Commit 0b1d6875 authored by Francois Tessier's avatar Francois Tessier

Change env variables. Add miniHACC-SoA example. Update Readme.md

parent 3638c509
Environment variables
----------------------
* TAPIOCA_STRATEGY = SHORTEST_PATH / LONGEST_PATH / TOPOLOGY_AWARE / CONTENTION_AWARE
* TAPIOCA_NBAGGR = Number of aggregators per file
* TAPIOCA_BUFFERSIZE = Buffer size in bytes. Use a multiple of the file system block size to avoid lock contention. Two allocations of this buffer size will be made to perform double-buffering
* TAPIOCA_COMMSPLIT = true / false. If true, MPI_Comm_split will be used to create one sub-communicator per aggregator. If false, the sub-communicator will be created from MPI_Groups. In the case of a single shared file as output on a large-scale run, set this variable to false can divide by two the time needed to elect the aggregators.
* TAPIOCA_DEVNULL = true / false. If true, instead of effectively writing the file, the write operation is made in /dev/null. Useful for aggregation time measurements.
Default values
**************
TAPIOCA_STRATEGY = SHORTEST_PATH
TAPIOCA_NBAGGR = 8
TAPIOCA_BUFFERSIZE = 16777216
TAPIOCA_COMMSPLIT = true
TAPIOCA_DEVNULL = false
\ No newline at end of file
......@@ -11,7 +11,7 @@ miniHACC-AoS.o:miniHACC-AoS.cpp
$(MPICC) $(MPI_CFLAGS) -c miniHACC-AoS.cpp
miniHACC-SoA:miniHACC-SoA.o
$(MPICC) $(MPI_CFLAGS) $(MPI_LDFLAGS) miniHACC-SoA.o -o $@
$(MPICC) $(MPI_CFLAGS) miniHACC-SoA.o -o $@ $(MPI_LDFLAGS)
miniHACC-SoA.o:miniHACC-SoA.cpp
$(MPICC) $(MPI_CFLAGS) -c miniHACC-SoA.cpp
......
......@@ -3,7 +3,6 @@
#include <assert.h>
#include <stdint.h>
#include <mpi.h>
#include <mpix.h>
#include "tapioca.hpp"
......@@ -24,12 +23,15 @@ int main (int argc, char * argv[])
MPI_File file_handle;
MPI_Status status;
char output[100];
Tapioca tp;
int64_t chunkCount[9], chunkOffset[9];
int chunkSize[9];
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &world_numtasks);
MPI_Comm_rank(MPI_COMM_WORLD, &world_myrank);
mycolor = MPIX_IO_link_id ();
mycolor = tp.topology.BridgeNodeId ();
mykey = world_myrank;
MPI_Comm_split (MPI_COMM_WORLD, mycolor, mykey, &sub_comm);
......@@ -94,10 +96,6 @@ int main (int argc, char * argv[])
/*****************/
/* INIT TAPIOCA */
/*****************/
Tapioca tp;
int64_t chunkCount[9], chunkOffset[9];
int chunkSize[9];
for ( i = 0; i < 9; i++ ) {
chunkCount[i] = num_particles;
}
......
#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include <stdint.h>
#include <mpi.h>
#include "tapioca.hpp"
#define RED "\x1b[31m"
#define GREEN "\x1b[32m"
#define BLUE "\x1b[34m"
#define RESET "\x1b[0m"
int main (int argc, char * argv[])
{
int world_numtasks, world_myrank, mycolor, mykey, sub_numtasks, sub_myrank, i;
int64_t num_particles = 25000;
int64_t sub_particles, tot_particles, particle_size, file_size, tot_size;
int64_t scan_size = 0, offset, hdr = 0;
double start_time, end_time, tot_time, max_time;
double io_bw;
MPI_Comm sub_comm;
MPI_File file_handle;
MPI_Status status;
char output[100];
Tapioca tp;
int64_t chunkCount[9], chunkOffset[9];
int chunkSize[9];
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &world_numtasks);
MPI_Comm_rank(MPI_COMM_WORLD, &world_myrank);
mycolor = tp.topology.BridgeNodeId ();
mykey = world_myrank;
MPI_Comm_split (MPI_COMM_WORLD, mycolor, mykey, &sub_comm);
MPI_Comm_size(sub_comm, &sub_numtasks);
MPI_Comm_rank(sub_comm, &sub_myrank);
snprintf (output, 100, "/projects/visualization/ftessier/debug/HACC-SOA-%08d.dat", mycolor);
/*****************/
/* WRITE */
/*****************/
float *xx, *yy, *zz, *vx, *vy, *vz, *phi;
int64_t* pid;
uint16_t* mask;
xx = new float[num_particles];
yy = new float[num_particles];
zz = new float[num_particles];
vx = new float[num_particles];
vy = new float[num_particles];
vz = new float[num_particles];
phi = new float[num_particles];
pid = new int64_t[num_particles];
mask = new uint16_t[num_particles];
for (uint64_t i = 0; i< num_particles; i++)
{
xx[i] = (float)i;
yy[i] = (float)i;
zz[i] = (float)i;
vx[i] = (float)i;
vy[i] = (float)i;
vz[i] = (float)i;
phi[i] = (float)i;
pid[i] = (int64_t)i;
mask[i] = (uint16_t)world_myrank;
}
MPI_Allreduce(&num_particles, &sub_particles, 1, MPI_LONG_LONG, MPI_SUM, sub_comm);
MPI_Allreduce(&num_particles, &tot_particles, 1, MPI_LONG_LONG, MPI_SUM, MPI_COMM_WORLD);
particle_size = (7 * sizeof(float)) + sizeof(int64_t) + sizeof(uint16_t);
file_size = particle_size * sub_particles;
tot_size = particle_size * tot_particles;
if (sub_myrank == 0) {
MPI_File_open(MPI_COMM_SELF, output,
MPI_MODE_WRONLY | MPI_MODE_CREATE, MPI_INFO_NULL, &file_handle);
MPI_File_set_size(file_handle, file_size);
MPI_File_close (&file_handle);
}
MPI_Exscan (&num_particles, &scan_size, 1, MPI_LONG_LONG, MPI_SUM, sub_comm);
if (0 == sub_myrank) {
fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Write output file (SoA data layout)\n", mycolor);
fprintf (stdout, GREEN "[INFO]" RESET " [%08d] --> %lld particles per rank\n", mycolor, num_particles);
fprintf (stdout, GREEN "[INFO]" RESET " [%08d] --> File size: %.2f MB (%lld particles)\n",
mycolor, (double)file_size/(1024*1024), sub_particles);
}
/*****************/
/* INIT TAPIOCA */
/*****************/
for ( i = 0; i < 9; i++ ) {
chunkCount[i] = num_particles;
}
chunkSize[0] = sizeof(float);
chunkSize[1] = sizeof(float);
chunkSize[2] = sizeof(float);
chunkSize[3] = sizeof(float);
chunkSize[4] = sizeof(float);
chunkSize[5] = sizeof(float);
chunkSize[6] = sizeof(float);
chunkSize[7] = sizeof(int64_t);
chunkSize[8] = sizeof(uint16_t);
chunkOffset[0] = hdr + scan_size * chunkSize[0];
for ( i = 1; i < 9; i++ ) {
chunkOffset[i] = chunkOffset[i - 1];
chunkOffset[i] += (sub_particles - scan_size) * chunkSize[i - 1];
chunkOffset[i] += scan_size * chunkSize[i];
}
tp.Initialize (chunkCount, chunkSize, chunkOffset, 9, hdr, STRUCTURE_OF_ARRAYS, sub_comm);
/*****************/
start_time = MPI_Wtime();
MPI_File_open(sub_comm, output,
MPI_MODE_WRONLY, MPI_INFO_NULL, &file_handle);
offset = scan_size * sizeof(float);
tp.Commit (file_handle, offset, xx, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, yy, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, zz, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, vx, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, vy, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, vz, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
tp.Commit (file_handle, offset, phi, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(int64_t);
tp.Commit (file_handle, offset, pid, num_particles, MPI_LONG_LONG, &status);
offset += (sub_particles - scan_size) * sizeof(int64_t) + scan_size * sizeof(uint16_t);
tp.Commit (file_handle, offset, mask, num_particles, MPI_UNSIGNED_SHORT, &status);
MPI_File_close (&file_handle);
end_time = MPI_Wtime();
tot_time = end_time - start_time;
MPI_Reduce (&tot_time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if (0 == world_myrank) {
io_bw = (double)tot_size / max_time / (1024 * 1024);
fprintf (stdout, BLUE "[TIMING]" RESET " Write I/O bandwidth: %.2f MBps (%.2f MB in %.2f ms)\n",
io_bw, (double)tot_size/(1024*1024), max_time * 1000);
}
MPI_Barrier (MPI_COMM_WORLD);
/*****************/
/* READ */
/*****************/
float *xx_r, *yy_r, *zz_r, *vx_r, *vy_r, *vz_r, *phi_r;
int64_t* pid_r;
uint16_t* mask_r;
xx_r = new float[num_particles];
yy_r = new float[num_particles];
zz_r = new float[num_particles];
vx_r = new float[num_particles];
vy_r = new float[num_particles];
vz_r = new float[num_particles];
phi_r = new float[num_particles];
pid_r = new int64_t[num_particles];
mask_r = new uint16_t[num_particles];
start_time = MPI_Wtime();
MPI_File_open(sub_comm, output,
MPI_MODE_RDONLY, MPI_INFO_NULL, &file_handle);
if (0 == sub_myrank)
fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Read output file\n", mycolor);
offset = scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, xx_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, yy_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, zz_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, vx_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, vy_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, vz_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(float);
MPI_File_read_at_all (file_handle, offset, phi_r, num_particles, MPI_FLOAT, &status);
offset += (sub_particles - scan_size) * sizeof(float) + scan_size * sizeof(int64_t);
MPI_File_read_at_all (file_handle, offset, pid_r, num_particles, MPI_LONG_LONG, &status);
offset += (sub_particles - scan_size) * sizeof(int64_t) + scan_size * sizeof(uint16_t);
MPI_File_read_at_all (file_handle, offset, mask_r, num_particles, MPI_UNSIGNED_SHORT, &status);
MPI_File_close (&file_handle);
end_time = MPI_Wtime();
tot_time = end_time - start_time;
MPI_Reduce (&tot_time, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if (0 == world_myrank) {
io_bw = (double)tot_size / max_time / (1024 * 1024);
fprintf (stdout, BLUE "[TIMING]" RESET " Read I/O bandwidth: %.2f MBps (%.2f MB in %.2f ms)\n",
io_bw, (double)tot_size/(1024*1024), max_time * 1000);
}
/*****************/
/* VERIFICATION */
/*****************/
for (uint64_t i = 0; i< num_particles; i++) {
if ((xx[i] != xx_r[i]) || (yy[i] != yy_r[i]) || (zz[i] != zz_r[i])
|| (vx[i] != vx_r[i]) || (vy[i] != vy_r[i]) || (vz[i] != vz_r[i])
|| (phi[i] != phi_r[i])|| (pid[i] != pid_r[i]) || (mask[i] != mask_r[i]))
{
fprintf (stdout, RED "[ERROR]" RESET " Wrong value for particle %d\n", i);
MPI_Abort (MPI_COMM_WORLD, -1);
}
}
if (0 == sub_myrank)
fprintf (stdout, GREEN "[INFO]" RESET " [%08d] Content verified and consistent\n", mycolor);
/*****************/
/* FREE */
/*****************/
delete [] xx;
delete [] xx_r;
delete [] yy;
delete [] yy_r;
delete [] zz;
delete [] zz_r;
delete [] vx;
delete [] vx_r;
delete [] vy;
delete [] vy_r;
delete [] vz;
delete [] vz_r;
delete [] phi;
delete [] phi_r;
delete [] pid;
delete [] pid_r;
delete [] mask;
delete [] mask_r;
MPI_Finalize ();
}
......@@ -328,9 +328,9 @@ void Tapioca::SetDefaultValues ()
this->rankDataSize_ = 0;
this->strategy_ = SHORTEST_PATH;
this->nAggr_ = 8;
this->bufferSize_ = 33554432;
this->bufferSize_ = 16777216;
this->amAnAggr_ = false;
this->commSplit_ = false;
this->commSplit_ = true;
this->currentRound_ = 0;
this->roundCounter_ = 0;
this->currentDataSize_ = 0;
......@@ -341,11 +341,11 @@ void Tapioca::SetDefaultValues ()
void Tapioca::ParseEnvVariables ()
{
char *envStrategy = getenv("AGGR_STRATEGY");
char *envNAggr = getenv("AGGR_NBAGGR");
char *envBufferSize = getenv("AGGR_BUFFERSIZE");
char *envSplit = getenv("AGGR_COMMSPLIT");
char *envDevNull = getenv("AGGR_DEVNULL");
char *envStrategy = getenv("TAPIOCA_STRATEGY");
char *envNAggr = getenv("TAPIOCA_NBAGGR");
char *envBufferSize = getenv("TAPIOCA_BUFFERSIZE");
char *envSplit = getenv("TAPIOCA_COMMSPLIT");
char *envDevNull = getenv("TAPIOCA_DEVNULL");
if (envStrategy != NULL) {
strcmp(envStrategy, "SHORTEST_PATH") ? 0 : this->strategy_ = SHORTEST_PATH;
......
......@@ -59,6 +59,7 @@ class Tapioca
void MPIIOInfo (MPI_File fileHandle);
Topology topology;
private:
/***********************/
/* INITIALIZATION */
......@@ -163,8 +164,6 @@ class Tapioca
/* TIMING */
double startAggrTime, endAggrTime, totAggrTime;
double startIOTime, endIOTime, totIOTime;
Topology topology;
};
#endif /* AGGREGATION_H */
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment