values in rte_eth_rxconf and rte_eth_txconf are not optimal

[Yi-y-yang]

In dpdk/vr_dpdk_ethdev.c, values in rte_eth_rxconf and rte_eth_txconf are not optimal.

Here are current values:

/* RX ring configuration */
static const struct rte_eth_rxconf rx_queue_conf = {
    .rx_thresh = {
        .pthresh = 8, /* Ring prefetch threshold */
        .hthresh = 8, /* Ring host threshold */
        .wthresh = 4, /* Ring writeback threshold */
    },
    /* Do not immediately free RX descriptors */
    .rx_free_thresh = VR_DPDK_RX_BURST_SZ, //32
};

/* TX ring configuration */
static const struct rte_eth_txconf tx_queue_conf = {
    .tx_thresh = {
        .pthresh = 36, /* Ring prefetch threshold */
        .hthresh = 0, /* Ring host threshold */
        .wthresh = 0, /* Ring writeback threshold */
    },
    .tx_free_thresh = 0,
    .tx_rs_thresh = 0, /* Use PMD default values */
    .txq_flags = 0 /* Set flags for the Tx queue */
};

I got some optimal values from dpdk source code test/test/test_pmd_perf.c

/*
 * RX and TX Prefetch, Host, and Write-back threshold values should be
 * carefully set for optimal performance. Consult the network
 * controller's datasheet and supporting DPDK documentation for guidance
 * on how these parameters should be set.
 */
#define RX_PTHRESH 8 /**< Default values of RX prefetch threshold reg. */
#define RX_HTHRESH 8 /**< Default values of RX host threshold reg. */
#define RX_WTHRESH 0 /**< Default values of RX write-back threshold reg. */

/*
 * These default values are optimized for use with the Intel(R) 82599 10 GbE
 * Controller and the DPDK ixgbe PMD. Consider using other values for other
 * network controllers and/or network drivers.
 */
#define TX_PTHRESH 32 /**< Default values of TX prefetch threshold reg. */
#define TX_HTHRESH 0 /**< Default values of TX host threshold reg. */
#define TX_WTHRESH 0 /**< Default values of TX write-back threshold reg. */

static struct rte_eth_rxconf rx_conf = {
        .rx_thresh = {
                .pthresh = RX_PTHRESH,
                .hthresh = RX_HTHRESH,
                .wthresh = RX_WTHRESH,
        },
        .rx_free_thresh = 32,
};

static struct rte_eth_txconf tx_conf = {
        .tx_thresh = {
                .pthresh = TX_PTHRESH,
                .hthresh = TX_HTHRESH,
                .wthresh = TX_WTHRESH,
        },
        .tx_free_thresh = 32, /* Use PMD default values */
        .tx_rs_thresh = 32, /* Use PMD default values */
        .txq_flags = (ETH_TXQ_FLAGS_NOMULTSEGS |
                      ETH_TXQ_FLAGS_NOVLANOFFL |
                      ETH_TXQ_FLAGS_NOXSUMSCTP |
                      ETH_TXQ_FLAGS_NOXSUMUDP |
                      ETH_TXQ_FLAGS_NOXSUMTCP)
};

I tried these values, they are obviously better than previous ones.

[OpenContrail Admin (ci-admin-f)]

Review in progress for https://review.opencontrail.org/45580
Submitter: Yi Yang (<email address hidden>)

[OpenContrail Admin (ci-admin-f)]

Reviewed: https://review.opencontrail.org/45580
Committed: http://github.com/Juniper/contrail-vrouter/commit/9462ef00336aa58a1c46543f702ee89120072aff
Submitter: Zuul v3 CI (<email address hidden>)
Branch: master

commit 9462ef00336aa58a1c46543f702ee89120072aff
Author: Yi Yang <email address hidden>
Date: Wed Aug 15 16:38:24 2018 +0800

Use optimal Tx config for physical ethenet NIC

struct rte_eth_txconf tx_queue_conf = {
.tx_thresh = {
.pthresh = 32, /* Ring prefetch threshold */
.hthresh = 0, /* Ring host threshold */
.wthresh = 0, /* Ring writeback threshold */
},
.tx_free_thresh = 32,
.tx_rs_thresh = 32, /* Use PMD default values */
.txq_flags = 0 /* Set flags for the Tx queue */
};

This can improve Tx performance obviously.

Change-Id: I63ad2cf61e00255dc8e2f3a3e9caa4c1653ec23c
Closes-Bug: #1787114
Signed-off-by: Yi Yang <email address hidden>