1 #include <math.h>
2 #include <string.h>
3 #include <inttypes.h>
4 #include <stdio.h>
5 #include <unistd.h>
6 #include <sys/types.h>
7 #include <fcntl.h>
8 #include "ieee754.h"
9 #include "../log.h"
10 #include "zipf.h"
11 #include "../minmax.h"
12 #include "../hash.h"
13
14 #define ZIPF_MAX_GEN 10000000UL
15
zipf_update(struct zipf_state * zs)16 static void zipf_update(struct zipf_state *zs)
17 {
18 unsigned long to_gen;
19 unsigned int i;
20
21 /*
22 * It can become very costly to generate long sequences. Just cap it at
23 * 10M max, that should be doable in 1-2s on even slow machines.
24 * Precision will take a slight hit, but nothing major.
25 */
26 to_gen = min(zs->nranges, (uint64_t) ZIPF_MAX_GEN);
27
28 for (i = 0; i < to_gen; i++)
29 zs->zetan += pow(1.0 / (double) (i + 1), zs->theta);
30 }
31
shared_rand_init(struct zipf_state * zs,unsigned long nranges,unsigned int seed)32 static void shared_rand_init(struct zipf_state *zs, unsigned long nranges,
33 unsigned int seed)
34 {
35 memset(zs, 0, sizeof(*zs));
36 zs->nranges = nranges;
37
38 init_rand_seed(&zs->rand, seed);
39 zs->rand_off = __rand(&zs->rand);
40 }
41
zipf_init(struct zipf_state * zs,unsigned long nranges,double theta,unsigned int seed)42 void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta,
43 unsigned int seed)
44 {
45 shared_rand_init(zs, nranges, seed);
46
47 zs->theta = theta;
48 zs->zeta2 = pow(1.0, zs->theta) + pow(0.5, zs->theta);
49
50 zipf_update(zs);
51 }
52
zipf_next(struct zipf_state * zs)53 unsigned long long zipf_next(struct zipf_state *zs)
54 {
55 double alpha, eta, rand_uni, rand_z;
56 unsigned long long n = zs->nranges;
57 unsigned long long val;
58
59 alpha = 1.0 / (1.0 - zs->theta);
60 eta = (1.0 - pow(2.0 / n, 1.0 - zs->theta)) / (1.0 - zs->zeta2 / zs->zetan);
61
62 rand_uni = (double) __rand(&zs->rand) / (double) FRAND_MAX;
63 rand_z = rand_uni * zs->zetan;
64
65 if (rand_z < 1.0)
66 val = 1;
67 else if (rand_z < (1.0 + pow(0.5, zs->theta)))
68 val = 2;
69 else
70 val = 1 + (unsigned long long)(n * pow(eta*rand_uni - eta + 1.0, alpha));
71
72 return (__hash_u64(val - 1) + zs->rand_off) % zs->nranges;
73 }
74
pareto_init(struct zipf_state * zs,unsigned long nranges,double h,unsigned int seed)75 void pareto_init(struct zipf_state *zs, unsigned long nranges, double h,
76 unsigned int seed)
77 {
78 shared_rand_init(zs, nranges, seed);
79 zs->pareto_pow = log(h) / log(1.0 - h);
80 }
81
pareto_next(struct zipf_state * zs)82 unsigned long long pareto_next(struct zipf_state *zs)
83 {
84 double rand = (double) __rand(&zs->rand) / (double) FRAND_MAX;
85 unsigned long long n = zs->nranges - 1;
86
87 return (__hash_u64(n * pow(rand, zs->pareto_pow)) + zs->rand_off) % zs->nranges;
88 }
89