Public bug reported:
we see degradation ~20% on ConnectX-5/4 in the following case:
TCP, 1 QP, 1 stream, unidir, single port.
Message sizes 1M and up show this degradation.
After changing the default TX moderation mode to off we see up to 40%
packet rate and up to 23% bandwidth degradtions.
There is an upstream commit that fix this issue, I will backport it and send it
to the kernel-t...@lists.ubuntu.com
commit 48bfc39791b8b4a25f165e711f18b9c1617cefbc
Author: Tal Gilboa <ta...@mellanox.com>
Date: Fri Mar 30 15:50:08 2018 -0700
net/mlx5e: Set EQE based as default TX interrupt moderation mode
The default TX moderation mode was mistakenly set to CQE based. The
intention was to add a control ability in order to improve some specific
use-cases. In general, we prefer to use EQE based moderation as it gives
much better numbers for the common cases.
CQE based causes a degradation in the common case since it resets the
moderation timer on CQE generation. This causes an issue when TSO is
well utilized (large TSO sessions). The timer is set to 16us so traffic
of ~64KB TSO sessions per second would mean timer reset (CQE per TSO
session -> long time between CQEs). In this case we quickly reach the
tcp_limit_output_bytes (256KB by default) and cause a halt in TX traffic.
By setting EQE based moderation we make sure timer would expire after
16us regardless of the packet rate.
This fixes an up to 40% packet rate and up to 23% bandwidth degradtions.
Fixes: 0088cbbc4b66 ("net/mlx5e: Enable CQE based moderation on TX CQ")
Signed-off-by: Tal Gilboa <ta...@mellanox.com>
Signed-off-by: Saeed Mahameed <sae...@mellanox.com>
Signed-off-by: David S. Miller <da...@davemloft.net>
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index c71f4f10283b..0aab3afc6885 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4137,7 +4137,7 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
struct mlx5e_params *params,
u16 max_channels, u16 mtu)
{
- u8 cq_period_mode = 0;
+ u8 rx_cq_period_mode;
params->sw_mtu = mtu;
params->hard_mtu = MLX5E_ETH_HARD_MTU;
@@ -4173,12 +4173,12 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
params->lro_timeout = mlx5e_choose_lro_timeout(mdev,
MLX5E_DEFAULT_LRO_TIMEOUT);
/* CQ moderation params */
- cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
+ rx_cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
- mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
- mlx5e_set_tx_cq_mode_params(params, cq_period_mode);
+ mlx5e_set_rx_cq_mode_params(params, rx_cq_period_mode);
+ mlx5e_set_tx_cq_mode_params(params, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
/* TX inline */
params->tx_min_inline_mode = mlx5e_params_calculate_tx_min_inline(mdev);
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1763325
Title:
[bionic] ConnectX5 Large message size throughput degradation in TCP
Status in linux package in Ubuntu:
New
Bug description:
we see degradation ~20% on ConnectX-5/4 in the following case:
TCP, 1 QP, 1 stream, unidir, single port.
Message sizes 1M and up show this degradation.
After changing the default TX moderation mode to off we see up to 40%
packet rate and up to 23% bandwidth degradtions.
There is an upstream commit that fix this issue, I will backport it and send
it to the kernel-t...@lists.ubuntu.com
commit 48bfc39791b8b4a25f165e711f18b9c1617cefbc
Author: Tal Gilboa <ta...@mellanox.com>
Date: Fri Mar 30 15:50:08 2018 -0700
net/mlx5e: Set EQE based as default TX interrupt moderation mode
The default TX moderation mode was mistakenly set to CQE based. The
intention was to add a control ability in order to improve some specific
use-cases. In general, we prefer to use EQE based moderation as it gives
much better numbers for the common cases.
CQE based causes a degradation in the common case since it resets the
moderation timer on CQE generation. This causes an issue when TSO is
well utilized (large TSO sessions). The timer is set to 16us so traffic
of ~64KB TSO sessions per second would mean timer reset (CQE per TSO
session -> long time between CQEs). In this case we quickly reach the
tcp_limit_output_bytes (256KB by default) and cause a halt in TX traffic.
By setting EQE based moderation we make sure timer would expire after
16us regardless of the packet rate.
This fixes an up to 40% packet rate and up to 23% bandwidth degradtions.
Fixes: 0088cbbc4b66 ("net/mlx5e: Enable CQE based moderation on TX CQ")
Signed-off-by: Tal Gilboa <ta...@mellanox.com>
Signed-off-by: Saeed Mahameed <sae...@mellanox.com>
Signed-off-by: David S. Miller <da...@davemloft.net>
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
index c71f4f10283b..0aab3afc6885 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
+++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c
@@ -4137,7 +4137,7 @@ void mlx5e_build_nic_params(struct mlx5_core_dev *mdev,
struct mlx5e_params *params,
u16 max_channels, u16 mtu)
{
- u8 cq_period_mode = 0;
+ u8 rx_cq_period_mode;
params->sw_mtu = mtu;
params->hard_mtu = MLX5E_ETH_HARD_MTU;
@@ -4173,12 +4173,12 @@ void mlx5e_build_nic_params(struct mlx5_core_dev
*mdev,
params->lro_timeout = mlx5e_choose_lro_timeout(mdev,
MLX5E_DEFAULT_LRO_TIMEOUT);
/* CQ moderation params */
- cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
+ rx_cq_period_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ?
MLX5_CQ_PERIOD_MODE_START_FROM_CQE :
MLX5_CQ_PERIOD_MODE_START_FROM_EQE;
params->rx_dim_enabled = MLX5_CAP_GEN(mdev, cq_moderation);
- mlx5e_set_rx_cq_mode_params(params, cq_period_mode);
- mlx5e_set_tx_cq_mode_params(params, cq_period_mode);
+ mlx5e_set_rx_cq_mode_params(params, rx_cq_period_mode);
+ mlx5e_set_tx_cq_mode_params(params, MLX5_CQ_PERIOD_MODE_START_FROM_EQE);
/* TX inline */
params->tx_min_inline_mode =
mlx5e_params_calculate_tx_min_inline(mdev);
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1763325/+subscriptions
--
Mailing list: https://launchpad.net/~kernel-packages
Post to : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help : https://help.launchpad.net/ListHelp
