LU-7795 quota: tuneable soft least qunit

[fs/lustre-release.git] / lustre / quota / qmt_internal.h

diff --git a/lustre/quota/qmt_internal.h b/lustre/quota/qmt_internal.h
index e21c78f..7f02256 100644 (file)
--- a/lustre/quota/qmt_internal.h
+++ b/lustre/quota/qmt_internal.h
@@ -139,6 +139,22 @@ struct qmt_pool_info {
        /* Global quota parameters which apply to all quota type */
        /* the least value of qunit */
        unsigned long            qpi_least_qunit;
+
+       /* Least value of qunit when soft limit is exceeded.
+        *
+        * When soft limit is exceeded, qunit will be shrinked to least_qunit
+        * (1M for block limit), that results in significant write performance
+        * drop since the client will turn to sync write from now on.
+        *
+        * To retain the write performance in an acceptable level, we choose
+        * to sacrifice grace time accuracy a bit and use a larger least_qunit
+        * when soft limit is exceeded. It's (qpi_least_qunit * 4) by default,
+        * and user may enlarge it via procfs to get even better performance
+        * (with the cost of losing more grace time accuracy).
+        *
+        * See qmt_calc_softlimit().
+        */
+       unsigned long            qpi_soft_least_qunit;
 };
 
 /*