This patch helps to distribute load across an afr cluster
when you have clients with the same access patterns, such
as in benchmarking or map-reduce. By randomly offsetting
the initial read_child_rr, clients should distribute reads
over the afr nodes on average.
A better solution could be to randomly shuffle the children
instead of chosing a different initial offset. This should
average the reads better across the nodes, but I'm not
sure if there are any other consequences to doing this.
---
xlators/cluster/afr/src/afr.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index a94f926..118e0fc 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -2494,6 +2494,9 @@ init (xlator_t *this)
i++;
}
+ /* Randomly pick the starting read child */
+ priv->read_child_rr = random() % priv->child_count;
+
ret = 0;
out:
return ret;
--
1.6.3.1
_______________________________________________
Gluster-devel mailing list
Gluster-devel@nongnu.org
http://lists.nongnu.org/mailman/listinfo/gluster-devel
