6 struct group_run_stats {
7 uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
8 uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
9 uint64_t io_kb[DDIR_RWDIR_CNT];
10 uint64_t agg[DDIR_RWDIR_CNT];
16 * How many depth levels to log
18 #define FIO_IO_U_MAP_NR 7
19 #define FIO_IO_U_LAT_U_NR 10
20 #define FIO_IO_U_LAT_M_NR 12
23 * Aggregate clat samples to report percentile(s) of them.
27 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
28 * value of resulting percentiles. The error will be approximately
29 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
31 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
32 * range being tracked for latency samples. The maximum value tracked
33 * accurately will be 2^(GROUP_NR + PLAT_BITS -1) microseconds.
35 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
36 * requirement of storing those aggregate counts. The memory used will
37 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
40 * FIO_IO_U_PLAT_NR is the total number of buckets.
44 * Suppose the clat varies from 0 to 999 (usec), the straightforward
45 * method is to keep an array of (999 + 1) buckets, in which a counter
46 * keeps the count of samples which fall in the bucket, e.g.,
47 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
48 * and can be avoided if an approximation is acceptable.
50 * One such method is to let the range of the bucket to be greater
51 * than one. This method has low accuracy when the value is small. For
52 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
53 * the represented value of each bucket be the mean of the range. Then
54 * a value 0 has an round-off error of 49.5. To improve on this, we
55 * use buckets with non-uniform ranges, while bounding the error of
56 * each bucket within a ratio of the sample value. A simple example
57 * would be when error_bound = 0.005, buckets are {
58 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
59 * {[900,909],[910,919]...} }. The total range is partitioned into
60 * groups with different ranges, then buckets with uniform ranges. An
61 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
63 * For better efficiency, we implement this using base two. We group
64 * samples by their Most Significant Bit (MSB), extract the next M bit
65 * of them as an index within the group, and discard the rest of the
68 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
69 * and use M bit for indexing
71 * | n | M bits | bit (n-M-1) ... bit 0 |
73 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
74 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
78 * e <= --------- <= ------- = ---
81 * Furthermore, we use "mean" of the range to represent the bucket,
82 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
83 * as the index, each group must contains 2^M buckets.
85 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
86 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
88 * Group MSB #discarded range of #buckets
90 * ----------------------------------------------------------------
96 * ... ... ... [...,...] ...
97 * 18 23 17 [8838608,+inf]** 64
99 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
100 * the value cannot be rounded off. Use all bits of the sample as
103 * ** If a sample's MSB is greater than 23, it will be counted as 23.
106 #define FIO_IO_U_PLAT_BITS 6
107 #define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
108 #define FIO_IO_U_PLAT_GROUP_NR 19
109 #define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
110 #define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
111 list of percentiles */
113 #define MAX_PATTERN_SIZE 512
114 #define FIO_JOBNAME_SIZE 128
115 #define FIO_VERROR_SIZE 128
118 char name[FIO_JOBNAME_SIZE];
119 char verror[FIO_VERROR_SIZE];
121 uint32_t thread_number;
124 char description[FIO_JOBNAME_SIZE];
128 * bandwidth and latency stats
130 struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
131 struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
132 struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
133 struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
134 struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
137 * fio system usage accounting
145 * IO depth and latency stats
147 uint64_t clat_percentiles;
148 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
150 uint32_t io_u_map[FIO_IO_U_MAP_NR];
151 uint32_t io_u_submit[FIO_IO_U_MAP_NR];
152 uint32_t io_u_complete[FIO_IO_U_MAP_NR];
153 uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
154 uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
155 uint32_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
156 uint64_t total_io_u[3];
157 uint64_t short_io_u[3];
158 uint64_t total_submit;
159 uint64_t total_complete;
161 uint64_t io_bytes[DDIR_RWDIR_CNT];
162 uint64_t runtime[DDIR_RWDIR_CNT];
163 uint64_t total_run_time;
166 * IO Error related stats
168 uint16_t continue_on_error;
169 uint64_t total_err_count;
170 uint32_t first_error;
180 uint32_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
181 uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
182 uint32_t rate[DDIR_RWDIR_CNT];
183 uint32_t iops[DDIR_RWDIR_CNT];
184 uint64_t elapsed_sec;
189 * Network 'copy' of run_str[]
195 extern void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs);
196 extern void show_group_stats(struct group_run_stats *rs);
197 extern int calc_thread_status(struct jobs_eta *je, int force);
198 extern void display_thread_status(struct jobs_eta *je);
199 extern void show_run_stats(void);
200 extern void show_running_run_stats(void);
201 extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr);
202 extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
203 extern void init_thread_stat(struct thread_stat *ts);
204 extern void init_group_run_stat(struct group_run_stats *gs);
205 extern void eta_to_str(char *str, unsigned long eta_sec);
206 extern int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, double *mean, double *dev);
207 extern unsigned int calc_clat_percentiles(unsigned int *io_u_plat, unsigned long nr, fio_fp64_t *plist, unsigned int **output, unsigned int *maxv, unsigned int *minv);
208 extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
209 extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
210 extern void stat_calc_dist(unsigned int *map, unsigned long total, double *io_u_dist);
212 static inline int usec_to_msec(unsigned long *min, unsigned long *max,
213 double *mean, double *dev)
215 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {