DAOS-18164 bio: bump default cluster_sz

Bump default BS cluster size from 32MB to 128MB in md-on-ssd mode,
make the cluster size configurable through DAOS_BS_CLUSTER_MB.

Signed-off-by: Niu Yawei <[email protected]>

Author: NiuYawei

src/bio/bio_xstream.c

@@ -1,7 +1,7 @@
 /**
  * (C) Copyright 2018-2024 Intel Corporation.
  * (C) Copyright 2025 Google LLC
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -31,8 +31,8 @@
 
 /* These Macros should be turned into DAOS configuration in the future */
 #define DAOS_MSG_RING_SZ	4096
-/* SPDK blob parameters */
-#define DAOS_BS_CLUSTER_SZ	(1ULL << 25)	/* 32MB */
+/* Default cluster size in MB */
+#define DAOS_DEFAULT_CLUSTER_MB 128
 /* DMA buffer parameters */
 #define DAOS_DMA_CHUNK_INIT_PCT 50      /* Default per-xstream init chunks, in percentage */
 #define DAOS_DMA_CHUNK_CNT_MAX	128	/* Default per-xstream max chunks, 1GB */
@@ -224,6 +224,7 @@ bio_nvme_init_ext(const char *nvme_conf, int numa_node, unsigned int mem_size,
 	char		*env;
 	int		 rc, fd;
 	unsigned int     size_mb = BIO_DMA_CHUNK_MB, io_timeout_secs = 0;
+	unsigned int     cluster_mb = DAOS_DEFAULT_CLUSTER_MB;
 
 	if (tgt_nr <= 0) {
 		D_ERROR("tgt_nr: %u should be > 0\n", tgt_nr);
@@ -323,8 +324,14 @@ bio_nvme_init_ext(const char *nvme_conf, int numa_node, unsigned int mem_size,
 	D_INFO("Set per-xstream DMA buffer upper bound to %u %uMB chunks, prealloc %u chunks\n",
 	       bio_chk_cnt_max, size_mb, init_chk_cnt());
 
+	d_getenv_uint("DAOS_BS_CLUSTER_MB", &cluster_mb);
+	if (cluster_mb < 32 || cluster_mb > 1024) {
+		D_WARN("DAOS_BS_CLUSTER_MB %u is invalid, default %u is used\n", cluster_mb,
+		       DAOS_DEFAULT_CLUSTER_MB);
+		cluster_mb = DAOS_DEFAULT_CLUSTER_MB;
+	}
 	spdk_bs_opts_init(&nvme_glb.bd_bs_opts, sizeof(nvme_glb.bd_bs_opts));
-	nvme_glb.bd_bs_opts.cluster_sz = DAOS_BS_CLUSTER_SZ;
+	nvme_glb.bd_bs_opts.cluster_sz      = (cluster_mb << 20);
 	nvme_glb.bd_bs_opts.max_channel_ops = BIO_BS_MAX_CHANNEL_OPS;
 
 	d_agetenv_str(&env, "VOS_BDEV_CLASS");
@@ -368,8 +375,9 @@ bio_nvme_init_ext(const char *nvme_conf, int numa_node, unsigned int mem_size,
 	if (!bio_nvme_configured(SMD_DEV_TYPE_META))
 		nvme_glb.bd_bs_opts.cluster_sz = (1UL << 30);	/* 1GB */
 
-	D_INFO("MD on SSD is %s\n",
-	       bio_nvme_configured(SMD_DEV_TYPE_META) ? "enabled" : "disabled");
+	D_INFO("MD on SSD is %s, %u cluster size is used\n",
+	       bio_nvme_configured(SMD_DEV_TYPE_META) ? "enabled" : "disabled",
+	       nvme_glb.bd_bs_opts.cluster_sz);
 
 	bio_spdk_inited = true;
 

src/tests/suite/daos_md_replication.c

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2017-2022 Intel Corporation.
+ * (C) Copyright 2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -11,6 +12,12 @@
 #include <daos/pool.h>
 #include "daos_test.h"
 
+/*
+ * Given the 128MB default blobstore cluster size, the minimal pool scm_size for
+ * an 8 targets engine would be 128MB * 8 = 1GB.
+ */
+#define MIN_SCM_SIZE (1ULL << 30)
+
 static void
 mdr_stop_pool_svc(void **argv)
 {
@@ -24,10 +31,8 @@ mdr_stop_pool_svc(void **argv)
 	/* Create the pool. */
 	if (arg->myrank == 0) {
 		print_message("creating pool\n");
-		rc = dmg_pool_create(dmg_config_file,
-				     geteuid(), getegid(), arg->group,
-				     NULL, 256 * 1024 * 1024, 0,
-				     NULL, arg->pool.svc, uuid);
+		rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL,
+				     MIN_SCM_SIZE, 0, NULL, arg->pool.svc, uuid);
 	}
 	par_bcast(PAR_COMM_WORLD, &rc, 1, PAR_INT, 0);
 	assert_rc_equal(rc, 0);
@@ -134,10 +139,8 @@ mdr_stop_cont_svc(void **argv)
 	int			rc;
 
 	print_message("creating pool\n");
-	rc = dmg_pool_create(dmg_config_file,
-			     geteuid(), getegid(), arg->group,
-			     NULL, 256 * 1024 * 1024, 0,
-			     NULL, arg->pool.svc, pool_uuid);
+	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL, MIN_SCM_SIZE,
+			     0, NULL, arg->pool.svc, pool_uuid);
 	assert_rc_equal(rc, 0);
 
 	if (arg->pool.svc->rl_nr < 3) {

src/tests/suite/daos_mgmt.c

@@ -1,6 +1,6 @@
 /**
  * (C) Copyright 2016-2024 Intel Corporation.
- * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
+ * (C) Copyright 2025-2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -20,6 +20,12 @@
 #include <daos_event.h>
 #include <daos/agent.h>
 
+/*
+ * Given the 128MB default blobstore cluster size, the minimal pool scm_size for
+ * an 8 targets engine would be 128MB * 8 = 1GB.
+ */
+#define MIN_SCM_SIZE (1ULL << 30)
+
 /** create/destroy pool on all tgts */
 static void
 pool_create_all(void **state)
@@ -36,11 +42,8 @@ pool_create_all(void **state)
 
 	/** create container */
 	print_message("creating pool synchronously ... ");
-	rc = dmg_pool_create(dmg_config_file,
-			     geteuid(), getegid(),
-			     arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */,
-			     0 /* nvme size */, NULL /* prop */,
+	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */,
 			     arg->pool.svc /* svc */, uuid);
 	assert_rc_equal(rc, 0);
 
@@ -341,11 +344,8 @@ pool_create_and_destroy_retry(void **state)
 
 	test_set_engine_fail_loc(arg, CRT_NO_RANK, DAOS_POOL_CREATE_FAIL_CORPC | DAOS_FAIL_ONCE);
 	print_message("creating pool synchronously ... ");
-	rc = dmg_pool_create(dmg_config_file,
-			     geteuid(), getegid(),
-			     arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */,
-			     0 /* nvme size */, NULL /* prop */,
+	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */,
 			     arg->pool.svc /* svc */, uuid);
 	assert_rc_equal(rc, 0);
 	print_message("success uuid = "DF_UUIDF"\n", DP_UUID(uuid));
@@ -435,8 +435,7 @@ pool_create_steps_down_from_up_empty(void **state)
 	svc.rl_ranks = &rank;
 	svc.rl_nr = 1;
 	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */, 0 /* nvme size */,
-			     NULL /* prop */, &svc, uuid);
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */, &svc, uuid);
 	assert_rc_equal(rc, 0);
 	print_message("success uuid = "DF_UUIDF"\n", DP_UUID(uuid));
 
@@ -466,8 +465,7 @@ pool_destroy_disconnect_all(void **state)
 
 	print_message("creating pool synchronously ... ");
 	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */, 0 /* nvme size */,
-			     NULL /* prop */, arg->pool.svc, uuid);
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */, arg->pool.svc, uuid);
 	assert_rc_equal(rc, 0);
 	print_message("success uuid = "DF_UUIDF"\n", DP_UUID(uuid));
 
@@ -515,8 +513,7 @@ pool_destroy_cancel_rfcheck(void **state)
 
 	print_message("creating pool synchronously ... ");
 	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */, 0 /* nvme size */,
-			     NULL /* prop */, arg->pool.svc, uuid);
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */, arg->pool.svc, uuid);
 	assert_rc_equal(rc, 0);
 	print_message("success uuid = "DF_UUIDF"\n", DP_UUID(uuid));
 
@@ -544,8 +541,7 @@ pool_create_query_fail(void **state)
 
 	print_message("creating pool synchronously ... ");
 	rc = dmg_pool_create(dmg_config_file, geteuid(), getegid(), arg->group, NULL /* tgts */,
-			     256 * 1024 * 1024 /* minimal size */, 0 /* nvme size */,
-			     NULL /* prop */, arg->pool.svc, uuid);
+			     MIN_SCM_SIZE, 0 /* nvme size */, NULL /* prop */, arg->pool.svc, uuid);
 	assert_rc_equal(rc, 0);
 	print_message("success uuid = " DF_UUIDF "\n", DP_UUID(uuid));
 

src/vos/tests/wal_ut.c

@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2023-2024 Intel Corporation.
+ * (C) Copyright 2026 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -669,8 +670,8 @@ ut_fill_wal(struct bio_ut_args *args, int tx_nr, struct ut_tx_array **txa_ptr)
 	tx = txa->ta_tx_ptrs[0];
 
 	/*
-	 * Each tx is roughly 800k, 40 txs will consume 32000k, which is more than
-	 * half of 50MB WAL size.
+	 * Each tx is roughly 800k, 100 txs will consume 80MB, which is more than
+	 * half of 128MB WAL size.
 	 */
 	for (i = 0; i < tx_nr; i++) {
 		tx = txa->ta_tx_ptrs[i];
@@ -705,11 +706,11 @@ static void
 wal_ut_wrap(void **state)
 {
 	struct bio_ut_args	*args = *state;
-	uint64_t		 meta_sz = (50ULL << 20);	/* 50 MB */
+	uint64_t                 meta_sz = (128ULL << 20); /* 128 MB */
 	struct ut_tx_array	*txa;
 	struct umem_wal_tx	*tx;
 	struct ut_fake_tx	*fake_tx;
-	int			 tx_nr = 40, rc;
+	int                      tx_nr = 100, rc;
 
 	rc = ut_mc_init(args, meta_sz, meta_sz, meta_sz);
 	assert_rc_equal(rc, 0);
@@ -745,11 +746,11 @@ static void
 wal_ut_wrap_many(void **state)
 {
 	struct bio_ut_args	*args = *state;
-	uint64_t		 meta_sz = (50ULL << 20);	/* 50 MB */
+	uint64_t                 meta_sz = (128ULL << 20); /* 128 MB */
 	struct ut_tx_array	*txa;
 	struct umem_wal_tx	*tx;
 	struct ut_fake_tx	*fake_tx;
-	int			 tx_nr = 40, rc;
+	int                      tx_nr = 100, rc;
 
 	rc = ut_mc_init(args, meta_sz, meta_sz, meta_sz);
 	assert_rc_equal(rc, 0);