提交 c3164d2f 编写于 作者: T Tal Gilboa 提交者: Saeed Mahameed

net/mlx5e: Added BW check for DIM decision mechanism

DIM (Dynamically-tuned Interrupt Moderation) is a mechanism designed for
changing the channel interrupt moderation values in order to reduce CPU
overhead for all traffic types.
Until now only interrupt and packet rate were sampled.
We found a scenario on which we get a false indication since a change in
DIM caused more aggregation and reduced packet rate while increasing BW.

We now regard a change as succesfull iff:
current_BW > (prev_BW + threshold) or
current_BW ~= prev_BW and current_PR > (prev_PR + threshold) or
current_BW ~= prev_BW and current_PR ~= prev_PR and
    current_IR < (prev_IR - threshold)
Where BW = Bandwidth, PR = Packet rate and IR = Interrupt rate

Improvements (ConnectX-4Lx 25GbE, single RX queue, LRO off)
    --------------------------------------------------
    packet size | before[Mb/s] | after[Mb/s] | gain  |
    2B          | 343.4        | 359.4       |  4.5% |
    16B         | 2739.7       | 2814.8      |  2.7% |
    64B         | 9739         | 10185.3     |  4.5% |

Fixes: cb3c7fd4 ("net/mlx5e: Support adaptive RX coalescing")
Signed-off-by: NTal Gilboa <talgi@mellanox.com>
Signed-off-by: NSaeed Mahameed <saeedm@mellanox.com>
上级 f729860a
......@@ -458,12 +458,14 @@ struct mlx5e_mpw_info {
struct mlx5e_rx_am_stats {
int ppms; /* packets per msec */
int bpms; /* bytes per msec */
int epms; /* events per msec */
};
struct mlx5e_rx_am_sample {
ktime_t time;
unsigned int pkt_ctr;
unsigned int byte_ctr;
u16 event_ctr;
};
......
......@@ -183,28 +183,27 @@ static void mlx5e_am_exit_parking(struct mlx5e_rx_am *am)
mlx5e_am_step(am);
}
#define IS_SIGNIFICANT_DIFF(val, ref) \
(((100 * abs((val) - (ref))) / (ref)) > 10) /* more than 10% difference */
static int mlx5e_am_stats_compare(struct mlx5e_rx_am_stats *curr,
struct mlx5e_rx_am_stats *prev)
{
int diff;
if (!prev->ppms)
return curr->ppms ? MLX5E_AM_STATS_BETTER :
if (!prev->bpms)
return curr->bpms ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_SAME;
diff = curr->ppms - prev->ppms;
if (((100 * abs(diff)) / prev->ppms) > 10) /* more than 10% diff */
return (diff > 0) ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_WORSE;
if (IS_SIGNIFICANT_DIFF(curr->bpms, prev->bpms))
return (curr->bpms > prev->bpms) ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_WORSE;
if (!prev->epms)
return curr->epms ? MLX5E_AM_STATS_WORSE :
MLX5E_AM_STATS_SAME;
if (IS_SIGNIFICANT_DIFF(curr->ppms, prev->ppms))
return (curr->ppms > prev->ppms) ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_WORSE;
diff = curr->epms - prev->epms;
if (((100 * abs(diff)) / prev->epms) > 10) /* more than 10% diff */
return (diff < 0) ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_WORSE;
if (IS_SIGNIFICANT_DIFF(curr->epms, prev->epms))
return (curr->epms < prev->epms) ? MLX5E_AM_STATS_BETTER :
MLX5E_AM_STATS_WORSE;
return MLX5E_AM_STATS_SAME;
}
......@@ -266,6 +265,7 @@ static void mlx5e_am_sample(struct mlx5e_rq *rq,
{
s->time = ktime_get();
s->pkt_ctr = rq->stats.packets;
s->byte_ctr = rq->stats.bytes;
s->event_ctr = rq->cq.event_ctr;
}
......@@ -278,12 +278,15 @@ static void mlx5e_am_calc_stats(struct mlx5e_rx_am_sample *start,
/* u32 holds up to 71 minutes, should be enough */
u32 delta_us = ktime_us_delta(end->time, start->time);
unsigned int npkts = end->pkt_ctr - start->pkt_ctr;
unsigned int nbytes = end->byte_ctr - start->byte_ctr;
if (!delta_us)
return;
curr_stats->ppms = (npkts * USEC_PER_MSEC) / delta_us;
curr_stats->epms = (MLX5E_AM_NEVENTS * USEC_PER_MSEC) / delta_us;
curr_stats->ppms = DIV_ROUND_UP(npkts * USEC_PER_MSEC, delta_us);
curr_stats->bpms = DIV_ROUND_UP(nbytes * USEC_PER_MSEC, delta_us);
curr_stats->epms = DIV_ROUND_UP(MLX5E_AM_NEVENTS * USEC_PER_MSEC,
delta_us);
}
void mlx5e_rx_am_work(struct work_struct *work)
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册