From 2c9b3ccdc4d6aa5853c37d327db0249f41dff278 Mon Sep 17 00:00:00 2001 From: Fabrice Bacchella Date: Wed, 8 Apr 2015 16:49:42 +0200 Subject: [PATCH] fio-genzipf results output are slightly wrong, it doesn't help to understand how to use it: * The last row was not displayed * some useless variables were used * the calculation for the output table missed a few events, so the sum of 'No Hits' column was wrong * Wrong sample in genzipf.c header, used old style arguments * the help message for -o options was misleading. * useless define of DEF_NR. * default of default row number to a more common value of 20, instead of 23. --- t/genzipf.c | 64 ++++++++++++++++++++++++----------------------------- 1 file changed, 29 insertions(+), 35 deletions(-) diff --git a/t/genzipf.c b/t/genzipf.c index c5f098c4..9ba19bf9 100644 --- a/t/genzipf.c +++ b/t/genzipf.c @@ -3,10 +3,10 @@ * what an access pattern would look like. * * For instance, the following would generate a zipf distribution - * with theta 1.2, using 100,000 values and split the reporting into + * with theta 1.2, using 262144 (1 GB / 4096) values and split the reporting into * 20 buckets: * - * t/genzipf zipf 1.2 100000 20 + * ./t/fio-genzipf -t zipf -i 1.2 -g 1 -b 4096 -o 20 * * Only the distribution type (zipf or pareto) and spread input need * to be given, if not given defaults are used. @@ -22,8 +22,7 @@ #include "../flist.h" #include "../hash.h" -#define DEF_NR 1000000 -#define DEF_NR_OUTPUT 23 +#define DEF_NR_OUTPUT 20 struct node { struct flist_head list; @@ -87,7 +86,7 @@ static void usage(void) printf("\t-i\tDistribution algorithm input (zipf theta or pareto power)\n"); printf("\t-b\tBlock size of a given range (in bytes)\n"); printf("\t-g\tSize of data set (in gigabytes)\n"); - printf("\t-o\tNumber of output columns\n"); + printf("\t-o\tNumber of output rows\n"); printf("\t-c\tOutput ranges in CSV format\n"); } @@ -171,7 +170,7 @@ static int node_cmp(const void *p1, const void *p2) int main(int argc, char *argv[]) { unsigned long offset; - unsigned long i, j, k, nr_vals, cur_vals, interval, total_vals, nnodes; + unsigned long i, j, k, nr_vals, cur_vals, interval_step, next_interval, total_vals, nnodes; unsigned long long nranges; struct output_sum *output_sums; struct node *nodes; @@ -205,7 +204,7 @@ int main(int argc, char *argv[]) nodes = malloc(nranges * sizeof(struct node)); - for (nr_vals = i = j = 0; i < nranges; i++) { + for (i = j = 0; i < nranges; i++) { struct node *n; if (dist_type == TYPE_ZIPF) @@ -221,7 +220,6 @@ int main(int argc, char *argv[]) j++; } - nr_vals++; } qsort(nodes, j, sizeof(struct node), node_cmp); @@ -230,47 +228,40 @@ int main(int argc, char *argv[]) if (output_csv) { printf("rank, count\n"); - for (k = 0; k < nnodes; k++) + for (k = 0; k < nnodes; k++) { printf("%lu, %lu\n", k, nodes[k].hits); + } } else { - interval = (nr_vals + output_nranges - 1) / output_nranges; - + unsigned long blocks = percentage * nranges / 100; + double hit_percent_sum = 0; + unsigned long long hit_sum = 0; + interval_step = (nr_vals - 1) / output_nranges + 1; + next_interval = interval_step; output_sums = malloc(output_nranges * sizeof(struct output_sum)); for (i = 0; i < output_nranges; i++) { output_sums[i].output = 0.0; - output_sums[i].nranges = 1; + output_sums[i].nranges = 0; } - total_vals = i = j = cur_vals = 0; + j = total_vals = cur_vals = 0; for (k = 0; k < nnodes; k++) { struct output_sum *os = &output_sums[j]; struct node *node = &nodes[k]; - - if (i >= interval) { - os->output = - (double)(cur_vals + 1) / (double)nranges; + cur_vals += node->hits; + total_vals += node->hits; + os->nranges += node->hits; + if (k == (next_interval) -1 || k == (nnodes - 1)) { + os->output = (double)(cur_vals) / (double)nranges; os->output *= 100.0; + cur_vals = 0; + next_interval += interval_step; j++; - cur_vals = node->hits; - interval += - (nr_vals + output_nranges - - 1) / output_nranges; - } else { - cur_vals += node->hits; - os->nranges += node->hits; } - i++; - total_vals += node->hits; - if (percentage) { - unsigned long blocks = - percentage * nranges / 100; - if (total_vals >= blocks) { - double cs = - i * block_size / (1024 * 1024); + double cs = k * block_size / (1024 * 1024); char p = 'M'; if (cs > 1024.0) { @@ -291,9 +282,9 @@ int main(int argc, char *argv[]) perc_i = 100.0 / (double)output_nranges; perc = 0.0; - printf("\n Rows Hits No Hits Size\n"); + printf("\n Rows Hits %% No Hits Size\n"); printf("--------------------------------------------------------\n"); - for (i = 0; i < j; i++) { + for (i = 0; i < output_nranges; i++) { struct output_sum *os = &output_sums[i]; double gb = (double)os->nranges * block_size / 1024.0; char p = 'K'; @@ -308,11 +299,14 @@ int main(int argc, char *argv[]) } perc += perc_i; + hit_percent_sum += os->output; + hit_sum += os->nranges; printf("%s %6.2f%%\t%6.2f%%\t\t%8u\t%6.2f%c\n", i ? "|->" : "Top", perc, os->output, os->nranges, gb, p); } - + printf("--------------------------------------------------------\n"); + printf("Total\t\t%6.2f%%\t\t%8llu\n", hit_percent_sum, hit_sum); free(output_sums); } -- 2.25.1