Add support for zones of random IO, with varying frequency of access

Let's say you want to ensure that 50% of the IO falls in the first 5%
of the flie, with the remaining 50% over the last 95%. Now that's
possible with random_distribution=zoned. For this particular case,
you would do:

random_distribution=zoned:50/5:50/95

Up to 64 ranges can be specified, and multiple data directions can
be given as well. The above would apply to both reads, writes, and
trims. If you wanted to have 50% of the writes fall in the first
10%, 25% in the next 10%, and the last 25% over the remaining 80%,
you could extend the above ala:

random_distribution=zoned:50/5:50/95,50/10:25/10:25/80

Signed-off-by: Jens Axboe <[email protected]>

HOWTO

@@ -964,6 +964,7 @@ random_distribution=str:float	By default, fio will use a completely uniform
 		zipf		Zipf distribution
 		pareto		Pareto distribution
 		gauss		Normal (guassian) distribution
+		zoned		Zoned random distribution
 
 		When using a zipf or pareto distribution, an input value
 		is also needed to define the access pattern. For zipf, this
@@ -976,6 +977,25 @@ random_distribution=str:float	By default, fio will use a completely uniform
 		the gauss distribution, a normal deviation is supplied as
 		a value between 0 and 100.
 
+		For a zoned distribution, fio supports specifying percentages
+		of IO access that should fall within what range of the file or
+		device. For example, given a criteria of:
+
+			60% of accesses should be to the first 10%
+			30% of accesses should be to the next 20%
+			8% of accesses should be to to the next 30%
+			2% of accesses should be to the next 40%
+
+		we can define that through zoning of the random accesses. For
+		the above example, the user would do:
+
+			random_distribution=zoned:60/10:30/20:8/30:2/40
+
+		similarly to how bssplit works for setting ranges and
+		percentages of block sizes. Like bssplit, it's possible to
+		specify separate zones for reads, writes, and trims. If just
+		one set is given, it'll apply to all of them.
+
 percentage_random=int	For a random workload, set how big a percentage should
 		be random. This defaults to 100%, in which case the workload
 		is fully random. It can be set from anywhere from 0 to 100.

cconv.c

@@ -23,6 +23,8 @@ static void __string_to_net(uint8_t *dst, const char *src, size_t dst_size)
 
 static void free_thread_options_to_cpu(struct thread_options *o)
 {
+	int i;
+
 	free(o->description);
 	free(o->name);
 	free(o->wait_for);
@@ -43,6 +45,11 @@ static void free_thread_options_to_cpu(struct thread_options *o)
 	free(o->ioscheduler);
 	free(o->profile);
 	free(o->cgroup);
+
+	for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+		free(o->bssplit[i]);
+		free(o->zone_split[i]);
+	}
 }
 
 void convert_thread_options_to_cpu(struct thread_options *o,
@@ -111,6 +118,16 @@ void convert_thread_options_to_cpu(struct thread_options *o,
 			}
 		}
 
+		o->zone_split_nr[i] = le32_to_cpu(top->zone_split_nr[i]);
+
+		if (o->zone_split_nr[i]) {
+			o->zone_split[i] = malloc(o->zone_split_nr[i] * sizeof(struct zone_split));
+			for (j = 0; j < o->zone_split_nr[i]; j++) {
+				o->zone_split[i][j].access_perc = top->zone_split[i][j].access_perc;
+				o->zone_split[i][j].size_perc = top->zone_split[i][j].size_perc;
+			}
+		}
+
 		o->rwmix[i] = le32_to_cpu(top->rwmix[i]);
 		o->rate[i] = le32_to_cpu(top->rate[i]);
 		o->ratemin[i] = le32_to_cpu(top->ratemin[i]);
@@ -453,6 +470,21 @@ void convert_thread_options_to_net(struct thread_options_pack *top,
 			}
 		}
 
+		top->zone_split_nr[i] = cpu_to_le32(o->zone_split_nr[i]);
+
+		if (o->zone_split_nr[i]) {
+			unsigned int zone_split_nr = o->zone_split_nr[i];
+
+			if (zone_split_nr > ZONESPLIT_MAX) {
+				log_err("fio: ZONESPLIT_MAX is too small\n");
+				zone_split_nr = ZONESPLIT_MAX;
+			}
+			for (j = 0; j < zone_split_nr; j++) {
+				top->zone_split[i][j].access_perc = o->zone_split[i][j].access_perc;
+				top->zone_split[i][j].size_perc = o->zone_split[i][j].size_perc;
+			}
+		}
+
 		top->rwmix[i] = cpu_to_le32(o->rwmix[i]);
 		top->rate[i] = cpu_to_le32(o->rate[i]);
 		top->ratemin[i] = cpu_to_le32(o->ratemin[i]);

examples/rand-zones.fio

@@ -0,0 +1,18 @@
+# Sample job file demonstrating how to use zoned random distributionss
+# to have skewed random accesses. This example has 50% of the accesses
+# to the first 5% of the file (50/5), 30% to the next 15% (30/15), and
+# finally 20% of the IO will end up in the remaining 80%.
+[zones]
+size=2g
+direct=1
+bs=4k
+rw=randread
+norandommap
+random_distribution=zoned:50/5:30/15:20/
+
+# The above applies to all of reads/writes/trims. If we wanted to do
+# something differently for writes, let's say 50% for the first 10%
+# and 50% for the remaining 90%, we could do it by adding a new section
+# after a a comma.
+
+# random_distribution=zoned:50/5:30/15:20/,50/10:50/90

fio.1

@@ -877,8 +877,10 @@ Pareto distribution
 .B gauss
 Normal (gaussian) distribution
 .TP
+.B zoned
+Zoned random distribution
+.TP
 .RE
-.P
 When using a \fBzipf\fR or \fBpareto\fR distribution, an input value is also
 needed to define the access pattern. For \fBzipf\fR, this is the zipf theta.
 For \fBpareto\fR, it's the pareto power. Fio includes a test program, genzipf,
@@ -887,6 +889,36 @@ hit rates. If you wanted to use \fBzipf\fR with a theta of 1.2, you would use
 random_distribution=zipf:1.2 as the option. If a non-uniform model is used,
 fio will disable use of the random map. For the \fBgauss\fR distribution, a
 normal deviation is supplied as a value between 0 and 100.
+.P
+.RS
+For a \fBzoned\fR distribution, fio supports specifying percentages of IO
+access that should fall within what range of the file or device. For example,
+given a criteria of:
+.P
+.RS
+60% of accesses should be to the first 10%
+.RE
+.RS
+30% of accesses should be to the next 20%
+.RE
+.RS
+8% of accesses should be to to the next 30%
+.RE
+.RS
+2% of accesses should be to the next 40%
+.RE
+.P
+we can define that through zoning of the random accesses. For the above
+example, the user would do:
+.P
+.RS
+.B random_distribution=zoned:60/10:30/20:8/30:2/40
+.RE
+.P
+similarly to how \fBbssplit\fR works for setting ranges and percentages of block
+sizes. Like \fBbssplit\fR, it's possible to specify separate zones for reads,
+writes, and trims. If just one set is given, it'll apply to all of them.
+.RE
 .TP
 .BI percentage_random \fR=\fPint
 For a random workload, set how big a percentage should be random. This defaults

fio.h

@@ -96,6 +96,7 @@ enum {
 	FIO_RAND_START_DELAY,
 	FIO_DEDUPE_OFF,
 	FIO_RAND_POISSON_OFF,
+	FIO_RAND_ZONE_OFF,
 	FIO_RAND_NR_OFFS,
 };
 
@@ -115,6 +116,11 @@ struct sk_out;
 void sk_out_assign(struct sk_out *);
 void sk_out_drop(void);
 
+struct zone_split_index {
+	uint8_t size_perc;
+	uint8_t size_perc_prev;
+};
+
 /*
  * This describes a single thread/process executing a fio job.
  */
@@ -200,6 +206,9 @@ struct thread_data {
 	struct frand_state buf_state;
 	struct frand_state buf_state_prev;
 	struct frand_state dedupe_state;
+	struct frand_state zone_state;
+
+	struct zone_split_index **zone_state_index;
 
 	unsigned int verify_batch;
 	unsigned int trim_batch;
@@ -712,6 +721,7 @@ enum {
 	FIO_RAND_DIST_ZIPF,
 	FIO_RAND_DIST_PARETO,
 	FIO_RAND_DIST_GAUSS,
+	FIO_RAND_DIST_ZONED,
 };
 
 #define FIO_DEF_ZIPF		1.1

init.c

@@ -968,6 +968,7 @@ void td_fill_rand_seeds(struct thread_data *td)
 	frand_copy(&td->buf_state_prev, &td->buf_state);
 
 	init_rand_seed(&td->dedupe_state, td->rand_seeds[FIO_DEDUPE_OFF], use64);
+	init_rand_seed(&td->zone_state, td->rand_seeds[FIO_RAND_ZONE_OFF], use64);
 }
 
 /*

io_u.c

@@ -86,24 +86,19 @@ struct rand_off {
 };
 
 static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f,
-				  enum fio_ddir ddir, uint64_t *b)
+				  enum fio_ddir ddir, uint64_t *b,
+				  uint64_t lastb)
 {
 	uint64_t r;
 
 	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE ||
 	    td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) {
-		uint64_t frand_max, lastb;
 
-		lastb = last_block(td, f, ddir);
-		if (!lastb)
-			return 1;
-
-		frand_max = rand_max(&td->random_state);
 		r = __rand(&td->random_state);
 
 		dprint(FD_RANDOM, "off rand %llu\n", (unsigned long long) r);
 
-		*b = lastb * (r / ((uint64_t) frand_max + 1.0));
+		*b = lastb * (r / (rand_max(&td->random_state) + 1.0));
 	} else {
 		uint64_t off = 0;
 
@@ -161,6 +156,70 @@ static int __get_next_rand_offset_gauss(struct thread_data *td,
 	return 0;
 }
 
+static int __get_next_rand_offset_zoned(struct thread_data *td,
+					struct fio_file *f, enum fio_ddir ddir,
+					uint64_t *b)
+{
+	unsigned int v, send, stotal;
+	uint64_t offset, lastb;
+	static int warned;
+	struct zone_split_index *zsi;
+
+	lastb = last_block(td, f, ddir);
+	if (!lastb)
+		return 1;
+
+	if (!td->o.zone_split_nr[ddir]) {
+bail:
+		return __get_next_rand_offset(td, f, ddir, b, lastb);
+	}
+
+	/*
+	 * Generate a value, v, between 1 and 100, both inclusive
+	 */
+	v = rand_between(&td->zone_state, 1, 100);
+
+	zsi = &td->zone_state_index[ddir][v - 1];
+	stotal = zsi->size_perc_prev;
+	send = zsi->size_perc;
+
+	/*
+	 * Should never happen
+	 */
+	if (send == -1U) {
+		if (!warned) {
+			log_err("fio: bug in zoned generation\n");
+			warned = 1;
+		}
+		goto bail;
+	}
+
+	/*
+	 * 'send' is some percentage below or equal to 100 that
+	 * marks the end of the current IO range. 'stotal' marks
+	 * the start, in percent.
+	 */
+	if (stotal)
+		offset = stotal * lastb / 100ULL;
+	else
+		offset = 0;
+
+	lastb = lastb * (send - stotal) / 100ULL;
+
+	/*
+	 * Generate index from 0..send-of-lastb
+	 */
+	if (__get_next_rand_offset(td, f, ddir, b, lastb) == 1)
+		return 1;
+
+	/*
+	 * Add our start offset, if any
+	 */
+	if (offset)
+		*b += offset;
+
+	return 0;
+}
 
 static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b)
 {
@@ -173,14 +232,22 @@ static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b)
 static int get_off_from_method(struct thread_data *td, struct fio_file *f,
 			       enum fio_ddir ddir, uint64_t *b)
 {
-	if (td->o.random_distribution == FIO_RAND_DIST_RANDOM)
-		return __get_next_rand_offset(td, f, ddir, b);
-	else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
+	if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) {
+		uint64_t lastb;
+
+		lastb = last_block(td, f, ddir);
+		if (!lastb)
+			return 1;
+
+		return __get_next_rand_offset(td, f, ddir, b, lastb);
+	} else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
 		return __get_next_rand_offset_zipf(td, f, ddir, b);
 	else if (td->o.random_distribution == FIO_RAND_DIST_PARETO)
 		return __get_next_rand_offset_pareto(td, f, ddir, b);
 	else if (td->o.random_distribution == FIO_RAND_DIST_GAUSS)
 		return __get_next_rand_offset_gauss(td, f, ddir, b);
+	else if (td->o.random_distribution == FIO_RAND_DIST_ZONED)
+		return __get_next_rand_offset_zoned(td, f, ddir, b);
 
 	log_err("fio: unknown random distribution: %d\n", td->o.random_distribution);
 	return 1;
@@ -207,16 +274,12 @@ static inline bool should_sort_io(struct thread_data *td)
 
 static bool should_do_random(struct thread_data *td, enum fio_ddir ddir)
 {
-	uint64_t frand_max;
 	unsigned int v;
-	unsigned long r;
 
 	if (td->o.perc_rand[ddir] == 100)
 		return true;
 
-	frand_max = rand_max(&td->seq_rand_state[ddir]);
-	r = __rand(&td->seq_rand_state[ddir]);
-	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
+	v = rand_between(&td->seq_rand_state[ddir], 1, 100);
 
 	return v <= td->o.perc_rand[ddir];
 }
@@ -536,12 +599,9 @@ static void set_rwmix_bytes(struct thread_data *td)
 
 static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 {
-	uint64_t frand_max = rand_max(&td->rwmix_state);
 	unsigned int v;
-	unsigned long r;
 
-	r = __rand(&td->rwmix_state);
-	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
+	v = rand_between(&td->rwmix_state, 1, 100);
 
 	if (v <= td->o.rwmix[DDIR_READ])
 		return DDIR_READ;
@@ -1895,9 +1955,7 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
  */
 static struct frand_state *get_buf_state(struct thread_data *td)
 {
-	uint64_t frand_max;
 	unsigned int v;
-	unsigned long r;
 
 	if (!td->o.dedupe_percentage)
 		return &td->buf_state;
@@ -1906,9 +1964,7 @@ static struct frand_state *get_buf_state(struct thread_data *td)
 		return &td->buf_state;
 	}
 
-	frand_max = rand_max(&td->dedupe_state);
-	r = __rand(&td->dedupe_state);
-	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
+	v = rand_between(&td->dedupe_state, 1, 100);
 
 	if (v <= td->o.dedupe_percentage)
 		return &td->buf_state_prev;