DAOS-16640 vos: tune aggregation credits for phase2 pool (#15735)

When md-on-ssd phase2 pool runs into space pressure, larger SCAN credits will be used to reduce yield & reprobe on iterating, larger DEL credits will be used to drop more punched objects to GC in one batch, so that GC will likely reclaim more objects when reclaiming a bucket. Though larger aggregation credits will lower front end I/O performance, it can greatly reduce page misses for GC when free space/page is tight. This PR also fixed the NE space pressure check, where the sys reserved space should not be taken into account. Signed-off-by: Niu Yawei <[email protected]>
daos-stack · Jan 20, 2025 · 0be2d10 · 0be2d10
1 parent 4a95784
commit 0be2d10
Show file tree

Hide file tree

Showing 2 changed files with 25 additions and 14 deletions.
diff --git a/src/engine/sched.c b/src/engine/sched.c
@@ -771,7 +771,7 @@ check_space_pressure(struct dss_xstream *dx, struct sched_pool_info *spi)
 {
 	struct sched_info	*info = &dx->dx_sched_info;
 	struct vos_pool_space	 vps = { 0 };
-	uint64_t		 scm_left, nvme_left, ne_left, ne_sys;
+	uint64_t                 scm_left, nvme_left, ne_left;
 	struct pressure_ratio	*pr;
 	int			 orig_pressure, rc;
 
@@ -807,12 +807,8 @@ check_space_pressure(struct dss_xstream *dx, struct sched_pool_info *spi)
 	if (vps.vps_ne_total == 0) {
 		ne_left = UINT64_MAX;
 	} else {
-		D_ASSERT(vps.vps_ne_total < SCM_TOTAL(&vps));
-		ne_sys = SCM_SYS(&vps) * vps.vps_ne_total / SCM_TOTAL(&vps);
-		if (vps.vps_ne_free > ne_sys)
-			ne_left = vps.vps_ne_free - ne_sys;
-		else
-			ne_left = 0;
+		ne_left = vps.vps_ne_free;
+		D_ASSERT(ne_left <= vps.vps_ne_total);
 	}
 
 	if (NVME_TOTAL(&vps) == 0)      /* NVMe not enabled */

diff --git a/src/vos/vos_aggregate.c b/src/vos/vos_aggregate.c
@@ -1,5 +1,6 @@
 /**
  * (C) Copyright 2019-2024 Intel Corporation.
+ * (C) Copyright 2025 Hewlett Packard Enterprise Development LP
  *
  * SPDX-License-Identifier: BSD-2-Clause-Patent
  */
@@ -177,10 +178,24 @@ struct vos_agg_param {
 };
 
 static inline void
-credits_set(struct vos_agg_credits *vac, bool tight)
+credits_set(struct vos_pool *pool, struct vos_agg_credits *vac, bool tight)
 {
-	vac->vac_creds_scan = tight ? AGG_CREDS_SCAN_TIGHT : AGG_CREDS_SCAN_SLACK;
-	vac->vac_creds_del = tight ? AGG_CREDS_DEL_TIGHT : AGG_CREDS_DEL_SLACK;
+	unsigned int multiplier = 1;
+
+	/*
+	 * When md-on-ssd phase2 pool runs into space pressure, larger SCAN credits will
+	 * be used to reduce yield & reprobe on iterating, larger DEL credits will be used
+	 * to drop more punched objects to GC in one batch, so that GC will likely reclaim
+	 * more objects when reclaiming a bucket.
+	 *
+	 * Though larger aggregation credits will lower front end I/O performance, it can
+	 * greatly reduce page misses for GC when free space/page is tight.
+	 */
+	if (tight && vos_pool_is_evictable(pool))
+		multiplier = 100;
+
+	vac->vac_creds_scan  = (tight ? AGG_CREDS_SCAN_TIGHT : AGG_CREDS_SCAN_SLACK) * multiplier;
+	vac->vac_creds_del   = (tight ? AGG_CREDS_DEL_TIGHT : AGG_CREDS_DEL_SLACK) * multiplier;
 	vac->vac_creds_merge = tight ? AGG_CREDS_MERGE_TIGHT : AGG_CREDS_MERGE_SLACK;
 }
 
@@ -323,7 +338,7 @@ vos_aggregate_yield(struct vos_agg_param *agg_param)
 
 	if (agg_param->ap_yield_func == NULL) {
 		bio_yield(agg_param->ap_umm);
-		credits_set(&agg_param->ap_credits, true);
+		credits_set(cont->vc_pool, &agg_param->ap_credits, true);
 		return false;
 	}
 
@@ -333,7 +348,7 @@ vos_aggregate_yield(struct vos_agg_param *agg_param)
 		return true;
 
 	/* rc == 0: tight mode; rc == 1: slack mode */
-	credits_set(&agg_param->ap_credits, rc == 0);
+	credits_set(cont->vc_pool, &agg_param->ap_credits, rc == 0);
 
 	return false;
 }
@@ -2702,7 +2717,7 @@ vos_aggregate(daos_handle_t coh, daos_epoch_range_t *epr,
 	/* Set aggregation parameters */
 	ad->ad_agg_param.ap_umm = &cont->vc_pool->vp_umm;
 	ad->ad_agg_param.ap_coh = coh;
-	credits_set(&ad->ad_agg_param.ap_credits, true);
+	credits_set(cont->vc_pool, &ad->ad_agg_param.ap_credits, true);
 	ad->ad_agg_param.ap_discard = 0;
 	ad->ad_agg_param.ap_yield_func = yield_func;
 	ad->ad_agg_param.ap_yield_arg = yield_arg;
@@ -2822,7 +2837,7 @@ vos_discard(daos_handle_t coh, daos_unit_oid_t *oidp, daos_epoch_range_t *epr,
 	/* Set aggregation parameters */
 	ad->ad_agg_param.ap_umm = &cont->vc_pool->vp_umm;
 	ad->ad_agg_param.ap_coh = coh;
-	credits_set(&ad->ad_agg_param.ap_credits, true);
+	credits_set(cont->vc_pool, &ad->ad_agg_param.ap_credits, true);
 	ad->ad_agg_param.ap_discard = 1;
 	ad->ad_agg_param.ap_yield_func = yield_func;
 	ad->ad_agg_param.ap_yield_arg = yield_arg;