lib/ftl: consider 3D TLC NAND read unit size

For the latest TLC NAND, one write buffer unit (rwb batch)
needs to be spread over three PUs instead of being allocated
to a single PU for better sequential read performance
since the optimal write size(ws_opt) of 3D TLC NAND is
3 times bigger than the optimal read size(rs_opt).

I added num_interleave_units in 'struct spdk_ftl_conf'
to configure the number of interleaving units per ws_opt.
If num_interleave_units is set as 1, the whole of the ws_opt
blocks are placed sequentially around each PU.
If num_interleave_units is set as N, the 1/N of the ws_opt
blocks are staggered. So consecutively numbered blocks
are separated by ws_opt / num_interleave_units.

The sequential read performance is improved from 1.9GiB/s
up to 2.97GiB/S with this patch on our system. No performance
degradation is observed on sequential writes or
4KB random reads/writes.

Please refer to the Trello card for more details.
https://trello.com/c/Osol93ZU

Change-Id: I371e72067b278ef43c3ac87a3d9ce9010d3fcb15
Signed-off-by: Claire J. In <claire.in@circuitblvd.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/450976
Reviewed-by: Young Tack Jin <youngtack.jin@circuitblvd.com>
Reviewed-by: Konrad Sztyber <konrad.sztyber@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Wojciech Malikowski <wojciech.malikowski@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>

include/spdk/ftl.h

@@ -86,6 +86,9 @@ struct spdk_ftl_conf {
 		/* User writes limits */
 		struct spdk_ftl_limit		limits[SPDK_FTL_LIMIT_MAX];
 	} defrag;
+
+	/* Number of interleaving units per ws_opt */
+	size_t                                  num_interleave_units;
 };
 
 /* Range of parallel units (inclusive) */

lib/ftl/ftl_core.c

@@ -609,17 +609,23 @@ ftl_process_shutdown(struct spdk_ftl_dev *dev)
 {
 	size_t size = ftl_rwb_num_acquired(dev->rwb, FTL_RWB_TYPE_INTERNAL) +
 		      ftl_rwb_num_acquired(dev->rwb, FTL_RWB_TYPE_USER);
+	size_t num_active = dev->xfer_size * ftl_rwb_get_active_batches(dev->rwb);
 
-	if (size >= dev->xfer_size) {
+	num_active = num_active ? num_active : dev->xfer_size;
+	if (size >= num_active) {
 		return;
 	}
 
 	/* If we reach this point we need to remove free bands */
 	/* and pad current wptr band to the end */
+	if (ftl_rwb_get_active_batches(dev->rwb) <= 1) {
 		ftl_remove_free_bands(dev);
+	}
 
 	/* Pad write buffer until band is full */
-	ftl_rwb_pad(dev, dev->xfer_size - size);
+	/* TODO : It would be better to request padding to as many as PUs possible */
+	/* instead of requesting to one PU at a time */
+	ftl_rwb_pad(dev, num_active - size);
 }
 
 static int
@@ -1268,7 +1274,7 @@ static void
 ftl_flush_pad_batch(struct spdk_ftl_dev *dev)
 {
 	struct ftl_rwb *rwb = dev->rwb;
-	size_t size;
+	size_t size, num_entries;
 
 	size = ftl_rwb_num_acquired(rwb, FTL_RWB_TYPE_INTERNAL) +
 	       ftl_rwb_num_acquired(rwb, FTL_RWB_TYPE_USER);
@@ -1280,8 +1286,9 @@ ftl_flush_pad_batch(struct spdk_ftl_dev *dev)
 	/* Only add padding when there's less than xfer size */
 	/* entries in the buffer. Otherwise we just have to wait */
 	/* for the entries to become ready. */
-	if (size < dev->xfer_size) {
+	num_entries = ftl_rwb_get_active_batches(dev->rwb) * dev->xfer_size;
-		ftl_rwb_pad(dev, dev->xfer_size - (size % dev->xfer_size));
+	if (size < num_entries) {
+		ftl_rwb_pad(dev, num_entries - (size % num_entries));
 	}
 }
 

lib/ftl/ftl_init.c

@@ -87,6 +87,9 @@ static const struct spdk_ftl_conf	g_default_conf = {
 	.max_active_relocs = 3,
 	/* IO pool size per user thread (this should be adjusted to thread IO qdepth) */
 	.user_io_pool_size = 2048,
+	/* Number of interleaving units per ws_opt */
+	/* 1 for default and 3 for 3D TLC NAND */
+	.num_interleave_units = 1,
 };
 
 static void ftl_dev_free_sync(struct spdk_ftl_dev *dev);
@@ -116,7 +119,8 @@ ftl_band_init_md(struct ftl_band *band)
 }
 
 static int
-ftl_check_conf(const struct spdk_ftl_conf *conf)
+ftl_check_conf(const struct spdk_ftl_conf *conf,
+	       const struct spdk_ocssd_geometry_data *geo)
 {
 	size_t i;
 
@@ -135,6 +139,9 @@ ftl_check_conf(const struct spdk_ftl_conf *conf)
 	if (conf->rwb_size % FTL_BLOCK_SIZE != 0) {
 		return -1;
 	}
+	if (geo->ws_opt % conf->num_interleave_units != 0) {
+		return -1;
+	}
 
 	for (i = 0; i < SPDK_FTL_LIMIT_MAX; ++i) {
 		if (conf->defrag.limits[i].limit > 100) {
@@ -157,7 +164,7 @@ ftl_check_init_opts(const struct spdk_ftl_dev_init_opts *opts,
 		return -1;
 	}
 
-	if (ftl_check_conf(opts->conf)) {
+	if (ftl_check_conf(opts->conf, geo)) {
 		return -1;
 	}
 
@@ -952,7 +959,7 @@ spdk_ftl_dev_init(const struct spdk_ftl_dev_init_opts *_opts, spdk_ftl_init_fn c
 		goto fail_sync;
 	}
 
-	dev->rwb = ftl_rwb_init(&dev->conf, dev->geo.ws_opt, dev->md_size);
+	dev->rwb = ftl_rwb_init(&dev->conf, dev->geo.ws_opt, dev->md_size, ftl_dev_num_punits(dev));
 	if (!dev->rwb) {
 		SPDK_ERRLOG("Unable to initialize rwb structures\n");
 		goto fail_sync;
@@ -1095,6 +1102,8 @@ spdk_ftl_dev_free(struct spdk_ftl_dev *dev, spdk_ftl_fn cb, void *cb_arg)
 	dev->halt_arg = cb_arg;
 	dev->halt = 1;
 
+	ftl_rwb_disable_interleaving(dev->rwb);
+
 	spdk_thread_send_msg(ftl_get_core_thread(dev), ftl_add_halt_poller, dev);
 	return 0;
 }

lib/ftl/ftl_rwb.c

@@ -70,6 +70,12 @@ struct ftl_rwb_batch {
 struct ftl_rwb {
 	/* Number of batches */
 	size_t					num_batches;
+
+	/* Information for interleaving */
+	size_t                                  interleave_offset;
+	/* Maximum number of active batches */
+	size_t                                  max_active_batches;
+
 	/* Number of entries per batch */
 	size_t					xfer_size;
 	/* Metadata's size */
@@ -80,11 +86,16 @@ struct ftl_rwb {
 	/* User/internal limits */
 	size_t					limits[FTL_RWB_TYPE_MAX];
 
-	/* Current batch */
+	/* Active batch queue */
-	struct ftl_rwb_batch			*current;
+	STAILQ_HEAD(, ftl_rwb_batch)            active_queue;
+	/* Number of active batches */
+	unsigned int                            num_active_batches;
 
 	/* Free batch queue */
 	STAILQ_HEAD(, ftl_rwb_batch)		free_queue;
+	/* Number of active batches */
+	unsigned int                            num_free_batches;
+
 	/* Submission batch queue */
 	struct spdk_ring			*submit_queue;
 	/* High-priority batch queue */
@@ -174,9 +185,9 @@ ftl_rwb_batch_init(struct ftl_rwb *rwb, struct ftl_rwb_batch *batch, unsigned in
 }
 
 struct ftl_rwb *
-ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size)
+ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size, size_t num_punits)
 {
-	struct ftl_rwb *rwb;
+	struct ftl_rwb *rwb = NULL;
 	struct ftl_rwb_batch *batch;
 	size_t i;
 
@@ -194,8 +205,10 @@ ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size)
 
 	assert(conf->rwb_size % xfer_size == 0);
 	rwb->xfer_size = xfer_size;
+	rwb->interleave_offset = xfer_size / conf->num_interleave_units;
+	rwb->max_active_batches = conf->num_interleave_units == 1 ? 1 : num_punits;
 	rwb->md_size = md_size;
-	rwb->num_batches = conf->rwb_size / (FTL_BLOCK_SIZE * xfer_size);
+	rwb->num_batches = conf->rwb_size / (FTL_BLOCK_SIZE * xfer_size) + rwb->max_active_batches;
 
 	rwb->batches = calloc(rwb->num_batches, sizeof(*rwb->batches));
 	if (!rwb->batches) {
@@ -219,6 +232,7 @@ ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size)
 	}
 
 	STAILQ_INIT(&rwb->free_queue);
+	STAILQ_INIT(&rwb->active_queue);
 
 	for (i = 0; i < rwb->num_batches; ++i) {
 		batch = &rwb->batches[i];
@@ -229,6 +243,7 @@ ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size)
 		}
 
 		STAILQ_INSERT_TAIL(&rwb->free_queue, batch, stailq);
+		rwb->num_free_batches++;
 	}
 
 	for (unsigned int i = 0; i < FTL_RWB_TYPE_MAX; ++i) {
@@ -293,6 +308,7 @@ ftl_rwb_batch_release(struct ftl_rwb_batch *batch)
 
 	pthread_spin_lock(&rwb->lock);
 	STAILQ_INSERT_TAIL(&rwb->free_queue, batch, stailq);
+	rwb->num_free_batches++;
 	pthread_spin_unlock(&rwb->lock);
 }
 
@@ -336,6 +352,12 @@ ftl_rwb_num_acquired(struct ftl_rwb *rwb, enum ftl_rwb_entry_type type)
 	return __atomic_load_n(&rwb->num_acquired[type], __ATOMIC_SEQ_CST);
 }
 
+size_t
+ftl_rwb_get_active_batches(const struct ftl_rwb *rwb)
+{
+	return rwb->num_active_batches;
+}
+
 void
 ftl_rwb_batch_revert(struct ftl_rwb_batch *batch)
 {
@@ -370,6 +392,28 @@ ftl_rwb_check_limits(struct ftl_rwb *rwb, enum ftl_rwb_entry_type type)
 	return ftl_rwb_num_acquired(rwb, type) >= rwb->limits[type];
 }
 
+static struct ftl_rwb_batch *
+_ftl_rwb_acquire_batch(struct ftl_rwb *rwb)
+{
+	struct ftl_rwb_batch *batch;
+	size_t i;
+
+	if (rwb->num_free_batches < rwb->max_active_batches) {
+		return NULL;
+	}
+
+	for (i = 0; i < rwb->max_active_batches; i++) {
+		batch = STAILQ_FIRST(&rwb->free_queue);
+		STAILQ_REMOVE(&rwb->free_queue, batch, ftl_rwb_batch, stailq);
+		rwb->num_free_batches--;
+
+		STAILQ_INSERT_TAIL(&rwb->active_queue, batch, stailq);
+		rwb->num_active_batches++;
+	}
+
+	return STAILQ_FIRST(&rwb->active_queue);
+}
+
 struct ftl_rwb_entry *
 ftl_rwb_acquire(struct ftl_rwb *rwb, enum ftl_rwb_entry_type type)
 {
@@ -382,22 +426,28 @@ ftl_rwb_acquire(struct ftl_rwb *rwb, enum ftl_rwb_entry_type type)
 
 	pthread_spin_lock(&rwb->lock);
 
-	current = rwb->current;
+	current = STAILQ_FIRST(&rwb->active_queue);
 	if (!current) {
-		current = STAILQ_FIRST(&rwb->free_queue);
+		current = _ftl_rwb_acquire_batch(rwb);
 		if (!current) {
 			goto error;
 		}
-
-		STAILQ_REMOVE(&rwb->free_queue, current, ftl_rwb_batch, stailq);
-		rwb->current = current;
 	}
 
 	entry = &current->entries[current->num_acquired++];
 
-	/* If the whole batch is filled, clear the current batch pointer */
 	if (current->num_acquired >= rwb->xfer_size) {
-		rwb->current = NULL;
+		/* If the whole batch is filled, */
+		/* remove the current batch from active_queue */
+		/* since it will need to move to submit_queue */
+		STAILQ_REMOVE(&rwb->active_queue, current, ftl_rwb_batch, stailq);
+		rwb->num_active_batches--;
+	} else if (current->num_acquired % rwb->interleave_offset == 0) {
+		/* If the current batch is filled by the interleaving offset, */
+		/* move the current batch at the tail of active_queue */
+		/* to place the next logical blocks into another batch. */
+		STAILQ_REMOVE(&rwb->active_queue, current, ftl_rwb_batch, stailq);
+		STAILQ_INSERT_TAIL(&rwb->active_queue, current, stailq);
 	}
 
 	pthread_spin_unlock(&rwb->lock);
@@ -408,6 +458,30 @@ error:
 	return NULL;
 }
 
+void
+ftl_rwb_disable_interleaving(struct ftl_rwb *rwb)
+{
+	struct ftl_rwb_batch *batch, *temp;
+
+	pthread_spin_lock(&rwb->lock);
+	rwb->max_active_batches = 1;
+	rwb->interleave_offset = rwb->xfer_size;
+
+	STAILQ_FOREACH_SAFE(batch, &rwb->active_queue, stailq, temp) {
+		if (batch->num_acquired == 0) {
+			STAILQ_REMOVE(&rwb->active_queue, batch, ftl_rwb_batch, stailq);
+			rwb->num_active_batches--;
+
+			assert(batch->num_ready == 0);
+			assert(batch->num_acquired == 0);
+
+			STAILQ_INSERT_TAIL(&rwb->free_queue, batch, stailq);
+			rwb->num_free_batches++;
+		}
+	}
+	pthread_spin_unlock(&rwb->lock);
+}
+
 struct ftl_rwb_batch *
 ftl_rwb_pop(struct ftl_rwb *rwb)
 {

lib/ftl/ftl_rwb.h

@@ -90,7 +90,9 @@ struct ftl_rwb_entry {
 	LIST_ENTRY(ftl_rwb_entry)		list_entry;
 };
 
-struct ftl_rwb *ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size, size_t md_size);
+struct ftl_rwb *ftl_rwb_init(const struct spdk_ftl_conf *conf, size_t xfer_size,
+			     size_t md_size, size_t num_punits);
+size_t	ftl_rwb_get_active_batches(const struct ftl_rwb *rwb);
 void	ftl_rwb_free(struct ftl_rwb *rwb);
 void	ftl_rwb_batch_release(struct ftl_rwb_batch *batch);
 void	ftl_rwb_push(struct ftl_rwb_entry *entry);
@@ -110,6 +112,7 @@ void	ftl_rwb_batch_revert(struct ftl_rwb_batch *batch);
 struct ftl_rwb_entry *ftl_rwb_batch_first_entry(struct ftl_rwb_batch *batch);
 void	*ftl_rwb_batch_get_data(struct ftl_rwb_batch *batch);
 void	*ftl_rwb_batch_get_md(struct ftl_rwb_batch *batch);
+void	ftl_rwb_disable_interleaving(struct ftl_rwb *rwb);
 
 static inline void
 _ftl_rwb_entry_set_valid(struct ftl_rwb_entry *entry, bool valid)

test/unit/lib/ftl/ftl_rwb.c/ftl_rwb_ut.c

@@ -38,19 +38,93 @@
 
 #include "ftl/ftl_rwb.c"
 
-#define RWB_SIZE	(1024 * 1024)
+struct ftl_rwb_ut {
-#define RWB_ENTRY_COUNT	(RWB_SIZE / FTL_BLOCK_SIZE)
+	/* configurations */
-#define XFER_SIZE	16
+	struct	spdk_ftl_conf conf;
-#define METADATA_SIZE	64
+	size_t	metadata_size;
+	size_t	num_punits;
+	size_t	xfer_size;
+
+	/* the fields below are calculated by the configurations */
+	size_t	max_batches;
+	size_t	max_active_batches;
+	size_t	max_entries;
+	size_t	max_allocable_entries;
+	size_t	interleave_offset;
+	size_t	num_entries_per_worker;
+};
 
 static struct ftl_rwb *g_rwb;
+static struct ftl_rwb_ut g_ut;
+
+static int _init_suite(void);
+
+static int
+init_suite1(void)
+{
+	g_ut.conf.rwb_size = 1024 * 1024;
+	g_ut.conf.num_interleave_units = 1;
+	g_ut.metadata_size = 64;
+	g_ut.num_punits = 4;
+	g_ut.xfer_size = 16;
+
+	return _init_suite();
+}
+
+static int
+init_suite2(void)
+{
+	g_ut.conf.rwb_size = 2 * 1024 * 1024;
+	g_ut.conf.num_interleave_units = 4;
+	g_ut.metadata_size = 64;
+	g_ut.num_punits = 8;
+	g_ut.xfer_size = 16;
+
+	return _init_suite();
+}
+
+static int
+_init_suite(void)
+{
+	struct spdk_ftl_conf *conf = &g_ut.conf;
+
+	if (conf->num_interleave_units == 0 ||
+	    g_ut.xfer_size % conf->num_interleave_units ||
+	    g_ut.num_punits == 0) {
+		return -1;
+	}
+
+	g_ut.max_batches = conf->rwb_size / (FTL_BLOCK_SIZE * g_ut.xfer_size);
+	if (conf->num_interleave_units > 1) {
+		g_ut.max_batches += g_ut.num_punits;
+		g_ut.max_active_batches = g_ut.num_punits;
+	} else {
+		g_ut.max_batches++;
+		g_ut.max_active_batches = 1;
+	}
+
+	g_ut.max_entries = g_ut.max_batches * g_ut.xfer_size;
+	g_ut.max_allocable_entries = (g_ut.max_batches / g_ut.max_active_batches) *
+				     g_ut.max_active_batches * g_ut.xfer_size;
+
+	g_ut.interleave_offset = g_ut.xfer_size / conf->num_interleave_units;
+
+	/* if max_batches is less than max_active_batches * 2, */
+	/* test_rwb_limits_applied will be failed. */
+	if (g_ut.max_batches < g_ut.max_active_batches * 2) {
+		return -1;
+	}
+
+	g_ut.num_entries_per_worker = 16 * g_ut.max_allocable_entries;
+
+	return 0;
+}
 
 static void
 setup_rwb(void)
 {
-	struct spdk_ftl_conf conf = { .rwb_size = RWB_SIZE };
+	g_rwb = ftl_rwb_init(&g_ut.conf, g_ut.xfer_size,
-
+			     g_ut.metadata_size, g_ut.num_punits);
-	g_rwb = ftl_rwb_init(&conf, XFER_SIZE, METADATA_SIZE);
 	SPDK_CU_ASSERT_FATAL(g_rwb != NULL);
 }
 
@@ -64,12 +138,12 @@ cleanup_rwb(void)
 static void
 test_rwb_acquire(void)
 {
-	size_t i;
 	struct ftl_rwb_entry *entry;
+	size_t i;
 
 	setup_rwb();
 	/* Verify that it's possible to acquire all of the entries */
-	for (i = 0; i < RWB_ENTRY_COUNT; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 		SPDK_CU_ASSERT_FATAL(entry);
 		ftl_rwb_push(entry);
@@ -85,52 +159,118 @@ test_rwb_pop(void)
 {
 	struct ftl_rwb_entry *entry;
 	struct ftl_rwb_batch *batch;
-	size_t entry_count, i;
+	size_t entry_count, i, i_reset = 0, i_offset = 0;
+	uint64_t expected_lba;
 
 	setup_rwb();
+
 	/* Acquire all entries */
-	for (i = 0; i < RWB_ENTRY_COUNT; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
+
 		SPDK_CU_ASSERT_FATAL(entry);
 		entry->lba = i;
 		ftl_rwb_push(entry);
 	}
 
 	/* Pop all batches and free them */
-	for (i = 0; i < RWB_ENTRY_COUNT / XFER_SIZE; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries / g_ut.xfer_size; ++i) {
 		batch = ftl_rwb_pop(g_rwb);
 		SPDK_CU_ASSERT_FATAL(batch);
 		entry_count = 0;
 
 		ftl_rwb_foreach(entry, batch) {
-			CU_ASSERT_EQUAL(entry->lba, i * XFER_SIZE + entry_count);
+			if (i % g_ut.max_active_batches == 0) {
+				i_offset = i * g_ut.xfer_size;
+			}
+
+			if (entry_count % g_ut.interleave_offset == 0) {
+				i_reset = i % g_ut.max_active_batches +
+					  (entry_count / g_ut.interleave_offset) *
+					  g_ut.max_active_batches;
+			}
+
+			expected_lba = i_offset +
+				       i_reset * g_ut.interleave_offset +
+				       entry_count % g_ut.interleave_offset;
+
+			CU_ASSERT_EQUAL(entry->lba, expected_lba);
 			entry_count++;
 		}
 
-		CU_ASSERT_EQUAL(entry_count, XFER_SIZE);
+		CU_ASSERT_EQUAL(entry_count, g_ut.xfer_size);
 		ftl_rwb_batch_release(batch);
 	}
 
 	/* Acquire all entries once more */
-	for (i = 0; i < RWB_ENTRY_COUNT; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 		SPDK_CU_ASSERT_FATAL(entry);
 		ftl_rwb_push(entry);
 	}
 
-	/* Pop one batch and check we can acquire XFER_SIZE entries */
+	/* Pop one batch and check we can acquire xfer_size entries */
+	for (i = 0; i < g_ut.max_active_batches; i++) {
 		batch = ftl_rwb_pop(g_rwb);
 		SPDK_CU_ASSERT_FATAL(batch);
 		ftl_rwb_batch_release(batch);
+	}
 
-	for (i = 0; i < XFER_SIZE; ++i) {
+	for (i = 0; i < g_ut.xfer_size * g_ut.max_active_batches; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
+
 		SPDK_CU_ASSERT_FATAL(entry);
 		ftl_rwb_push(entry);
 	}
 
 	entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 	CU_ASSERT_PTR_NULL(entry);
+
+	/* Pop and Release all batches */
+	for (i = 0; i < g_ut.max_allocable_entries / g_ut.xfer_size; ++i) {
+		batch = ftl_rwb_pop(g_rwb);
+		SPDK_CU_ASSERT_FATAL(batch);
+		ftl_rwb_batch_release(batch);
+	}
+
+	cleanup_rwb();
+}
+
+static void
+test_rwb_disable_interleaving(void)
+{
+	struct ftl_rwb_entry *entry;
+	struct ftl_rwb_batch *batch;
+	size_t entry_count, i;
+
+	setup_rwb();
+
+	ftl_rwb_disable_interleaving(g_rwb);
+
+	/* Acquire all entries and assign sequential lbas */
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
+		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
+
+		SPDK_CU_ASSERT_FATAL(entry);
+		entry->lba = i;
+		ftl_rwb_push(entry);
+	}
+
+	/* Check for expected lbas */
+	for (i = 0; i < g_ut.max_allocable_entries / g_ut.xfer_size; ++i) {
+		batch = ftl_rwb_pop(g_rwb);
+		SPDK_CU_ASSERT_FATAL(batch);
+		entry_count = 0;
+
+		ftl_rwb_foreach(entry, batch) {
+			CU_ASSERT_EQUAL(entry->lba, i * g_ut.xfer_size + entry_count);
+			entry_count++;
+		}
+
+		CU_ASSERT_EQUAL(entry_count, g_ut.xfer_size);
+		ftl_rwb_batch_release(batch);
+	}
+
 	cleanup_rwb();
 }
 
@@ -142,7 +282,7 @@ test_rwb_batch_revert(void)
 	size_t i;
 
 	setup_rwb();
-	for (i = 0; i < RWB_ENTRY_COUNT; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 		SPDK_CU_ASSERT_FATAL(entry);
 		ftl_rwb_push(entry);
@@ -155,7 +295,7 @@ test_rwb_batch_revert(void)
 	ftl_rwb_batch_revert(batch);
 
 	/* Verify all of the batches */
-	for (i = 0; i < RWB_ENTRY_COUNT / XFER_SIZE; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries / g_ut.xfer_size; ++i) {
 		batch = ftl_rwb_pop(g_rwb);
 		CU_ASSERT_PTR_NOT_NULL_FATAL(batch);
 	}
@@ -170,7 +310,7 @@ test_rwb_entry_from_offset(void)
 	size_t i;
 
 	setup_rwb();
-	for (i = 0; i < RWB_ENTRY_COUNT; ++i) {
+	for (i = 0; i < g_ut.max_allocable_entries; ++i) {
 		ppa.offset = i;
 
 		entry = ftl_rwb_entry_from_offset(g_rwb, i);
@@ -182,12 +322,11 @@ test_rwb_entry_from_offset(void)
 static void *
 test_rwb_worker(void *ctx)
 {
-#define ENTRIES_PER_WORKER (16 * RWB_ENTRY_COUNT)
 	struct ftl_rwb_entry *entry;
 	unsigned int *num_done = ctx;
 	size_t i;
 
-	for (i = 0; i < ENTRIES_PER_WORKER; ++i) {
+	for (i = 0; i < g_ut.num_entries_per_worker; ++i) {
 		while (1) {
 			entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 			if (entry) {
@@ -251,7 +390,7 @@ test_rwb_parallel(void)
 		}
 	}
 
-	CU_ASSERT_TRUE(num_entries == NUM_PARALLEL_WORKERS * ENTRIES_PER_WORKER);
+	CU_ASSERT_TRUE(num_entries == NUM_PARALLEL_WORKERS * g_ut.num_entries_per_worker);
 	cleanup_rwb();
 }
 
@@ -282,6 +421,7 @@ test_rwb_limits_set(void)
 	size_t i;
 
 	setup_rwb();
+
 	/* Check valid limits */
 	ftl_rwb_get_limits(g_rwb, limits);
 	memcpy(check, limits, sizeof(limits));
@@ -307,9 +447,11 @@ test_rwb_limits_applied(void)
 	struct ftl_rwb_entry *entry;
 	struct ftl_rwb_batch *batch;
 	size_t limits[FTL_RWB_TYPE_MAX];
+	const size_t test_limit = g_ut.xfer_size * g_ut.max_active_batches;
 	size_t i;
 
 	setup_rwb();
+
 	/* Check that it's impossible to acquire any entries when the limits are */
 	/* set to 0 */
 	ftl_rwb_get_limits(g_rwb, limits);
@@ -325,11 +467,10 @@ test_rwb_limits_applied(void)
 	CU_ASSERT_PTR_NULL(entry);
 
 	/* Check positive limits */
-#define TEST_LIMIT XFER_SIZE
 	limits[FTL_RWB_TYPE_USER] = ftl_rwb_entry_cnt(g_rwb);
-	limits[FTL_RWB_TYPE_INTERNAL] = TEST_LIMIT;
+	limits[FTL_RWB_TYPE_INTERNAL] = test_limit;
 	ftl_rwb_set_limits(g_rwb, limits);
-	for (i = 0; i < TEST_LIMIT; ++i) {
+	for (i = 0; i < test_limit; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_INTERNAL);
 		SPDK_CU_ASSERT_FATAL(entry);
 		entry->flags = FTL_IO_INTERNAL;
@@ -340,20 +481,22 @@ test_rwb_limits_applied(void)
 	entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_INTERNAL);
 	CU_ASSERT_PTR_NULL(entry);
 
+	for (i = 0; i < test_limit / g_ut.xfer_size; ++i) {
 		/* Complete the entries and check we can retrieve the entries once again */
 		batch = ftl_rwb_pop(g_rwb);
 		SPDK_CU_ASSERT_FATAL(batch);
 		ftl_rwb_batch_release(batch);
+	}
 
 	entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_INTERNAL);
-	CU_ASSERT_PTR_NOT_NULL_FATAL(entry);
+	SPDK_CU_ASSERT_FATAL(entry);
 	entry->flags = FTL_IO_INTERNAL;
 
 	/* Set the same limit but this time for user entries */
-	limits[FTL_RWB_TYPE_USER] = TEST_LIMIT;
+	limits[FTL_RWB_TYPE_USER] = test_limit;
 	limits[FTL_RWB_TYPE_INTERNAL] = ftl_rwb_entry_cnt(g_rwb);
 	ftl_rwb_set_limits(g_rwb, limits);
-	for (i = 0; i < TEST_LIMIT; ++i) {
+	for (i = 0; i < test_limit; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_USER);
 		SPDK_CU_ASSERT_FATAL(entry);
 		ftl_rwb_push(entry);
@@ -365,45 +508,72 @@ test_rwb_limits_applied(void)
 
 	/* Check that we're still able to acquire a number of internal entries */
 	/* while the user entires are being throttled */
-	for (i = 0; i < TEST_LIMIT; ++i) {
+	for (i = 0; i < g_ut.xfer_size; ++i) {
 		entry = ftl_rwb_acquire(g_rwb, FTL_RWB_TYPE_INTERNAL);
 		SPDK_CU_ASSERT_FATAL(entry);
 	}
+
 	cleanup_rwb();
 }
 
 int
 main(int argc, char **argv)
 {
-	CU_pSuite suite;
+	CU_pSuite suite1, suite2;
 	unsigned int num_failures;
 
 	if (CU_initialize_registry() != CUE_SUCCESS) {
 		return CU_get_error();
 	}
 
-	suite = CU_add_suite("ftl_rwb_suite", NULL, NULL);
+	suite1 = CU_add_suite("suite1", init_suite1, NULL);
-	if (!suite) {
+	if (!suite1) {
+		CU_cleanup_registry();
+		return CU_get_error();
+	}
+
+	suite2 = CU_add_suite("suite2", init_suite2, NULL);
+	if (!suite2) {
 		CU_cleanup_registry();
 		return CU_get_error();
 	}
 
 	if (
-		CU_add_test(suite, "test_rwb_acquire",
+		CU_add_test(suite1, "test_rwb_acquire",
 			    test_rwb_acquire) == NULL
-		|| CU_add_test(suite, "test_rwb_pop",
+		|| CU_add_test(suite1, "test_rwb_pop",
 			       test_rwb_pop) == NULL
-		|| CU_add_test(suite, "test_rwb_batch_revert",
+		|| CU_add_test(suite1, "test_rwb_disable_interleaving",
+			       test_rwb_disable_interleaving) == NULL
+		|| CU_add_test(suite1, "test_rwb_batch_revert",
 			       test_rwb_batch_revert) == NULL
-		|| CU_add_test(suite, "test_rwb_entry_from_offset",
+		|| CU_add_test(suite1, "test_rwb_entry_from_offset",
 			       test_rwb_entry_from_offset) == NULL
-		|| CU_add_test(suite, "test_rwb_parallel",
+		|| CU_add_test(suite1, "test_rwb_parallel",
 			       test_rwb_parallel) == NULL
-		|| CU_add_test(suite, "test_rwb_limits_base",
+		|| CU_add_test(suite1, "test_rwb_limits_base",
 			       test_rwb_limits_base) == NULL
-		|| CU_add_test(suite, "test_rwb_limits_set",
+		|| CU_add_test(suite1, "test_rwb_limits_set",
 			       test_rwb_limits_set) == NULL
-		|| CU_add_test(suite, "test_rwb_limits_applied",
+		|| CU_add_test(suite1, "test_rwb_limits_applied",
+			       test_rwb_limits_applied) == NULL
+		|| CU_add_test(suite2, "test_rwb_acquire",
+			       test_rwb_acquire) == NULL
+		|| CU_add_test(suite2, "test_rwb_pop",
+			       test_rwb_pop) == NULL
+		|| CU_add_test(suite2, "test_rwb_disable_interleaving",
+			       test_rwb_disable_interleaving) == NULL
+		|| CU_add_test(suite2, "test_rwb_batch_revert",
+			       test_rwb_batch_revert) == NULL
+		|| CU_add_test(suite2, "test_rwb_entry_from_offset",
+			       test_rwb_entry_from_offset) == NULL
+		|| CU_add_test(suite2, "test_rwb_parallel",
+			       test_rwb_parallel) == NULL
+		|| CU_add_test(suite2, "test_rwb_limits_base",
+			       test_rwb_limits_base) == NULL
+		|| CU_add_test(suite2, "test_rwb_limits_set",
+			       test_rwb_limits_set) == NULL
+		|| CU_add_test(suite2, "test_rwb_limits_applied",
 			       test_rwb_limits_applied) == NULL
 	) {
 		CU_cleanup_registry();