X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=lib%2Fzipf.c;h=9b6ce6334836baa1f0f3c0c238da24f319ad4c3b;hb=2c24c93850cc4fa2dfe6b521231a69d6c116bba4;hp=527ae294cf9afd6212151e004e2220163e25f888;hpb=8e600258bad065fbdfd6a1b2856077d12cd521e5;p=fio.git diff --git a/lib/zipf.c b/lib/zipf.c index 527ae294..9b6ce633 100644 --- a/lib/zipf.c +++ b/lib/zipf.c @@ -10,104 +10,48 @@ #include "zipf.h" #include "../minmax.h" #include "../hash.h" -#include "../os/os.h" -struct fio_zipf_disk { - uint64_t ver_magic; - uint64_t nranges; - uint64_t zetan; -}; - -#define FIO_ZIPF_DISK_MAGIC 0x7a697066 -#define FIO_ZIPF_DISK_VER 1 -#define FIO_ZIPF_MAGIC ((FIO_ZIPF_DISK_MAGIC << 16) | FIO_ZIPF_DISK_VER) - -static void write_zipf(struct zipf_state *zs) -{ - struct fio_zipf_disk f; - char tmp[80]; - int fd; - - sprintf(tmp, "fio.zipf.%f.%llu", zs->theta, (unsigned long long) zs->nranges); - fd = open(tmp, O_CREAT | O_WRONLY, 0644); - if (fd == -1) - return; - - f.ver_magic = __cpu_to_le64(FIO_ZIPF_MAGIC); - f.nranges = __cpu_to_le64(zs->nranges); - f.zetan = __cpu_to_le64(fio_double_to_uint64(zs->zetan)); - if (write(fd, &f, sizeof(f)) != sizeof(f)) - unlink(tmp); - - close(fd); -} +#define ZIPF_MAX_GEN 10000000 static void zipf_update(struct zipf_state *zs) { + unsigned long to_gen; unsigned int i; - log_info("fio: generating zetan for theta=%f, ranges=%lu\n", zs->theta, zs->nranges); + /* + * It can become very costly to generate long sequences. Just cap it at + * 10M max, that should be doable in 1-2s on even slow machines. + * Precision will take a slight hit, but nothing major. + */ + to_gen = min(zs->nranges, ZIPF_MAX_GEN); - for (i = 0; i < zs->nranges; i++) + for (i = 0; i < to_gen; i++) zs->zetan += pow(1.0 / (double) (i + 1), zs->theta); - - write_zipf(zs); } -static void zipf_load_gen_zeta(struct zipf_state *zs) +static void shared_rand_init(struct zipf_state *zs, unsigned long nranges, + unsigned int seed) { - struct fio_zipf_disk f; - char tmp[80]; - int fd; - - sprintf(tmp, "fio.zipf.%f.%llu", zs->theta, (unsigned long long) zs->nranges); - fd = open(tmp, O_RDONLY); - if (fd == -1) { -punt: - zipf_update(zs); - return; - } - - if (read(fd, &f, sizeof(f)) != sizeof(f)) { - close(fd); - goto punt; - } - - close(fd); - - f.ver_magic = le64_to_cpu(f.ver_magic); - f.nranges = le64_to_cpu(f.nranges); - f.zetan = le64_to_cpu(f.zetan); - - if (f.ver_magic != FIO_ZIPF_MAGIC) { - unlink(tmp); - goto punt; - } - - zs->zetan = fio_uint64_to_double(f.zetan); - zs->nranges = f.nranges; + memset(zs, 0, sizeof(*zs)); + zs->nranges = nranges; + + init_rand_seed(&zs->rand, seed); + zs->rand_off = __rand(&zs->rand); } -void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta) +void zipf_init(struct zipf_state *zs, unsigned long nranges, double theta, + unsigned int seed) { - unsigned int i; + shared_rand_init(zs, nranges, seed); - memset(zs, 0, sizeof(*zs)); - - zs->nranges = nranges; zs->theta = theta; + zs->zeta2 = pow(1.0, zs->theta) + pow(0.5, zs->theta); - for (i = 1; i <= 2; i++) - zs->zeta2 += pow(1.0 / (double) i, zs->theta); - - init_rand(&zs->rand); - - zipf_load_gen_zeta(zs); + zipf_update(zs); } unsigned long long zipf_next(struct zipf_state *zs) { - double alpha, eta, rand_uni, rand_z; unsigned long long n = zs->nranges; unsigned long long val; @@ -125,17 +69,14 @@ unsigned long long zipf_next(struct zipf_state *zs) else val = 1 + (unsigned long long)(n * pow(eta*rand_uni - eta + 1.0, alpha)); - return __hash_long(val - 1) % zs->nranges; + return (__hash_u64(val - 1) + zs->rand_off) % zs->nranges; } -void pareto_init(struct zipf_state *zs, unsigned long nranges, double h) +void pareto_init(struct zipf_state *zs, unsigned long nranges, double h, + unsigned int seed) { - memset(zs, 0, sizeof(*zs)); - - zs->nranges = nranges; + shared_rand_init(zs, nranges, seed); zs->pareto_pow = log(h) / log(1.0 - h); - - init_rand(&zs->rand); } unsigned long long pareto_next(struct zipf_state *zs) @@ -143,5 +84,5 @@ unsigned long long pareto_next(struct zipf_state *zs) double rand = (double) __rand(&zs->rand) / (double) FRAND_MAX; unsigned long long n = zs->nranges - 1; - return __hash_long(n * pow(rand, zs->pareto_pow)) % zs->nranges; + return (__hash_u64(n * pow(rand, zs->pareto_pow)) + zs->rand_off) % zs->nranges; }