINT("enable_mpi_debug", enable_debug, "enable debugging of MPI sim layer (works with sync=1 only)"),
     TWOPT_UINT("sampling_interval", sampling_interval, "sampling interval for MPI operations"),
 	TWOPT_UINT("enable_sampling", enable_sampling, "enable sampling (only works in sequential mode)"),