Re: [dpdk-dev] [PATCH v2 1/7] rte_sched: make RED optional at runtime

["Dumitrescu, Cristian" <cristian.dumitrescu@intel.com>]

> -----Original Message-----
> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Stephen
> Hemminger
> Sent: Thursday, February 5, 2015 6:05 AM
> To: dev@dpdk.org
> Cc: Stephen Hemminger
> Subject: [dpdk-dev] [PATCH v2 1/7] rte_sched: make RED optional at runtime
> 
> From: Stephen Hemminger <shemming@brocade.com>
> 
> Want to be able to build with RTE_SCHED_RED enabled but
> allow disabling RED on a per-queue basis at runtime.
> 
> RED is disabled unless min/max thresholds set.
> 
> Signed-off-by: Stephen Hemmminger <stephen@networkplumber.org>
> ---
>  lib/librte_sched/rte_sched.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/lib/librte_sched/rte_sched.c b/lib/librte_sched/rte_sched.c
> index 95dee27..6928c98 100644
> --- a/lib/librte_sched/rte_sched.c
> +++ b/lib/librte_sched/rte_sched.c
> @@ -636,6 +636,12 @@ rte_sched_port_config(struct
> rte_sched_port_params *params)
>  		uint32_t j;
> 
>  		for (j = 0; j < e_RTE_METER_COLORS; j++) {
> +			/* if min/max are both zero, then RED is disabled */
> +			if ((params->red_params[i][j].min_th |
> +			     params->red_params[i][j].max_th) == 0) {
> +				continue;
> +			}
> +
>  			if (rte_red_config_init(&port->red_config[i][j],
>  				params->red_params[i][j].wq_log2,
>  				params->red_params[i][j].min_th,
> @@ -1069,6 +1075,9 @@ rte_sched_port_red_drop(struct rte_sched_port
> *port, struct rte_mbuf *pkt, uint3
>  	color = rte_sched_port_pkt_read_color(pkt);
>  	red_cfg = &port->red_config[tc_index][color];
> 
> +	if ( (red_cfg->min_th | red_cfg->max_th) == 0)
> +		return 0;
> +
>  	qe = port->queue_extra + qindex;
>  	red = &qe->red;
> 
> --
> 2.1.4

Acked by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>

I agree with this approach of leaving RED as a library option that can be enabled/disabled at build-time in order to avoid any performance penalty when RED is not required.

On enqueue side (when RED is enabled):
-Makes sense to avoid the RED processing when RED is not enabled for the current queue;

On dequeue side (when RED is enabled):
-Not sure what's best when RED is not enabled for the current queue: whether to go ahead anyway and apply the timestamp, as it is harmless (but might trigger a cache miss) or check (which might trigger cache miss). Makes sense to go with the former option, which is implemented by this patch.

--------------------------------------------------------------
Intel Shannon Limited
Registered in Ireland
Registered Office: Collinstown Industrial Park, Leixlip, County Kildare
Registered Number: 308263
Business address: Dromore House, East Park, Shannon, Co. Clare

This e-mail and any attachments may contain confidential material for the sole use of the intended recipient(s). Any review or distribution by others is strictly prohibited. If you are not the intended recipient, please contact the sender and delete all copies.