man: sync "JOB FILE FORMAT" section with HOWTO
[fio.git] / stat.h
CommitLineData
a64e88da
JA
1#ifndef FIO_STAT_H
2#define FIO_STAT_H
3
ec41265e 4#include "iolog.h"
a666cab8 5#include "lib/output_buffer.h"
ec41265e 6
a64e88da 7struct group_run_stats {
6eaf09d6
SL
8 uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
9 uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
af7f87cb 10 uint64_t iobytes[DDIR_RWDIR_CNT];
6eaf09d6 11 uint64_t agg[DDIR_RWDIR_CNT];
a64e88da 12 uint32_t kb_base;
ad705bcb 13 uint32_t unit_base;
a64e88da 14 uint32_t groupid;
771e58be 15 uint32_t unified_rw_rep;
e5021b34 16} __attribute__((packed));
a64e88da
JA
17
18/*
19 * How many depth levels to log
20 */
21#define FIO_IO_U_MAP_NR 7
d6bb626e 22#define FIO_IO_U_LAT_N_NR 10
a64e88da
JA
23#define FIO_IO_U_LAT_U_NR 10
24#define FIO_IO_U_LAT_M_NR 12
25
26/*
27 * Aggregate clat samples to report percentile(s) of them.
28 *
29 * EXECUTIVE SUMMARY
30 *
31 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
32 * value of resulting percentiles. The error will be approximately
33 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
34 *
35 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
36 * range being tracked for latency samples. The maximum value tracked
37 * accurately will be 2^(GROUP_NR + PLAT_BITS -1) microseconds.
38 *
39 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
40 * requirement of storing those aggregate counts. The memory used will
41 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
42 * bytes.
43 *
44 * FIO_IO_U_PLAT_NR is the total number of buckets.
45 *
46 * DETAILS
47 *
48 * Suppose the clat varies from 0 to 999 (usec), the straightforward
49 * method is to keep an array of (999 + 1) buckets, in which a counter
50 * keeps the count of samples which fall in the bucket, e.g.,
51 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
52 * and can be avoided if an approximation is acceptable.
53 *
54 * One such method is to let the range of the bucket to be greater
55 * than one. This method has low accuracy when the value is small. For
56 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
57 * the represented value of each bucket be the mean of the range. Then
58 * a value 0 has an round-off error of 49.5. To improve on this, we
59 * use buckets with non-uniform ranges, while bounding the error of
60 * each bucket within a ratio of the sample value. A simple example
61 * would be when error_bound = 0.005, buckets are {
62 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
63 * {[900,909],[910,919]...} }. The total range is partitioned into
64 * groups with different ranges, then buckets with uniform ranges. An
65 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
66 *
67 * For better efficiency, we implement this using base two. We group
68 * samples by their Most Significant Bit (MSB), extract the next M bit
69 * of them as an index within the group, and discard the rest of the
70 * bits.
71 *
72 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
73 * and use M bit for indexing
74 *
75 * | n | M bits | bit (n-M-1) ... bit 0 |
76 *
77 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
78 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
79 * error
80 *
81 * 2^(n-M)-1 2^(n-M) 1
82 * e <= --------- <= ------- = ---
83 * 2^n 2^n 2^M
84 *
85 * Furthermore, we use "mean" of the range to represent the bucket,
86 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
87 * as the index, each group must contains 2^M buckets.
88 *
89 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
90 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
91 *
92 * Group MSB #discarded range of #buckets
93 * error_bits value
94 * ----------------------------------------------------------------
95 * 0* 0~5 0 [0,63] 64
96 * 1* 6 0 [64,127] 64
97 * 2 7 1 [128,255] 64
98 * 3 8 2 [256,511] 64
99 * 4 9 3 [512,1023] 64
100 * ... ... ... [...,...] ...
101 * 18 23 17 [8838608,+inf]** 64
102 *
103 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
104 * the value cannot be rounded off. Use all bits of the sample as
105 * index.
106 *
107 * ** If a sample's MSB is greater than 23, it will be counted as 23.
108 */
109
110#define FIO_IO_U_PLAT_BITS 6
111#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
d6bb626e 112#define FIO_IO_U_PLAT_GROUP_NR 29
a64e88da
JA
113#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
114#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
115 list of percentiles */
116
66347cfa
DE
117/*
118 * Trim cycle count measurements
119 */
120#define MAX_NR_BLOCK_INFOS 8192
121#define BLOCK_INFO_STATE_SHIFT 29
122#define BLOCK_INFO_TRIMS(block_info) \
123 ((block_info) & ((1 << BLOCK_INFO_STATE_SHIFT) - 1))
124#define BLOCK_INFO_STATE(block_info) \
125 ((block_info) >> BLOCK_INFO_STATE_SHIFT)
126#define BLOCK_INFO(state, trim_cycles) \
fc8d6d05 127 ((trim_cycles) | ((unsigned int) (state) << BLOCK_INFO_STATE_SHIFT))
66347cfa
DE
128#define BLOCK_INFO_SET_STATE(block_info, state) \
129 BLOCK_INFO(state, BLOCK_INFO_TRIMS(block_info))
130enum block_info_state {
131 BLOCK_STATE_UNINIT,
132 BLOCK_STATE_TRIMMED,
133 BLOCK_STATE_WRITTEN,
134 BLOCK_STATE_TRIM_FAILURE,
135 BLOCK_STATE_WRITE_FAILURE,
136 BLOCK_STATE_COUNT,
1d6d3455 137};
66347cfa 138
a64e88da
JA
139#define MAX_PATTERN_SIZE 512
140#define FIO_JOBNAME_SIZE 128
4e59d0f3 141#define FIO_JOBDESC_SIZE 256
a64e88da
JA
142#define FIO_VERROR_SIZE 128
143
144struct thread_stat {
145 char name[FIO_JOBNAME_SIZE];
146 char verror[FIO_VERROR_SIZE];
ddcc0b69 147 uint32_t error;
2f122b13 148 uint32_t thread_number;
ddcc0b69 149 uint32_t groupid;
a64e88da 150 uint32_t pid;
4e59d0f3 151 char description[FIO_JOBDESC_SIZE];
a64e88da 152 uint32_t members;
771e58be 153 uint32_t unified_rw_rep;
a64e88da
JA
154
155 /*
156 * bandwidth and latency stats
157 */
6eaf09d6
SL
158 struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
159 struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
160 struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
161 struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
162 struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
a64e88da
JA
163
164 /*
165 * fio system usage accounting
166 */
167 uint64_t usr_time;
168 uint64_t sys_time;
169 uint64_t ctx;
170 uint64_t minf, majf;
171
172 /*
173 * IO depth and latency stats
174 */
175 uint64_t clat_percentiles;
435d195a 176 uint64_t percentile_precision;
802ad4a8 177 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
a64e88da
JA
178
179 uint32_t io_u_map[FIO_IO_U_MAP_NR];
180 uint32_t io_u_submit[FIO_IO_U_MAP_NR];
181 uint32_t io_u_complete[FIO_IO_U_MAP_NR];
d6bb626e 182 uint32_t io_u_lat_n[FIO_IO_U_LAT_N_NR];
a64e88da
JA
183 uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
184 uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
6eaf09d6 185 uint32_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
3d0ebb30
GG
186 uint32_t pad;
187
de5cdfea
JA
188 uint64_t total_io_u[DDIR_RWDIR_CNT];
189 uint64_t short_io_u[DDIR_RWDIR_CNT];
190 uint64_t drop_io_u[DDIR_RWDIR_CNT];
a64e88da
JA
191 uint64_t total_submit;
192 uint64_t total_complete;
193
6eaf09d6
SL
194 uint64_t io_bytes[DDIR_RWDIR_CNT];
195 uint64_t runtime[DDIR_RWDIR_CNT];
a64e88da
JA
196 uint64_t total_run_time;
197
198 /*
199 * IO Error related stats
200 */
3d0ebb30
GG
201 union {
202 uint16_t continue_on_error;
11e955e3 203 uint32_t pad2;
3d0ebb30 204 };
ddcc0b69 205 uint32_t first_error;
11e955e3 206 uint64_t total_err_count;
a64e88da 207
66347cfa
DE
208 uint64_t nr_block_infos;
209 uint32_t block_infos[MAX_NR_BLOCK_INFOS];
210
a64e88da 211 uint32_t kb_base;
ad705bcb 212 uint32_t unit_base;
3e260a46
JA
213
214 uint32_t latency_depth;
11e955e3 215 uint32_t pad3;
3e260a46
JA
216 uint64_t latency_target;
217 fio_fp64_t latency_percentile;
218 uint64_t latency_window;
16e56d25 219
bb49c8bd
VF
220 uint64_t ss_dur;
221 uint32_t ss_state;
222 uint32_t ss_head;
bb49c8bd 223
bb49c8bd
VF
224 fio_fp64_t ss_limit;
225 fio_fp64_t ss_slope;
226 fio_fp64_t ss_deviation;
227 fio_fp64_t ss_criterion;
cb84f1fa
VF
228
229 union {
230 uint64_t *ss_iops_data;
11e955e3 231 uint64_t pad4;
cb84f1fa
VF
232 };
233
234 union {
235 uint64_t *ss_bw_data;
11e955e3 236 uint64_t pad5;
cb84f1fa 237 };
e5021b34 238} __attribute__((packed));
a64e88da 239
b75a394f
JA
240struct jobs_eta {
241 uint32_t nr_running;
242 uint32_t nr_ramp;
819db5f5 243
b75a394f 244 uint32_t nr_pending;
714e85f3 245 uint32_t nr_setting_up;
819db5f5 246
90eff1c9 247 uint64_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
90eff1c9 248 uint64_t rate[DDIR_RWDIR_CNT];
dbf285f7 249 uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
6eaf09d6 250 uint32_t iops[DDIR_RWDIR_CNT];
b75a394f
JA
251 uint64_t elapsed_sec;
252 uint64_t eta_sec;
b7f05eb0 253 uint32_t is_pow2;
ad705bcb 254 uint32_t unit_base;
1d1f45ae 255
dbf285f7
SW
256 uint32_t files_open;
257
1d1f45ae
JA
258 /*
259 * Network 'copy' of run_str[]
260 */
261 uint32_t nr_threads;
372aecb9 262 uint8_t run_str[];
e5021b34 263} __attribute__((packed));
b75a394f 264
65a4d15c
KC
265struct io_u_plat_entry {
266 struct flist_head list;
267 unsigned int io_u_plat[FIO_IO_U_PLAT_NR];
268};
269
e5437a07
VT
270extern struct fio_mutex *stat_mutex;
271
c5103619 272extern struct jobs_eta *get_jobs_eta(bool force, size_t *size);
723297c9 273
cef9175e
JA
274extern void stat_init(void);
275extern void stat_exit(void);
276
0279b880 277extern struct json_object * show_thread_status(struct thread_stat *ts, struct group_run_stats *rs, struct flist_head *, struct buf_output *);
a666cab8 278extern void show_group_stats(struct group_run_stats *rs, struct buf_output *);
8aa89d70 279extern bool calc_thread_status(struct jobs_eta *je, int force);
cf451d1e 280extern void display_thread_status(struct jobs_eta *je);
5b9babb7 281extern void show_run_stats(void);
83f7b64e 282extern void __show_run_stats(void);
5ddc6707 283extern void __show_running_run_stats(void);
b852e7cf 284extern void show_running_run_stats(void);
06464907 285extern void check_for_running_stats(void);
fd595830 286extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, bool first);
37f0c1ae
JA
287extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
288extern void init_thread_stat(struct thread_stat *ts);
289extern void init_group_run_stat(struct group_run_stats *gs);
3e47bd25 290extern void eta_to_str(char *str, unsigned long eta_sec);
d6bb626e
VF
291extern bool calc_lat(struct io_stat *is, unsigned long long *min, unsigned long long *max, double *mean, double *dev);
292extern unsigned int calc_clat_percentiles(unsigned int *io_u_plat, unsigned long nr, fio_fp64_t *plist, unsigned long long **output, unsigned long long *maxv, unsigned long long *minv);
fea76a0c 293extern void stat_calc_lat_n(struct thread_stat *ts, double *io_u_lat);
e5bd1347
JA
294extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
295extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
2e33101f 296extern void stat_calc_dist(unsigned int *map, unsigned long total, double *io_u_dist);
6bb58215 297extern void reset_io_stats(struct thread_data *);
210dd0fc
JA
298extern void update_rusage_stat(struct thread_data *);
299extern void clear_rusage_stat(struct thread_data *);
2e33101f 300
d6bb626e 301extern void add_lat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
cf8f852a 302 unsigned int, uint64_t);
d6bb626e 303extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
cf8f852a
JA
304 unsigned int, uint64_t);
305extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long,
306 unsigned int, uint64_t);
af2fde19 307extern void add_agg_sample(union io_sample_data, enum fio_ddir, unsigned int);
a47591e4
JA
308extern void add_iops_sample(struct thread_data *, struct io_u *,
309 unsigned int);
310extern void add_bw_sample(struct thread_data *, struct io_u *,
d6bb626e 311 unsigned int, unsigned long long);
a47591e4 312extern int calc_log_samples(void);
cf8f852a
JA
313
314extern struct io_log *agg_io_log[DDIR_RWDIR_CNT];
315extern int write_bw_log;
316
74558486
JA
317static inline bool nsec_to_usec(unsigned long long *min,
318 unsigned long long *max, double *mean,
319 double *dev)
b29ad562 320{
d6bb626e 321 if (*min > 2000 && *max > 99999 && *dev > 1000.0) {
b29ad562
JA
322 *min /= 1000;
323 *max /= 1000;
324 *mean /= 1000.0;
325 *dev /= 1000.0;
8aa89d70 326 return true;
b29ad562
JA
327 }
328
8aa89d70 329 return false;
b29ad562 330}
d6bb626e 331
74558486
JA
332static inline bool nsec_to_msec(unsigned long long *min,
333 unsigned long long *max, double *mean,
334 double *dev)
d6bb626e
VF
335{
336 if (*min > 2000000 && *max > 99999999ULL && *dev > 1000000.0) {
337 *min /= 1000000;
338 *max /= 1000000;
339 *mean /= 1000000.0;
340 *dev /= 1000000.0;
341 return true;
342 }
343
344 return false;
345}
74558486 346
723297c9
JA
347/*
348 * Worst level condensing would be 1:5, so allow enough room for that
349 */
350#define __THREAD_RUNSTR_SZ(nr) ((nr) * 5)
11155857
JA
351#define THREAD_RUNSTR_SZ __THREAD_RUNSTR_SZ(thread_number)
352
66347cfa
DE
353uint32_t *io_u_block_info(struct thread_data *td, struct io_u *io_u);
354
a64e88da 355#endif