static void batch_queue_wq_option_update (struct batch_queue *q, const char *what, const char *value) { if(strcmp(what, "password") == 0) { if(value) work_queue_specify_password(q->data, value); } else if(strcmp(what, "master-mode") == 0) { if(strcmp(value, "catalog") == 0) work_queue_specify_master_mode(q->data, WORK_QUEUE_MASTER_MODE_CATALOG); else if(strcmp(value, "standalone") == 0) work_queue_specify_master_mode(q->data, WORK_QUEUE_MASTER_MODE_STANDALONE); } else if(strcmp(what, "name") == 0) { if(value) work_queue_specify_name(q->data, value); } else if(strcmp(what, "priority") == 0) { if(value) work_queue_specify_priority(q->data, atoi(value)); else work_queue_specify_priority(q->data, 0); } else if(strcmp(what, "fast-abort") == 0) { if(value) work_queue_activate_fast_abort(q->data, atof(value)); } else if(strcmp(what, "estimate-capacity") == 0) { work_queue_specify_estimate_capacity_on(q->data, string_istrue(value)); } else if(strcmp(what, "keepalive-interval") == 0) { if(value) work_queue_specify_keepalive_interval(q->data, atoi(value)); else work_queue_specify_keepalive_interval(q->data, WORK_QUEUE_DEFAULT_KEEPALIVE_INTERVAL); } else if(strcmp(what, "keepalive-timeout") == 0) { if(value) work_queue_specify_keepalive_timeout(q->data, atoi(value)); else work_queue_specify_keepalive_timeout(q->data, WORK_QUEUE_DEFAULT_KEEPALIVE_TIMEOUT); } else if(strcmp(what, "master-preferred-connection") == 0) { if(value) work_queue_master_preferred_connection(q->data, value); else work_queue_master_preferred_connection(q->data, "by_ip"); } else if(strcmp(what, "category-limits") == 0) { struct rmsummary *s = rmsummary_parse_string(value); if(s) { work_queue_specify_category_max_resources(q->data, s->category, s); rmsummary_delete(s); } else { debug(D_NOTICE, "Could no parse '%s' as a summary of resorces encoded in JSON\n", value); } } }
static void batch_queue_wq_option_update (struct batch_queue *q, const char *what, const char *value) { if(strcmp(what, "password") == 0) { if(value) work_queue_specify_password(q->data, value); } else if(strcmp(what, "master-mode") == 0) { if(strcmp(value, "catalog") == 0) work_queue_specify_master_mode(q->data, WORK_QUEUE_MASTER_MODE_CATALOG); else if(strcmp(value, "standalone") == 0) work_queue_specify_master_mode(q->data, WORK_QUEUE_MASTER_MODE_STANDALONE); } else if(strcmp(what, "name") == 0) { if(value) work_queue_specify_name(q->data, value); } else if(strcmp(what, "priority") == 0) { if(value) work_queue_specify_priority(q->data, atoi(value)); else work_queue_specify_priority(q->data, 0); } else if(strcmp(what, "estimate-capacity") == 0) { work_queue_specify_estimate_capacity_on(q->data, string_istrue(value)); } else if(strcmp(what, "keepalive-interval") == 0) { if(value) work_queue_specify_keepalive_interval(q->data, atoi(value)); else work_queue_specify_keepalive_interval(q->data, WORK_QUEUE_DEFAULT_KEEPALIVE_INTERVAL); } else if(strcmp(what, "keepalive-timeout") == 0) { if(value) work_queue_specify_keepalive_timeout(q->data, atoi(value)); else work_queue_specify_keepalive_timeout(q->data, WORK_QUEUE_DEFAULT_KEEPALIVE_TIMEOUT); } else if(strcmp(what, "wait-queue-size") == 0) { if(value) work_queue_activate_worker_waiting(q->data, atoi(value)); else work_queue_activate_worker_waiting(q->data, 0); } else if(strcmp(what, "master-preferred-connection") == 0) { if(value) work_queue_master_preferred_connection(q->data, value); else work_queue_master_preferred_connection(q->data, "by_ip"); } }
int main(int argc, char *argv[]) { struct work_queue *q; int port = 0; //pick an arbitrary port int c; char *sort_arguments = NULL; const char *proj_name = NULL; char *outfile= NULL; int auto_partition = 0; int sample_env = 0; int print_runtime_estimates = 0; int estimate_partition= 0; struct timeval current; long long unsigned int execn_start_time, execn_time, workload_runtime; int keepalive_interval = 300; int keepalive_timeout = 30; unsigned long long records = 0; int partitions = PARTITION_DEFAULT; int sample_size = SAMPLE_SIZE_DEFAULT; gettimeofday(¤t, 0); execn_start_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; debug_flags_set("all"); if(argc < 3) { show_help(argv[0]); return 0; } while((c = getopt(argc, argv, "N:k:o:ASs:p:MR:L:I:T:B:h")) != (char) -1) { switch (c) { case 'N': proj_name = strdup(optarg); break; case 'k': partitions = atoi(optarg); break; case 'o': outfile = strdup(optarg); break; case 'A': auto_partition = 1; break; case 's': sample_size = atoi(optarg); break; case 'S': sample_env = 1; break; case 'p': sort_arguments = strdup(optarg); break; case 'M': print_runtime_estimates = 1; break; case 'R': estimate_partition = atoi(optarg); break; case 'L': records = atoll(optarg); break; case 'I': keepalive_interval = atoi(optarg); break; case 'T': keepalive_timeout = atoi(optarg); break; case 'B': bandwidth_bytes_per_sec = atoi(optarg) * 1000000; break; case 'h': show_help(argv[0]); return 0; default: show_help(argv[0]); return -1; } } char sort_executable[256], infile[256]; off_t last_partition_offset_end = 0; int optimal_partitions, optimal_resources, current_optimal_partitions; double current_optimal_time = DBL_MAX; double optimal_times[5]; int sample_partition_offset_end = 0; int i; sprintf(sort_executable, "%s", argv[optind]); sprintf(infile, "%s", argv[optind+1]); if(!outfile){ char *infile_dup = strdup(infile); outfile = (char *) malloc((strlen(infile)+8)*sizeof(char)); sprintf(outfile, "%s.sorted", basename(infile_dup)); free(infile_dup); } if(records == 0) { records = get_total_lines(infile); fprintf(stdout, "Input file %s has %llu records to sort\n", infile, records); if(records == 0) { fprintf(stderr, "Error in reading records. Quitting...\n"); return 0; } } if(estimate_partition) { double *estimated_runtimes = (double *)malloc(sizeof(double) * 5); for (i = 1; i <= 2*estimate_partition; i++) { estimated_runtimes = sort_estimate_runtime(infile, sort_executable, records, i, estimate_partition); if(estimated_runtimes[0] < current_optimal_time) { current_optimal_time = estimated_runtimes[0]; optimal_times[0] = estimated_runtimes[0]; optimal_times[1] = estimated_runtimes[1]; optimal_times[2] = estimated_runtimes[2]; optimal_times[3] = estimated_runtimes[3]; optimal_times[4] = estimated_runtimes[4]; optimal_resources = i; } } fprintf(stdout, "For partition %d: %d %f %f %f %f %f\n", estimate_partition, optimal_resources, optimal_times[0], optimal_times[1], optimal_times[2], optimal_times[3], optimal_times[4]); free(estimated_runtimes); return 1; } if(print_runtime_estimates) { fprintf(stdout, "Resources \t Partitions \t Runtime \t Part time \t Merge time \t Task time \t Transfer time\n"); for (i = 1; i <= 100; i++) { optimal_partitions = get_optimal_runtimes(infile, sort_executable, i, records, optimal_times); fprintf(stdout, "%d \t \t %d \t %f \t %f \t %f \t %f \t %f\n", i, optimal_partitions, optimal_times[0], optimal_times[1], optimal_times[2], optimal_times[3], optimal_times[4]); } return 1; } q = work_queue_create(port); if(!q) { fprintf(stderr, "couldn't listen on port %d: %s\n", port, strerror(errno)); return 1; } fprintf(stdout, "listening on port %d...\n", work_queue_port(q)); if(proj_name){ work_queue_specify_master_mode(q, WORK_QUEUE_MASTER_MODE_CATALOG); work_queue_specify_name(q, proj_name); } work_queue_specify_keepalive_interval(q, keepalive_interval); work_queue_specify_keepalive_timeout(q, keepalive_timeout); free((void *)proj_name); fprintf(stdout, "%s will be run to sort contents of %s\n", sort_executable, infile); long long unsigned int sample_start_time, sample_end_time, sample_time; if(sample_env) { gettimeofday(¤t, 0); sample_start_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; int sample_record_size = (5*records)/100; //sample size is 5% of the total records char *sample_partition_file_prefix = (char *) malloc((strlen(outfile)+8) * sizeof(char)); sprintf(sample_partition_file_prefix, "%s.sample", outfile); char *sample_outfile = (char *) malloc((strlen(outfile)+3) * sizeof(char)); sprintf(sample_outfile, "%s.0", outfile); sample_partition_offset_end = sample_run(q, sort_executable, sort_arguments, infile, 0, sample_partition_file_prefix, sample_outfile, sample_size, sample_record_size); records -= sample_record_size; free(sample_partition_file_prefix); free(sample_outfile); gettimeofday(¤t, 0); sample_end_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; sample_time = sample_end_time - sample_start_time; fprintf(stdout, "Sampling time is %llu\n", sample_time); } if(auto_partition) { fprintf(stdout, "Determining optimal partition size for %s\n", infile); for (i = 1; i <= 100; i++) { current_optimal_partitions = get_optimal_runtimes(infile, sort_executable, i, records, optimal_times); if (optimal_times[0] < current_optimal_time) { current_optimal_time = optimal_times[0]; optimal_partitions = current_optimal_partitions; optimal_resources = i; } } fprintf(stdout, "Optimal partition size is %d that runs the workload in %f\n", optimal_partitions, current_optimal_time); fprintf(stdout, "--> Please allocate %d resources for running this workload in a cost-efficient manner.\n", optimal_resources); partitions = optimal_partitions; } long long unsigned int part_start_time, part_end_time, part_time; gettimeofday(¤t, 0); part_start_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; last_partition_offset_end = partition_tasks(q, sort_executable, sort_arguments, infile, 0+sample_partition_offset_end, outfile, partitions, records); if(last_partition_offset_end <= 0) { fprintf(stderr, "Partitioning failed. Quitting...\n"); return 0; } gettimeofday(¤t, 0); part_end_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; part_time = part_end_time - part_start_time; fprintf(stdout, "Partition time is %llu\n", part_time); free(sort_arguments); fprintf(stdout, "Waiting for tasks to complete...\n"); long long unsigned int parallel_start_time, parallel_end_time, parallel_time; gettimeofday(¤t, 0); parallel_start_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; char *record_task_times_file = (char *)malloc((strlen(outfile)+11) * sizeof(char)); sprintf(record_task_times_file, "%s.tasktimes", outfile); wait_partition_tasks(q, 5, record_task_times_file); free(record_task_times_file); gettimeofday(¤t, 0); parallel_end_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; parallel_time = parallel_end_time - parallel_start_time; fprintf(stdout, "Parallel execution time is %llu\n", parallel_time); long long unsigned int merge_start_time, merge_end_time, merge_time; gettimeofday(¤t, 0); merge_start_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; merge_sorted_outputs(outfile, outfile, created_partitions); gettimeofday(¤t, 0); merge_end_time = ((long long unsigned int) current.tv_sec) * 1000000 + current.tv_usec; merge_time = merge_end_time - merge_start_time; fprintf(stdout, "Merge time is %llu\n", merge_time); fprintf(stdout, "Sorting complete. Output is at: %s!\n", outfile); execn_time = merge_end_time - execn_start_time; workload_runtime = merge_end_time - part_start_time; fprintf(stdout, "Workload execn time is %llu\n", workload_runtime); fprintf(stdout, "Total execn time is %llu\n", execn_time); FILE *time_file = fopen("wq_sort.times", "w"); if (time_file) { fprintf(time_file, "Partition time: %llu\n", part_time); fprintf(time_file, "Parallel time: %llu\n", parallel_time); fprintf(time_file, "Merge time: %llu\n", merge_time); if(sample_env) fprintf(time_file, "Sampling time: %llu\n", sample_time); fprintf(time_file, "Workload execution time: %llu\n", workload_runtime); fprintf(time_file, "Total execution time: %llu\n", execn_time); } fclose(time_file); work_queue_delete(q); free(outfile); return 0; }