Browse Source

drbd: Improve asender performance

Shorten receive path in the asender thread. Reduces CPU utilisation
of asender when receiving packets, and with that increases IOPs.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Lars Ellenberg 11 years ago
parent
commit
abde9cc6a5
2 changed files with 7 additions and 10 deletions
  1. 6 0
      drivers/block/drbd/drbd_receiver.c
  2. 1 10
      drivers/block/drbd/drbd_worker.c

+ 6 - 0
drivers/block/drbd/drbd_receiver.c

@@ -5561,6 +5561,7 @@ int drbd_asender(struct drbd_thread *thi)
 		 * rv <  expected: "woken" by signal during receive
 		 * rv == 0	 : "connection shut down by peer"
 		 */
+received_more:
 		if (likely(rv > 0)) {
 			received += rv;
 			buf	 += rv;
@@ -5636,6 +5637,11 @@ int drbd_asender(struct drbd_thread *thi)
 			expect	 = header_size;
 			cmd	 = NULL;
 		}
+		if (test_bit(SEND_PING, &connection->flags))
+			continue;
+		rv = drbd_recv_short(connection->meta.socket, buf, expect-received, MSG_DONTWAIT);
+		if (rv > 0)
+			goto received_more;
 	}
 
 	if (0) {

+ 1 - 10
drivers/block/drbd/drbd_worker.c

@@ -1994,22 +1994,13 @@ static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *
 	return !list_empty(work_list);
 }
 
-static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list)
-{
-	spin_lock_irq(&queue->q_lock);
-	if (!list_empty(&queue->q))
-		list_move(queue->q.next, work_list);
-	spin_unlock_irq(&queue->q_lock);
-	return !list_empty(work_list);
-}
-
 static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list)
 {
 	DEFINE_WAIT(wait);
 	struct net_conf *nc;
 	int uncork, cork;
 
-	dequeue_work_item(&connection->sender_work, work_list);
+	dequeue_work_batch(&connection->sender_work, work_list);
 	if (!list_empty(work_list))
 		return;