It's desirable to lower peer credits because smaller values allow us
to cancel outstanding traffic to down peers faster (because there is
less traffic in flight). Testing shows that peer_credits 16 does not
perform any worse than our current default. Let's make 16 the
new default.
In addition, testing shows a benefit for further increasing the
default number of immediate receive buffers. Increase this to 8.
HPE-bug-id: LUS-11421
Test-Parameters: trivial
Signed-off-by: Chris Horn <chris.horn@hpe.com>
Change-Id: I877fe408b276071f33a99c8b3b50d13f597aaa29
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/49592
Reviewed-by: Ron Gredvig <ron.gredvig@hpe.com>
Reviewed-by: Ian Ziemba <ian.ziemba@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
module_param(eq_size, uint, 0444);
MODULE_PARM_DESC(eq_size, "Default event queue size used by all kfi LNet NIs");
module_param(eq_size, uint, 0444);
MODULE_PARM_DESC(eq_size, "Default event queue size used by all kfi LNet NIs");
-unsigned int immediate_rx_buf_count = 2;
+unsigned int immediate_rx_buf_count = 8;
module_param(immediate_rx_buf_count, uint, 0444);
MODULE_PARM_DESC(immediate_rx_buf_count,
"Number of immediate multi-receive buffers posted per CPT");
module_param(immediate_rx_buf_count, uint, 0444);
MODULE_PARM_DESC(immediate_rx_buf_count,
"Number of immediate multi-receive buffers posted per CPT");
module_param(credits, int, 0444);
MODULE_PARM_DESC(credits, "Number of concurrent sends on network");
module_param(credits, int, 0444);
MODULE_PARM_DESC(credits, "Number of concurrent sends on network");
-static int peer_credits = 128;
+static int peer_credits = 16;
module_param(peer_credits, int, 0444);
MODULE_PARM_DESC(peer_credits, "Number of concurrent sends to 1 peer");
module_param(peer_credits, int, 0444);
MODULE_PARM_DESC(peer_credits, "Number of concurrent sends to 1 peer");