4 struct group_run_stats {
5 uint64_t max_run[2], min_run[2];
6 uint64_t max_bw[2], min_bw[2];
14 * How many depth levels to log
16 #define FIO_IO_U_MAP_NR 7
17 #define FIO_IO_U_LAT_U_NR 10
18 #define FIO_IO_U_LAT_M_NR 12
21 * Aggregate clat samples to report percentile(s) of them.
25 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
26 * value of resulting percentiles. The error will be approximately
27 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
29 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
30 * range being tracked for latency samples. The maximum value tracked
31 * accurately will be 2^(GROUP_NR + PLAT_BITS -1) microseconds.
33 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
34 * requirement of storing those aggregate counts. The memory used will
35 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
38 * FIO_IO_U_PLAT_NR is the total number of buckets.
42 * Suppose the clat varies from 0 to 999 (usec), the straightforward
43 * method is to keep an array of (999 + 1) buckets, in which a counter
44 * keeps the count of samples which fall in the bucket, e.g.,
45 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
46 * and can be avoided if an approximation is acceptable.
48 * One such method is to let the range of the bucket to be greater
49 * than one. This method has low accuracy when the value is small. For
50 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
51 * the represented value of each bucket be the mean of the range. Then
52 * a value 0 has an round-off error of 49.5. To improve on this, we
53 * use buckets with non-uniform ranges, while bounding the error of
54 * each bucket within a ratio of the sample value. A simple example
55 * would be when error_bound = 0.005, buckets are {
56 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
57 * {[900,909],[910,919]...} }. The total range is partitioned into
58 * groups with different ranges, then buckets with uniform ranges. An
59 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
61 * For better efficiency, we implement this using base two. We group
62 * samples by their Most Significant Bit (MSB), extract the next M bit
63 * of them as an index within the group, and discard the rest of the
66 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
67 * and use M bit for indexing
69 * | n | M bits | bit (n-M-1) ... bit 0 |
71 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
72 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
76 * e <= --------- <= ------- = ---
79 * Furthermore, we use "mean" of the range to represent the bucket,
80 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
81 * as the index, each group must contains 2^M buckets.
83 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
84 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
86 * Group MSB #discarded range of #buckets
88 * ----------------------------------------------------------------
94 * ... ... ... [...,...] ...
95 * 18 23 17 [8838608,+inf]** 64
97 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
98 * the value cannot be rounded off. Use all bits of the sample as
101 * ** If a sample's MSB is greater than 23, it will be counted as 23.
104 #define FIO_IO_U_PLAT_BITS 6
105 #define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
106 #define FIO_IO_U_PLAT_GROUP_NR 19
107 #define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
108 #define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
109 list of percentiles */
111 #define MAX_PATTERN_SIZE 512
112 #define FIO_JOBNAME_SIZE 128
113 #define FIO_VERROR_SIZE 128
116 char name[FIO_JOBNAME_SIZE];
117 char verror[FIO_VERROR_SIZE];
121 char description[FIO_JOBNAME_SIZE];
125 * bandwidth and latency stats
127 struct io_stat clat_stat[2]; /* completion latency */
128 struct io_stat slat_stat[2]; /* submission latency */
129 struct io_stat lat_stat[2]; /* total latency */
130 struct io_stat bw_stat[2]; /* bandwidth stats */
131 struct io_stat iops_stat[2]; /* IOPS stats */
134 * fio system usage accounting
142 * IO depth and latency stats
144 uint64_t clat_percentiles;
145 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
147 uint32_t io_u_map[FIO_IO_U_MAP_NR];
148 uint32_t io_u_submit[FIO_IO_U_MAP_NR];
149 uint32_t io_u_complete[FIO_IO_U_MAP_NR];
150 uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
151 uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
152 uint32_t io_u_plat[2][FIO_IO_U_PLAT_NR];
153 uint64_t total_io_u[3];
154 uint64_t short_io_u[3];
155 uint64_t total_submit;
156 uint64_t total_complete;
158 uint64_t io_bytes[2];
160 uint64_t total_run_time;
163 * IO Error related stats
165 uint16_t continue_on_error;
166 uint64_t total_err_count;
167 uint32_t first_error;
177 uint32_t m_rate, t_rate;
178 uint32_t m_iops, t_iops;
181 uint64_t elapsed_sec;
185 * Network 'copy' of run_str[]
191 extern void show_thread_status(struct thread_stat *ts, struct group_run_stats *rs);
192 extern void show_group_stats(struct group_run_stats *rs);
193 extern int calc_thread_status(struct jobs_eta *je, int force);
194 extern void display_thread_status(struct jobs_eta *je);
195 extern void show_run_stats(void);
196 extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr);
197 extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
198 extern void init_thread_stat(struct thread_stat *ts);
199 extern void init_group_run_stat(struct group_run_stats *gs);