Whamcloud - gitweb
LU-12065 lnd: increase CQ entries 27/34427/6
authorAmir Shehata <ashehata@whamcloud.com>
Wed, 20 Mar 2019 18:10:34 +0000 (11:10 -0700)
committerAndreas Dilger <adilger@whamcloud.com>
Thu, 21 Mar 2019 14:55:04 +0000 (14:55 +0000)
Several sites have reported RDMA timeouts. Most of the timeouts
are occurring for transmits on the active_tx queue. Transmits are
placed on the active_tx queue until a completion is received. If
there isn't enough CQ entries available, it's possible for a
completions events to be delayed, causing these timeouts.

Lustre-change: https://review.whamcloud.com/34473
Lustre-commit: bf3fc7f1a7bf82c02181c64810bc3f93b3303703

Signed-off-by: Amir Shehata <ashehata@whamcloud.com>
Change-Id: I9edad734b5860ce20af4977b4c1cdc07f25f078e
Reviewed-by: Sonia Sharma <sharmaso@whamcloud.com>
Reviewed-by: James Simmons <uja.ornl@yahoo.com>
Reviewed-on: https://review.whamcloud.com/34427
Tested-by: Jenkins
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
lnet/klnds/o2iblnd/o2iblnd.h

index e60a92d..8322382 100644 (file)
@@ -163,9 +163,7 @@ extern kib_tunables_t  kiblnd_tunables;
 #define IBLND_RECV_WRS(c)            IBLND_RX_MSGS(c)
 
 /* 2 = LNet msg + Transfer chain */
-#define IBLND_CQ_ENTRIES(c)    \
-       (IBLND_RECV_WRS(c) + 2 * kiblnd_concurrent_sends(c->ibc_version, \
-                                                        c->ibc_peer->ibp_ni))
+#define IBLND_CQ_ENTRIES(c)    (IBLND_RECV_WRS(c) + kiblnd_send_wrs(c))
 
 struct kib_hca_dev;