1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349
|
/*
* Generate/analyze pareto/zipf distributions to better understand
* what an access pattern would look like.
*
* For instance, the following would generate a zipf distribution
* with theta 1.2, using 262144 (1 GiB / 4096) values and split the
* reporting into 20 buckets:
*
* ./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 <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include "../lib/zipf.h"
#include "../lib/gauss.h"
#include "../flist.h"
#include "../hash.h"
#define DEF_NR_OUTPUT 20
struct node {
struct flist_head list;
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;
enum {
TYPE_NONE = 0,
TYPE_ZIPF,
TYPE_PARETO,
TYPE_NORMAL,
};
static const char *dist_types[] = { "None", "Zipf", "Pareto", "Normal" };
enum {
OUTPUT_NORMAL,
OUTPUT_CSV,
};
static int dist_type = TYPE_ZIPF;
static unsigned long gib_size = 500;
static unsigned long block_size = 4096;
static unsigned long output_nranges = DEF_NR_OUTPUT;
static double percentage;
static double dist_val;
static int output_type = OUTPUT_NORMAL;
#define DEF_ZIPF_VAL 1.2
#define DEF_PARETO_VAL 0.3
static unsigned int hashv(unsigned long long val)
{
return jhash(&val, sizeof(val), 0) & (hash_size - 1);
}
static struct node *hash_lookup(unsigned long long val)
{
struct flist_head *l = &hash[hashv(val)];
struct flist_head *entry;
struct node *n;
flist_for_each(entry, l) {
n = flist_entry(entry, struct node, list);
if (n->val == val)
return n;
}
return NULL;
}
static void hash_insert(struct node *n, unsigned long long val)
{
struct flist_head *l = &hash[hashv(val)];
n->val = val;
n->hits = 1;
flist_add_tail(&n->list, l);
}
static void usage(void)
{
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, pareto, or normal)\n");
printf("\t-i\tDistribution algorithm input (zipf theta, pareto power,\n"
"\t\tor normal %% deviation)\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: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 'b':
block_size = strtoul(optarg, NULL, 10);
break;
case 't':
if (!strncmp(optarg, "zipf", 4))
dist_type = TYPE_ZIPF;
else if (!strncmp(optarg, "pareto", 6))
dist_type = TYPE_PARETO;
else if (!strncmp(optarg, "normal", 6))
dist_type = TYPE_NORMAL;
else {
printf("wrong dist type: %s\n", optarg);
return 1;
}
break;
case 'g':
gib_size = strtoul(optarg, NULL, 10);
break;
case 'i':
dist_val = atof(optarg);
dist_val_set = 1;
break;
case 'o':
output_nranges = strtoul(optarg, NULL, 10);
break;
case 'c':
output_type = OUTPUT_CSV;
break;
default:
printf("bad option %c\n", c);
return 1;
}
}
if (dist_type == TYPE_PARETO) {
if ((dist_val >= 1.00 || dist_val < 0.00)) {
printf("pareto input must be > 0.00 and < 1.00\n");
return 1;
}
if (!dist_val_set)
dist_val = DEF_PARETO_VAL;
} else if (dist_type == TYPE_ZIPF) {
if (dist_val == 1.0) {
printf("zipf input must be different than 1.0\n");
return 1;
}
if (!dist_val_set)
dist_val = DEF_ZIPF_VAL;
}
return 0;
}
struct output_sum {
double output;
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;
}
static void output_csv(struct node *nodes, unsigned long nnodes)
{
unsigned long i;
printf("rank, count\n");
for (i = 0; i < nnodes; i++)
printf("%lu, %lu\n", i, nodes[i].hits);
}
static void output_normal(struct node *nodes, unsigned long nnodes,
unsigned long nranges)
{
unsigned long i, j, cur_vals, interval_step, next_interval, total_vals;
unsigned long blocks = percentage * nnodes / 100;
double hit_percent_sum = 0;
unsigned long long hit_sum = 0;
double perc, perc_i;
struct output_sum *output_sums;
interval_step = (nnodes - 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;
}
j = total_vals = cur_vals = 0;
for (i = 0; i < nnodes; i++) {
struct output_sum *os = &output_sums[j];
struct node *node = &nodes[i];
cur_vals += node->hits;
total_vals += node->hits;
os->nranges += node->hits;
if (i == (next_interval) -1 || i == nnodes - 1) {
os->output = (double) cur_vals / (double) nranges;
os->output *= 100.0;
cur_vals = 0;
next_interval += interval_step;
j++;
}
if (percentage) {
if (total_vals >= blocks) {
double cs = (double) i * block_size / (1024.0 * 1024.0);
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;
}
}
}
perc_i = 100.0 / (double)output_nranges;
perc = 0.0;
printf("\n Rows Hits %% Sum %% # 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';
if (gb > 1024.0) {
p = 'M';
gb /= 1024.0;
}
if (gb > 1024.0) {
p = 'G';
gb /= 1024.0;
}
perc += perc_i;
hit_percent_sum += os->output;
hit_sum += os->nranges;
printf("%s %6.2f%%\t%6.2f%%\t\t%6.2f%%\t\t%8u\t%6.2f%c\n",
i ? "|->" : "Top", perc, os->output, hit_percent_sum,
os->nranges, gb, p);
}
printf("-----------------------------------------------------------------------\n");
printf("Total\t\t\t\t\t\t%8llu\n", hit_sum);
free(output_sums);
}
int main(int argc, char *argv[])
{
unsigned long offset;
unsigned long long nranges;
unsigned long nnodes;
struct node *nodes;
struct zipf_state zs;
struct gauss_state gs;
int i, j;
if (parse_options(argc, argv))
return 1;
if (output_type != OUTPUT_CSV)
printf("Generating %s distribution with %f input and %lu GiB size and %lu block_size.\n",
dist_types[dist_type], dist_val, gib_size, block_size);
nranges = gib_size * 1024 * 1024 * 1024ULL;
nranges /= block_size;
if (dist_type == TYPE_ZIPF)
zipf_init(&zs, nranges, dist_val, 1);
else if (dist_type == TYPE_PARETO)
pareto_init(&zs, nranges, dist_val, 1);
else
gauss_init(&gs, nranges, dist_val, 1);
hash_bits = 0;
hash_size = nranges;
while ((hash_size >>= 1) != 0)
hash_bits++;
hash_size = 1 << hash_bits;
hash = calloc(hash_size, sizeof(struct flist_head));
for (i = 0; i < hash_size; i++)
INIT_FLIST_HEAD(&hash[i]);
nodes = malloc(nranges * sizeof(struct node));
for (i = j = 0; i < nranges; i++) {
struct node *n;
if (dist_type == TYPE_ZIPF)
offset = zipf_next(&zs);
else if (dist_type == TYPE_PARETO)
offset = pareto_next(&zs);
else
offset = gauss_next(&gs);
n = hash_lookup(offset);
if (n)
n->hits++;
else {
hash_insert(&nodes[j], offset);
j++;
}
}
qsort(nodes, j, sizeof(struct node), node_cmp);
nnodes = j;
if (output_type == OUTPUT_CSV)
output_csv(nodes, nnodes);
else
output_normal(nodes, nnodes, nranges);
free(hash);
free(nodes);
return 0;
}
|