[fio.git] / stat.h

#ifndef FIO_STAT_H
#define FIO_STAT_H

#include "iolog.h"

struct group_run_stats {
	uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
	uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
	uint64_t io_kb[DDIR_RWDIR_CNT];
	uint64_t agg[DDIR_RWDIR_CNT];
	uint32_t kb_base;
	uint32_t unit_base;
	uint32_t groupid;
	uint32_t unified_rw_rep;
} __attribute__((packed));

/*
 * How many depth levels to log
 */
#define FIO_IO_U_MAP_NR	7
#define FIO_IO_U_LAT_U_NR 10
#define FIO_IO_U_LAT_M_NR 12

/*
 * Aggregate clat samples to report percentile(s) of them.
 *
 * EXECUTIVE SUMMARY
 *
 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
 * value of resulting percentiles. The error will be approximately
 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
 *
 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
 * range being tracked for latency samples. The maximum value tracked
 * accurately will be 2^(GROUP_NR + PLAT_BITS -1) microseconds.
 *
 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
 * requirement of storing those aggregate counts. The memory used will
 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
 * bytes.
 *
 * FIO_IO_U_PLAT_NR is the total number of buckets.
 *
 * DETAILS
 *
 * Suppose the clat varies from 0 to 999 (usec), the straightforward
 * method is to keep an array of (999 + 1) buckets, in which a counter
 * keeps the count of samples which fall in the bucket, e.g.,
 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
 * and can be avoided if an approximation is acceptable.
 *
 * One such method is to let the range of the bucket to be greater
 * than one. This method has low accuracy when the value is small. For
 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
 * the represented value of each bucket be the mean of the range. Then
 * a value 0 has an round-off error of 49.5. To improve on this, we
 * use buckets with non-uniform ranges, while bounding the error of
 * each bucket within a ratio of the sample value. A simple example
 * would be when error_bound = 0.005, buckets are {
 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
 * {[900,909],[910,919]...}  }. The total range is partitioned into
 * groups with different ranges, then buckets with uniform ranges. An
 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
 *
 * For better efficiency, we implement this using base two. We group
 * samples by their Most Significant Bit (MSB), extract the next M bit
 * of them as an index within the group, and discard the rest of the
 * bits.
 *
 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
 * and use M bit for indexing
 *
 *        | n |    M bits   | bit (n-M-1) ... bit 0 |
 *
 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
 * error
 *
 *           2^(n-M)-1    2^(n-M)    1
 *      e <= --------- <= ------- = ---
 *             2^n          2^n     2^M
 *
 * Furthermore, we use "mean" of the range to represent the bucket,
 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
 * as the index, each group must contains 2^M buckets.
 *
 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
 *      Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
 *
 *	Group	MSB	#discarded	range of		#buckets
 *			error_bits	value
 *	----------------------------------------------------------------
 *	0*	0~5	0		[0,63]			64
 *	1*	6	0		[64,127]		64
 *	2	7	1		[128,255]		64
 *	3	8	2		[256,511]		64
 *	4	9	3		[512,1023]		64
 *	...	...	...		[...,...]		...
 *	18	23	17		[8838608,+inf]**	64
 *
 *  * Special cases: when n < (M-1) or when n == (M-1), in both cases,
 *    the value cannot be rounded off. Use all bits of the sample as
 *    index.
 *
 *  ** If a sample's MSB is greater than 23, it will be counted as 23.
 */

#define FIO_IO_U_PLAT_BITS 6
#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
#define FIO_IO_U_PLAT_GROUP_NR 19
#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
					list of percentiles */

#define MAX_PATTERN_SIZE	512
#define FIO_JOBNAME_SIZE	128
#define FIO_JOBDESC_SIZE	256
#define FIO_VERROR_SIZE		128

struct thread_stat {
	char name[FIO_JOBNAME_SIZE];
	char verror[FIO_VERROR_SIZE];
	uint32_t error;
	uint32_t thread_number;
	uint32_t groupid;
	uint32_t pid;
	char description[FIO_JOBDESC_SIZE];
	uint32_t members;
	uint32_t unified_rw_rep;

	/*
	 * bandwidth and latency stats
	 */
	struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
	struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
	struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
	struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
	struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */

	/*
	 * fio system usage accounting
	 */
	uint64_t usr_time;
	uint64_t sys_time;
	uint64_t ctx;
	uint64_t minf, majf;

	/*
	 * IO depth and latency stats
	 */
	uint64_t clat_percentiles;
	uint64_t percentile_precision;
	fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];

	uint32_t io_u_map[FIO_IO_U_MAP_NR];
	uint32_t io_u_submit[FIO_IO_U_MAP_NR];
	uint32_t io_u_complete[FIO_IO_U_MAP_NR];
	uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
	uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
	uint32_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
	uint32_t pad;

	uint64_t total_io_u[3];
	uint64_t short_io_u[3];
	uint64_t drop_io_u[3];
	uint64_t total_submit;
	uint64_t total_complete;

	uint64_t io_bytes[DDIR_RWDIR_CNT];
	uint64_t runtime[DDIR_RWDIR_CNT];
	uint64_t total_run_time;

	/*
	 * IO Error related stats
	 */
	union {
		uint16_t continue_on_error;
		uint64_t pad2;
	};
	uint64_t total_err_count;
	uint32_t first_error;

	uint32_t kb_base;
	uint32_t unit_base;

	uint32_t latency_depth;
	uint64_t latency_target;
	fio_fp64_t latency_percentile;
	uint64_t latency_window;
} __attribute__((packed));

struct jobs_eta {
	uint32_t nr_running;
	uint32_t nr_ramp;

	uint32_t nr_pending;
	uint32_t nr_setting_up;

	uint32_t files_open;

	uint32_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
	uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
	uint32_t rate[DDIR_RWDIR_CNT];
	uint32_t iops[DDIR_RWDIR_CNT];
	uint64_t elapsed_sec;
	uint64_t eta_sec;
	uint32_t is_pow2;
	uint32_t unit_base;

	/*
	 * Network 'copy' of run_str[]
	 */
	uint32_t nr_threads;
	uint8_t run_str[];
} __attribute__((packed));

extern struct fio_mutex *stat_mutex;

extern struct jobs_eta *get_jobs_eta(int force, size_t *size);

extern void stat_init(void);
extern void stat_exit(void);

extern struct json_object * show_thread_status(struct thread_stat *ts, struct group_run_stats *rs);
extern void show_group_stats(struct group_run_stats *rs);
extern int calc_thread_status(struct jobs_eta *je, int force);
extern void display_thread_status(struct jobs_eta *je);
extern void show_run_stats(void);
extern void __show_run_stats(void);
extern void __show_running_run_stats(void);
extern void show_running_run_stats(void);
extern void check_for_running_stats(void);
extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, int nr);
extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
extern void init_thread_stat(struct thread_stat *ts);
extern void init_group_run_stat(struct group_run_stats *gs);
extern void eta_to_str(char *str, unsigned long eta_sec);
extern int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, double *mean, double *dev);
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);
extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
extern void stat_calc_dist(unsigned int *map, unsigned long total, double *io_u_dist);
extern void reset_io_stats(struct thread_data *);

static inline int usec_to_msec(unsigned long *min, unsigned long *max,
			       double *mean, double *dev)
{
	if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
		*min /= 1000;
		*max /= 1000;
		*mean /= 1000.0;
		*dev /= 1000.0;
		return 0;
	}

	return 1;
}
/*
 * Worst level condensing would be 1:5, so allow enough room for that
 */
#define __THREAD_RUNSTR_SZ(nr)	((nr) * 5)
#define THREAD_RUNSTR_SZ	__THREAD_RUNSTR_SZ(thread_number)

#endif
Commit	Line	Data
	1	#ifndef FIO_STAT_H
	2	#define FIO_STAT_H
	3
	4	#include "iolog.h"
	5
	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];
	11	uint32_t kb_base;
	12	uint32_t unit_base;
	13	uint32_t groupid;
	14	uint32_t unified_rw_rep;
	15	} __attribute__((packed));
	16
	17	/*
	18	* How many depth levels to log
	19	*/
	20	#define FIO_IO_U_MAP_NR 7
	21	#define FIO_IO_U_LAT_U_NR 10
	22	#define FIO_IO_U_LAT_M_NR 12
	23
	24	/*
	25	* Aggregate clat samples to report percentile(s) of them.
	26	*
	27	* EXECUTIVE SUMMARY
	28	*
	29	* FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
	30	* value of resulting percentiles. The error will be approximately
	31	* 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
	32	*
	33	* FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
	34	* range being tracked for latency samples. The maximum value tracked
	35	* accurately will be 2^(GROUP_NR + PLAT_BITS -1) microseconds.
	36	*
	37	* FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
	38	* requirement of storing those aggregate counts. The memory used will
	39	* be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
	40	* bytes.
	41	*
	42	* FIO_IO_U_PLAT_NR is the total number of buckets.
	43	*
	44	* DETAILS
	45	*
	46	* Suppose the clat varies from 0 to 999 (usec), the straightforward
	47	* method is to keep an array of (999 + 1) buckets, in which a counter
	48	* keeps the count of samples which fall in the bucket, e.g.,
	49	* {[0],[1],...,[999]}. However this consumes a huge amount of space,
	50	* and can be avoided if an approximation is acceptable.
	51	*
	52	* One such method is to let the range of the bucket to be greater
	53	* than one. This method has low accuracy when the value is small. For
	54	* example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
	55	* the represented value of each bucket be the mean of the range. Then
	56	* a value 0 has an round-off error of 49.5. To improve on this, we
	57	* use buckets with non-uniform ranges, while bounding the error of
	58	* each bucket within a ratio of the sample value. A simple example
	59	* would be when error_bound = 0.005, buckets are {
	60	* {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
	61	* {[900,909],[910,919]...} }. The total range is partitioned into
	62	* groups with different ranges, then buckets with uniform ranges. An
	63	* upper bound of the error is (range_of_bucket/2)/value_of_bucket
	64	*
	65	* For better efficiency, we implement this using base two. We group
	66	* samples by their Most Significant Bit (MSB), extract the next M bit
	67	* of them as an index within the group, and discard the rest of the
	68	* bits.
	69	*
	70	* E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
	71	* and use M bit for indexing
	72	*
	73	* \| n \| M bits \| bit (n-M-1) ... bit 0 \|
	74	*
	75	* Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
	76	* (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
	77	* error
	78	*
	79	* 2^(n-M)-1 2^(n-M) 1
	80	* e <= --------- <= ------- = ---
	81	* 2^n 2^n 2^M
	82	*
	83	* Furthermore, we use "mean" of the range to represent the bucket,
	84	* the error e can be lowered by half to 1 / 2^(M+1). By using M bits
	85	* as the index, each group must contains 2^M buckets.
	86	*
	87	* E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
	88	* Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
	89	*
	90	* Group MSB #discarded range of #buckets
	91	* error_bits value
	92	* ----------------------------------------------------------------
	93	* 0* 0~5 0 [0,63] 64
	94	* 1* 6 0 [64,127] 64
	95	* 2 7 1 [128,255] 64
	96	* 3 8 2 [256,511] 64
	97	* 4 9 3 [512,1023] 64
	98	* ... ... ... [...,...] ...
	99	* 18 23 17 [8838608,+inf]** 64
	100	*
	101	* * Special cases: when n < (M-1) or when n == (M-1), in both cases,
	102	* the value cannot be rounded off. Use all bits of the sample as
	103	* index.
	104	*
	105	* ** If a sample's MSB is greater than 23, it will be counted as 23.
	106	*/
	107
	108	#define FIO_IO_U_PLAT_BITS 6
	109	#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
	110	#define FIO_IO_U_PLAT_GROUP_NR 19
	111	#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
	112	#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
	113	list of percentiles */
	114
	115	#define MAX_PATTERN_SIZE 512
	116	#define FIO_JOBNAME_SIZE 128
	117	#define FIO_JOBDESC_SIZE 256
	118	#define FIO_VERROR_SIZE 128
	119
	120	struct thread_stat {
	121	char name[FIO_JOBNAME_SIZE];
	122	char verror[FIO_VERROR_SIZE];
	123	uint32_t error;
	124	uint32_t thread_number;
	125	uint32_t groupid;
	126	uint32_t pid;
	127	char description[FIO_JOBDESC_SIZE];
	128	uint32_t members;
	129	uint32_t unified_rw_rep;
	130
	131	/*
	132	* bandwidth and latency stats
	133	*/
	134	struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
	135	struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
	136	struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
	137	struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
	138	struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
	139
	140	/*
	141	* fio system usage accounting
	142	*/
	143	uint64_t usr_time;
	144	uint64_t sys_time;
	145	uint64_t ctx;
	146	uint64_t minf, majf;
	147
	148	/*
	149	* IO depth and latency stats
	150	*/
	151	uint64_t clat_percentiles;
	152	uint64_t percentile_precision;
	153	fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
	154
	155	uint32_t io_u_map[FIO_IO_U_MAP_NR];
	156	uint32_t io_u_submit[FIO_IO_U_MAP_NR];
	157	uint32_t io_u_complete[FIO_IO_U_MAP_NR];
	158	uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
	159	uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
	160	uint32_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
	161	uint32_t pad;
	162
	163	uint64_t total_io_u[3];
	164	uint64_t short_io_u[3];
	165	uint64_t drop_io_u[3];
	166	uint64_t total_submit;
	167	uint64_t total_complete;
	168
	169	uint64_t io_bytes[DDIR_RWDIR_CNT];
	170	uint64_t runtime[DDIR_RWDIR_CNT];
	171	uint64_t total_run_time;
	172
	173	/*
	174	* IO Error related stats
	175	*/
	176	union {
	177	uint16_t continue_on_error;
	178	uint64_t pad2;
	179	};
	180	uint64_t total_err_count;
	181	uint32_t first_error;
	182
	183	uint32_t kb_base;
	184	uint32_t unit_base;
	185
	186	uint32_t latency_depth;
	187	uint64_t latency_target;
	188	fio_fp64_t latency_percentile;
	189	uint64_t latency_window;
	190	} __attribute__((packed));
	191
	192	struct jobs_eta {
	193	uint32_t nr_running;
	194	uint32_t nr_ramp;
	195
	196	uint32_t nr_pending;
	197	uint32_t nr_setting_up;
	198
	199	uint32_t files_open;
	200
	201	uint32_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
	202	uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
	203	uint32_t rate[DDIR_RWDIR_CNT];
	204	uint32_t iops[DDIR_RWDIR_CNT];
	205	uint64_t elapsed_sec;
	206	uint64_t eta_sec;
	207	uint32_t is_pow2;
	208	uint32_t unit_base;
	209
	210	/*
	211	* Network 'copy' of run_str[]
	212	*/
	213	uint32_t nr_threads;
	214	uint8_t run_str[];
	215	} __attribute__((packed));
	216
	217	extern struct fio_mutex *stat_mutex;
	218
	219	extern struct jobs_eta get_jobs_eta(int force, size_t size);
	220
	221	extern void stat_init(void);
	222	extern void stat_exit(void);
	223
	224	extern struct json_object * show_thread_status(struct thread_stat ts, struct group_run_stats rs);
	225	extern void show_group_stats(struct group_run_stats *rs);
	226	extern int calc_thread_status(struct jobs_eta *je, int force);
	227	extern void display_thread_status(struct jobs_eta *je);
	228	extern void show_run_stats(void);
	229	extern void __show_run_stats(void);
	230	extern void __show_running_run_stats(void);
	231	extern void show_running_run_stats(void);
	232	extern void check_for_running_stats(void);
	233	extern void sum_thread_stats(struct thread_stat dst, struct thread_stat src, int nr);
	234	extern void sum_group_stats(struct group_run_stats dst, struct group_run_stats src);
	235	extern void init_thread_stat(struct thread_stat *ts);
	236	extern void init_group_run_stat(struct group_run_stats *gs);
	237	extern void eta_to_str(char *str, unsigned long eta_sec);
	238	extern int calc_lat(struct io_stat is, unsigned long min, unsigned long max, double mean, double *dev);
	239	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);
	240	extern void stat_calc_lat_m(struct thread_stat ts, double io_u_lat);
	241	extern void stat_calc_lat_u(struct thread_stat ts, double io_u_lat);
	242	extern void stat_calc_dist(unsigned int map, unsigned long total, double io_u_dist);
	243	extern void reset_io_stats(struct thread_data *);
	244
	245	static inline int usec_to_msec(unsigned long min, unsigned long max,
	246	double mean, double dev)
	247	{
	248	if (min > 1000 && max > 1000 && mean > 1000.0 && dev > 1000.0) {
	249	*min /= 1000;
	250	*max /= 1000;
	251	*mean /= 1000.0;
	252	*dev /= 1000.0;
	253	return 0;
	254	}
	255
	256	return 1;
	257	}
	258	/*
	259	* Worst level condensing would be 1:5, so allow enough room for that
	260	*/
	261	#define __THREAD_RUNSTR_SZ(nr) ((nr) * 5)
	262	#define THREAD_RUNSTR_SZ __THREAD_RUNSTR_SZ(thread_number)
	263
	264	#endif