Skip to content

GitLab

Commit 66da80cc authored by Shane Snyder's avatar Shane Snyder
Browse files

Replay tool/darshan timing optimizations

parent ec4c6370
Hide whitespace changes
Inline Side-by-side
src/workload/codes-darshan-io-wrkld.c
View file @ 66da80cc
......@@ -12,10 +12,10 @@
#include "darshan-logutils.h"
#define DEF_INTER_IO_DELAY_PCT 0.2
#define DEF_INTER_CYC_DELAY_PCT 0.4
#define DEF_INTER_IO_DELAY_PCT 0.4
#define DEF_INTER_CYC_DELAY_PCT 0.2
#define DARSHAN_NEGLIGIBLE_DELAY .001
#define DARSHAN_NEGLIGIBLE_DELAY 0.001
#define RANK_HASH_TABLE_SIZE 397
......@@ -70,13 +70,12 @@ static double generate_psx_close_event(struct darshan_file *file, double meta_op
static double generate_barrier_event(struct darshan_file *file, int64_t root, double cur_time,
struct rank_io_context *io_context);
static double generate_psx_ind_io_events(struct darshan_file *file, int64_t io_ops_this_cycle,
double inter_io_delay, double meta_op_time,
double cur_time, struct rank_io_context *io_context);
double inter_io_delay, double cur_time,
struct rank_io_context *io_context);
static double generate_psx_coll_io_events(struct darshan_file *file, int64_t ind_io_ops_this_cycle,
int64_t coll_io_ops_this_cycle, int64_t nprocs,
int64_t aggregator_cnt, double inter_io_delay,
double meta_op_time, double cur_time,
struct rank_io_context *io_context);
double cur_time, struct rank_io_context *io_context);
static void determine_io_params(struct darshan_file *file, int write_flag, int64_t io_this_op,
int64_t proc_count, size_t *io_sz, off_t *io_off);
static void calc_io_delays(struct darshan_file *file, int64_t num_opens, int64_t num_io_ops,
......@@ -107,6 +106,7 @@ static int darshan_io_workload_load(const char *params, int rank)
struct darshan_job job;
struct darshan_file next_file;
struct rank_io_context *my_ctx;
struct darshan_io_op end_op;
int ret;
if (!d_params)
......@@ -162,8 +162,10 @@ static int darshan_io_workload_load(const char *params, int rank)
/* generate collective i/o events and store them in the rank context */
generate_psx_coll_file_events(&next_file, my_ctx, job.nprocs, d_params->aggregator_cnt);
}
else
else if (next_file.rank < rank)
continue;
else
break;
assert(next_file.counters[CP_POSIX_OPENS] == 0);
assert(next_file.counters[CP_POSIX_READS] == 0);
......@@ -174,6 +176,12 @@ static int darshan_io_workload_load(const char *params, int rank)
darshan_log_close(logfile_fd);
/* tack a workload end file on the end of the events list */
/* NOTE: this helps us generate a delay from the last close to program termination */
end_op.codes_op.op_type = CODES_WK_END;
end_op.start_time = end_op.end_time = job.end_time - job.start_time + 1;
darshan_insert_next_io_op(my_ctx->io_op_dat, &end_op);
/* finalize the rank's i/o context so i/o ops may be retrieved later (in order) */
darshan_finalize_io_op_dat(my_ctx->io_op_dat);
......@@ -359,16 +367,12 @@ static void darshan_remove_next_io_op(
{
/* no more events just end the workload */
io_op->codes_op.op_type = CODES_WK_END;
/* free data structures */
free(array->op_array);
free(array);
}
else
{
struct darshan_io_op *tmp = &(array->op_array[array->op_arr_ndx]);
if ((tmp->start_time - last_op_time) < DARSHAN_NEGLIGIBLE_DELAY)
if ((tmp->start_time - last_op_time) <= DARSHAN_NEGLIGIBLE_DELAY)
{
/* there is no delay, just return the next op in the array */
*io_op = *tmp;
......@@ -383,6 +387,15 @@ static void darshan_remove_next_io_op(
io_op->end_time = tmp->start_time;
}
}
/* if this is the end op, free data structures */
if (io_op->codes_op.op_type == CODES_WK_END)
{
free(array->op_array);
free(array);
}
return;
}
/* sort the dynamic array in order of i/o op start time */
......@@ -522,8 +535,7 @@ static void generate_psx_ind_file_events(
/* calculate average meta op time (for i/o and opens/closes) */
/* TODO: this needs to be updated when we add in stat, seek, etc. */
meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / ((2 * file->counters[CP_POSIX_OPENS]) +
file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES]);
meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / (2 * file->counters[CP_POSIX_OPENS]);
/* set the create flag if the file was written to */
if (file->counters[CP_BYTES_WRITTEN])
......@@ -549,7 +561,7 @@ static void generate_psx_ind_file_events(
/* perform the calculated number of i/o operations for this file open */
cur_time = generate_psx_ind_io_events(file, io_ops_this_cycle, inter_io_delay,
meta_op_time, cur_time, io_context);
cur_time, io_context);
/* account for potential delay from last io to close */
cur_time += close_delay;
......@@ -677,8 +689,7 @@ void generate_psx_coll_file_events(
assert(extra_opens <= open_cycles);
/* calculate average meta op time (for i/o and opens/closes) */
meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / ((2 * file->counters[CP_POSIX_OPENS]) +
file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES]);
meta_op_time = file->fcounters[CP_F_POSIX_META_TIME] / (2 * file->counters[CP_POSIX_OPENS]);
/* it is rare to overwrite existing files, so set the create flag */
if (file->counters[CP_BYTES_WRITTEN])
......@@ -704,7 +715,7 @@ void generate_psx_coll_file_events(
if (!file->counters[CP_COLL_OPENS] && !file->counters[CP_INDEP_OPENS])
{
cur_time = generate_psx_coll_io_events(file, 1, 0, nprocs, nprocs, 0.0,
meta_op_time, cur_time, io_context);
cur_time, io_context);
extra_io_ops--;
}
......@@ -728,8 +739,7 @@ void generate_psx_coll_file_events(
ind_io_ops_this_cycle = ceil(((double)total_ind_io_ops / total_ind_opens) * rank_cnt);
cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle, 0, nprocs,
nprocs, inter_io_delay, meta_op_time,
cur_time, io_context);
nprocs, inter_io_delay, cur_time, io_context);
total_ind_io_ops -= ind_io_ops_this_cycle;
cur_time += close_delay;
......@@ -766,8 +776,7 @@ void generate_psx_coll_file_events(
(file->counters[CP_COLL_OPENS] / nprocs));
cur_time = generate_psx_coll_io_events(file, ind_io_ops_this_cycle,
coll_io_ops_this_cycle, nprocs, in_agg_cnt,
inter_io_delay, meta_op_time,
cur_time, io_context);
inter_io_delay, cur_time, io_context);
total_ind_io_ops -= ind_io_ops_this_cycle;
total_coll_io_ops -= coll_io_ops_this_cycle;
......@@ -860,7 +869,7 @@ static double generate_barrier_event(
/* generate all i/o events for one independent file open and store them with the rank context */
static double generate_psx_ind_io_events(
struct darshan_file *file, int64_t io_ops_this_cycle, double inter_io_delay,
double meta_op_time, double cur_time, struct rank_io_context *io_context)
double cur_time, struct rank_io_context *io_context)
{
static int rw = -1; /* rw = 1 for write, 0 for read, -1 for uninitialized */
static int64_t io_ops_this_rw;
......@@ -924,7 +933,7 @@ static double generate_psx_ind_io_events(
io_op_time = (io_sz / rd_bw);
/* update time, accounting for metadata time */
cur_time += (io_op_time + meta_op_time);
cur_time += io_op_time;
next_io_op.end_time = cur_time;
file->counters[CP_POSIX_READS]--;
}
......@@ -944,7 +953,7 @@ static double generate_psx_ind_io_events(
io_op_time = (io_sz / wr_bw);
/* update time, accounting for metadata time */
cur_time += (io_op_time + meta_op_time);
cur_time += io_op_time;
next_io_op.end_time = cur_time;
file->counters[CP_POSIX_WRITES]--;
}
......@@ -990,8 +999,8 @@ static double generate_psx_ind_io_events(
static double generate_psx_coll_io_events(
struct darshan_file *file, int64_t ind_io_ops_this_cycle, int64_t coll_io_ops_this_cycle,
int64_t nprocs, int64_t aggregator_cnt, double inter_io_delay, double meta_op_time,
double cur_time, struct rank_io_context *io_context)
int64_t nprocs, int64_t aggregator_cnt, double inter_io_delay, double cur_time,
struct rank_io_context *io_context)
{
static int rw = -1; /* rw = 1 for write, 0 for read, -1 for uninitialized */
static int64_t io_ops_this_rw;
......@@ -1123,7 +1132,7 @@ static double generate_psx_coll_io_events(
else
io_op_time = (io_sz / rd_bw);
next_io_op.end_time = cur_time + io_op_time + meta_op_time;
next_io_op.end_time = cur_time + io_op_time;
file->counters[CP_POSIX_READS]--;
}
else
......@@ -1141,7 +1150,7 @@ static double generate_psx_coll_io_events(
else
io_op_time = (io_sz / wr_bw);
next_io_op.end_time = cur_time + io_op_time + meta_op_time;
next_io_op.end_time = cur_time + io_op_time;
file->counters[CP_POSIX_WRITES]--;
}
psx_rw_ops_remaining--;
......@@ -1585,8 +1594,7 @@ static void file_sanity_check(
/* reduce total meta time by percentage of ops not currently implemented */
ops_not_implemented = file->counters[CP_POSIX_SEEKS] + file->counters[CP_POSIX_STATS] +
file->counters[CP_POSIX_FSYNCS];
ops_total = ops_not_implemented + (2 * file->counters[CP_POSIX_OPENS]) +
file->counters[CP_POSIX_READS] + file->counters[CP_POSIX_WRITES];
ops_total = ops_not_implemented + (2 * file->counters[CP_POSIX_OPENS]);
file->fcounters[CP_F_POSIX_META_TIME] *= (1 - ((double)ops_not_implemented / ops_total));
return;
......
tests/workload/codes-workload-mpi-replay.c
View file @ 66da80cc
......@@ -26,6 +26,8 @@
#define WORKLOAD_PRINT 1
#define WORKLOAD_DELAY_PCT 0.95
/* hash table entry for looking up file descriptor of a workload file id */
struct file_info
{
......@@ -193,6 +195,8 @@ int main(int argc, char *argv[])
int workload_id;
struct codes_workload_op next_op;
long long int replay_op_number = 1;
double load_start, load_end;
int first_op = 1;
int ret = 0;
/* parse command line args */
......@@ -203,6 +207,9 @@ int main(int argc, char *argv[])
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
/* start workload load timer */
load_start = MPI_Wtime();
/* change the working directory to be the test directory */
ret = chdir(replay_test_path);
if (ret < 0)
......@@ -239,14 +246,32 @@ int main(int argc, char *argv[])
goto error_exit;
}
/* loading is finished */
load_end = MPI_Wtime();
/* replay loop */
while (1)
{
/* get the next replay operation from the workload generator */
codes_workload_get_next(workload_id, myrank, &next_op);
/* subtract the load time from any initial delay to be fair */
if (first_op && (next_op.op_type == CODES_WK_DELAY))
{
if (next_op.u.delay.seconds > (load_end - load_start))
next_op.u.delay.seconds -= (load_end - load_start);
else
continue;
first_op = 0;
}
if (next_op.op_type != CODES_WK_END)
{
if (next_op.op_type == CODES_WK_DELAY)
next_op.u.delay.seconds *= WORKLOAD_DELAY_PCT;
/* replay the next workload operation */
ret = replay_workload_op(next_op, myrank, replay_op_number++);
if (ret < 0)
......@@ -486,3 +511,13 @@ int hash_file_compare(void *key, struct qlist_head *link)
return 0;
}
/*
* Local variables:
* c-indent-level: 4
* c-basic-offset: 4
* End:
*
* vim: ft=c ts=8 sts=4 sw=4 expandtab
*/
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