LU-12385 lnet: update opa defaults

Testing reveals no significant performance improvements
when using peer_credits > 32. Adjusted the default
peer_credits, peer_credits_hiw and concurrent_sends
to take that into account.

This has the advantage of avoiding an issue observed
on multiple opa sites where the qp can not be created because
of large initial queue_depth. The queue depth is then
reduced gradually until the qp creation succeeds.

Signed-off-by: Amir Shehata <ashehata@whamcloud.com>
Change-Id: I6036ec1da7063e30b567446e5db89040f21bc701
Reviewed-on: https://review.whamcloud.com/36072
Reviewed-by: Chris Horn <hornc@cray.com>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/conf/ko2iblnd.conf b/lustre/conf/ko2iblnd.conf
index b486895..829c601 100644 (file)
--- a/lustre/conf/ko2iblnd.conf
+++ b/lustre/conf/ko2iblnd.conf
@@ -13,6 +13,6 @@
 # Lustre is first mounted.
 
 alias ko2iblnd-opa ko2iblnd
-options ko2iblnd-opa peer_credits=128 peer_credits_hiw=64 credits=1024 concurrent_sends=256 ntx=2048 map_on_demand=256 fmr_pool_size=2048 fmr_flush_trigger=512 fmr_cache=1 conns_per_peer=4
+options ko2iblnd-opa peer_credits=32 peer_credits_hiw=16 credits=1024 concurrent_sends=64 ntx=2048 map_on_demand=256 fmr_pool_size=2048 fmr_flush_trigger=512 fmr_cache=1 conns_per_peer=4
 
 install ko2iblnd /usr/sbin/ko2iblnd-probe