* 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.
#include "../lib/zipf.h"
#include "../flist.h"
#include "../hash.h"
-#include "../rbtree.h"
-#define DEF_NR 1000000
-#define DEF_NR_OUTPUT 23
+#define DEF_NR_OUTPUT 20
struct node {
struct flist_head list;
- struct rb_node rb;
unsigned long long val;
unsigned long hits;
};
static struct flist_head *hash;
static unsigned long hash_bits = 24;
static unsigned long hash_size = 1 << 24;
-static struct rb_root rb;
enum {
TYPE_NONE = 0,
static unsigned long output_nranges = DEF_NR_OUTPUT;
static double percentage;
static double dist_val;
+static int output_csv = 0;
#define DEF_ZIPF_VAL 1.2
#define DEF_PARETO_VAL 0.3
return NULL;
}
-static void hash_insert(unsigned long long val)
+static struct node *hash_insert(struct node *n, unsigned long long val)
{
struct flist_head *l = &hash[hash_long(val, hash_bits)];
- struct node *n = malloc(sizeof(*n));
n->val = val;
n->hits = 1;
flist_add_tail(&n->list, l);
+ return n;
}
-static void rb_insert(struct node *n)
+static void usage(void)
{
- struct rb_node **p, *parent;
-
- memset(&n->rb, 0, sizeof(n->rb));
- p = &rb.rb_node;
- parent = NULL;
- while (*p) {
- struct node *__n;
-
- parent = *p;
- __n = rb_entry(parent, struct node, rb);
- if (n->hits > __n->hits)
- p = &(*p)->rb_left;
- else
- p = &(*p)->rb_right;
- }
-
- rb_link_node(&n->rb, parent, p);
- rb_insert_color(&n->rb, &rb);
-}
-
-static unsigned long rb_add(struct flist_head *list)
-{
- struct flist_head *entry;
- unsigned long ret = 0;
- struct node *n;
-
- flist_for_each(entry, list) {
- n = flist_entry(entry, struct node, list);
-
- rb_insert(n);
- ret++;
- }
-
- return ret;
-}
-
-static unsigned long rb_gen(void)
-{
- unsigned long ret = 0;
- unsigned int i;
-
- for (i = 0; i < hash_size; i++)
- ret += rb_add(&hash[i]);
-
- return ret;
+ printf("genzipf: test zipf/pareto values for fio input\n");
+ printf("\t-h\tThis help screen\n");
+ printf("\t-p\tGenerate size of data set that are hit by this percentage\n");
+ printf("\t-t\tDistribution type (zipf or pareto)\n");
+ 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 rows\n");
+ printf("\t-c\tOutput ranges in CSV format\n");
}
static int parse_options(int argc, char *argv[])
{
- const char *optstring = "t:g:i:o:b:p:";
+ const char *optstring = "t:g:i:o:b:p:ch";
int c, dist_val_set = 0;
while ((c = getopt(argc, argv, optstring)) != -1) {
switch (c) {
+ case 'h':
+ usage();
+ return 1;
case 'p':
percentage = atof(optarg);
break;
case 'o':
output_nranges = strtoul(optarg, NULL, 10);
break;
+ case 'c':
+ output_csv = 1;
+ break;
default:
printf("bad option %c\n", c);
return 1;
unsigned int nranges;
};
+static int node_cmp(const void *p1, const void *p2)
+{
+ const struct node *n1 = p1;
+ const struct node *n2 = p2;
+
+ return n2->hits - n1->hits;
+}
+
int main(int argc, char *argv[])
{
unsigned long offset;
- unsigned long i, j, nr_vals, cur_vals, interval, total_vals;
+ 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;
double perc, perc_i;
struct zipf_state zs;
- struct rb_node *n;
if (parse_options(argc, argv))
return 1;
- printf("Generating %s distribution with %f input and %lu GB size and %lu block_size.\n", dist_types[dist_type], dist_val, gb_size, block_size);
+ if( !output_csv )
+ printf("Generating %s distribution with %f input and %lu GB size and %lu block_size.\n", dist_types[dist_type], dist_val, gb_size, block_size);
nranges = gb_size * 1024 * 1024 * 1024ULL;
nranges /= block_size;
for (i = 0; i < hash_size; i++)
INIT_FLIST_HEAD(&hash[i]);
- for (nr_vals = i = 0; i < nranges; i++) {
+ nodes = malloc(nranges * sizeof(struct node));
+
+ for (i = j = 0; i < nranges; i++) {
struct node *n;
if (dist_type == TYPE_ZIPF)
n = hash_lookup(offset);
if (n)
n->hits++;
- else
- hash_insert(offset);
+ else {
+ hash_insert(&nodes[j], offset);
+ j++;
+ }
- nr_vals++;
}
- nr_vals = rb_gen();
+ qsort(nodes, j, sizeof(struct node), node_cmp);
+ nnodes = j;
+ nr_vals = nnodes;
- interval = (nr_vals + output_nranges - 1) / output_nranges;
-
- 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;
- }
+ if (output_csv) {
+ printf("rank, count\n");
+ for (k = 0; k < nnodes; k++) {
+ printf("%lu, %lu\n", k, nodes[k].hits);
+ }
+ } else {
+ 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 = 0;
+ }
- total_vals = i = j = cur_vals = 0;
-
- n = rb_first(&rb);
- while (n) {
- struct node *node = rb_entry(n, struct node, rb);
- struct output_sum *os = &output_sums[j];
+ j = total_vals = cur_vals = 0;
- if (i >= interval) {
- os->output = (double) (cur_vals + 1) / (double) nranges;
- os->output *= 100.0;
- j++;
- cur_vals = node->hits;
- interval += (nr_vals + output_nranges - 1) / output_nranges;
- } else {
+ for (k = 0; k < nnodes; k++) {
+ struct output_sum *os = &output_sums[j];
+ struct node *node = &nodes[k];
cur_vals += node->hits;
+ total_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);
- char p = 'M';
+ 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++;
+ }
- if (cs > 1024.0) {
- cs /= 1024.0;
- p = 'G';
- }
- if (cs > 1024.0) {
- cs /= 1024.0;
- p = 'T';
+ if (percentage) {
+ if (total_vals >= blocks) {
+ double cs = k * block_size / (1024 * 1024);
+ char p = 'M';
+
+ if (cs > 1024.0) {
+ cs /= 1024.0;
+ p = 'G';
+ }
+ if (cs > 1024.0) {
+ cs /= 1024.0;
+ p = 'T';
+ }
+
+ printf("%.2f%% of hits satisfied in %.3f%cB of cache\n", percentage, cs, p);
+ percentage = 0.0;
}
-
- printf("%.2f%% of hits satisfied in %.3f%cB of cache\n", percentage, cs, p);
- percentage = 0.0;
}
}
- n = rb_next(n);
- }
+ perc_i = 100.0 / (double)output_nranges;
+ perc = 0.0;
- perc_i = 100.0 / (double) output_nranges;
- perc = 0.0;
+ printf("\n Rows Hits %% No Hits Size\n");
+ printf("--------------------------------------------------------\n");
+ 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';
- printf("\n Rows Hits No Hits Size\n");
- printf("--------------------------------------------------------\n");
- for (i = 0; i < j; i++) {
- struct output_sum *os = &output_sums[i];
- double gb = (double) os->nranges * block_size / 1024.0;
- char p = 'K';
+ if (gb > 1024.0) {
+ p = 'M';
+ gb /= 1024.0;
+ }
+ if (gb > 1024.0) {
+ p = 'G';
+ gb /= 1024.0;
+ }
- if (gb > 1024.0) {
- p = 'M';
- gb /= 1024.0;
+ 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);
}
- if (gb > 1024.0) {
- p = 'G';
- gb /= 1024.0;
- }
-
- perc += perc_i;
- 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);
}
- free(output_sums);
free(hash);
+ free(nodes);
return 0;
}
projects / fio.git / blobdiff