From: Konstantin Ananyev <konstantin.ananyev@huawei.com>
To: haijie <haijie1@huawei.com>, "dev@dpdk.org" <dev@dpdk.org>,
"thomas@monjalon.net" <thomas@monjalon.net>,
"ferruh.yigit@amd.com" <ferruh.yigit@amd.com>,
"stephen@networkplumber.org" <stephen@networkplumber.org>,
"lihuisong (C)" <lihuisong@huawei.com>,
Fengchengwen <fengchengwen@huawei.com>,
"Morten Brørup" <mb@smartsharesystems.com>
Cc: huangdengdui <huangdengdui@huawei.com>
Subject: RE: [PATCH v2] examples/l3fwd: add option to set Tx burst size
Date: Fri, 21 Feb 2025 08:45:34 +0000 [thread overview]
Message-ID: <64b206a0090347a8b321928eb21544e7@huawei.com> (raw)
In-Reply-To: <aa08f058-6ae4-f4fa-ab9e-c1563701a144@huawei.com>
> >> On 2024/12/4 10:06, Jie Hai wrote:
> >>> The application send packets only when the buffer is full, or the
> >>> buffer is empty and the packets to be sent extends TX_PKT_BURST.
> >>> The change of MAX_PKT_BURST make TX buffer size and TX_PKT_BURST
> >>> increase, while the default cache size is 256. The packets in
> >>> the TX direction occupy the cache. As a result, the performance
> >>> deteriorates.
> >>>
> >>> Restore the default Tx burst and add option '--tx-burst' to set
> >>> the Tx burst size. To ensure consistency, rename the option
> >>> '--burst' to '--rx-burst'. The valid range of the user-provided
> >>> value is (0, MAX_PKT_BURST] for both directions.
> >>>
> >>> Fixes: d5c4897ecfb2 ("examples/l3fwd: add option to set RX burst size")
> >>> Cc: stable@dpdk.org
> >
> > Wouldn't that patch:
> > https://patchwork.dpdk.org/project/dpdk/patch/20250212045416.2393001-1-sivaprasad.tummala@amd.com/
> > fix that issue too?
> > And probably in a less disruptive manner?
> >
> Thanks for your comment.
>
> I think different drivers may need different values for better
> performance, it makes sense to add options.
Well, we used to run l3fwd for more then 10 years by now,
and I don’t' remember that anyone claim that it is needed.
Again, AFAIR, our other sample apps including test-pmd uses
one burst size for both RX and TX.
Do you have any perf numbers to back-up your claim?
>
> >>> Signed-off-by: Jie Hai <haijie1@huawei.com>
> >>> ---
> >>> examples/l3fwd/l3fwd.h | 14 +++---
> >>> examples/l3fwd/l3fwd_acl.c | 2 +-
> >>> examples/l3fwd/l3fwd_common.h | 2 +-
> >>> examples/l3fwd/l3fwd_em.c | 2 +-
> >>> examples/l3fwd/l3fwd_fib.c | 2 +-
> >>> examples/l3fwd/l3fwd_lpm.c | 2 +-
> >>> examples/l3fwd/main.c | 89 ++++++++++++++++++-----------------
> >>> 7 files changed, 59 insertions(+), 54 deletions(-)
> >>>
> >>> diff --git a/examples/l3fwd/l3fwd.h b/examples/l3fwd/l3fwd.h
> >>> index 0cce3406ee7d..a4e23b817edf 100644
> >>> --- a/examples/l3fwd/l3fwd.h
> >>> +++ b/examples/l3fwd/l3fwd.h
> >>> @@ -32,10 +32,6 @@
> >>>
> >>> #define VECTOR_SIZE_DEFAULT MAX_PKT_BURST
> >>> #define VECTOR_TMO_NS_DEFAULT 1E6 /* 1ms */
> >>> -/*
> >>> - * Try to avoid TX buffering if we have at least MAX_TX_BURST packets to send.
> >>> - */
> >>> -#define MAX_TX_BURST (MAX_PKT_BURST / 2)
> >>>
> >>> #define NB_SOCKETS 8
> >>>
> >>> @@ -116,7 +112,11 @@ extern struct acl_algorithms acl_alg[];
> >>>
> >>> extern uint32_t max_pkt_len;
> >>>
> >>> -extern uint32_t nb_pkt_per_burst;
> >>> +extern uint32_t rx_pkt_burst;
> >>> +/*
> >>> + * Try to avoid TX buffering if we have at least tx_pkt_burst packets to send.
> >>> + */
> >>> +extern uint32_t tx_pkt_burst;
> >>> extern uint32_t mb_mempool_cache_size;
> >>>
> >>> /* Send burst of packets on an output interface */
> >>> @@ -152,8 +152,8 @@ send_single_packet(struct lcore_conf *qconf,
> >>> len++;
> >>>
> >>> /* enough pkts to be sent */
> >>> - if (unlikely(len == MAX_PKT_BURST)) {
> >>> - send_burst(qconf, MAX_PKT_BURST, port);
> >>> + if (unlikely(len == tx_pkt_burst)) {
> >>> + send_burst(qconf, tx_pkt_burst, port);
> >>> len = 0;
> >>> }
> >>>
> >>> diff --git a/examples/l3fwd/l3fwd_acl.c b/examples/l3fwd/l3fwd_acl.c
> >>> index 4fc4b986cce6..a5af82357a03 100644
> >>> --- a/examples/l3fwd/l3fwd_acl.c
> >>> +++ b/examples/l3fwd/l3fwd_acl.c
> >>> @@ -1136,7 +1136,7 @@ acl_main_loop(__rte_unused void *dummy)
> >>> portid = qconf->rx_queue_list[i].port_id;
> >>> queueid = qconf->rx_queue_list[i].queue_id;
> >>> nb_rx = rte_eth_rx_burst(portid, queueid,
> >>> - pkts_burst, nb_pkt_per_burst);
> >>> + pkts_burst, rx_pkt_burst);
> >>>
> >>> if (nb_rx > 0) {
> >>> nb_drop = acl_process_pkts(pkts_burst, hops,
> >>> diff --git a/examples/l3fwd/l3fwd_common.h b/examples/l3fwd/l3fwd_common.h
> >>> index d94e5f135791..34fe70b9415c 100644
> >>> --- a/examples/l3fwd/l3fwd_common.h
> >>> +++ b/examples/l3fwd/l3fwd_common.h
> >>> @@ -71,7 +71,7 @@ send_packetsx4(struct lcore_conf *qconf, uint16_t port, struct rte_mbuf *m[],
> >>> * If TX buffer for that queue is empty, and we have enough packets,
> >>> * then send them straightway.
> >>> */
> >>> - if (num >= MAX_TX_BURST && len == 0) {
> >>> + if (num >= tx_pkt_burst / 2 && len == 0) {
> >>> n = rte_eth_tx_burst(port, qconf->tx_queue_id[port], m, num);
> >>> if (unlikely(n < num)) {
> >>> do {
> >>> diff --git a/examples/l3fwd/l3fwd_em.c b/examples/l3fwd/l3fwd_em.c
> >>> index da9c45e3a482..ea74506ed971 100644
> >>> --- a/examples/l3fwd/l3fwd_em.c
> >>> +++ b/examples/l3fwd/l3fwd_em.c
> >>> @@ -644,7 +644,7 @@ em_main_loop(__rte_unused void *dummy)
> >>> portid = qconf->rx_queue_list[i].port_id;
> >>> queueid = qconf->rx_queue_list[i].queue_id;
> >>> nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
> >>> - nb_pkt_per_burst);
> >>> + rx_pkt_burst);
> >>> if (nb_rx == 0)
> >>> continue;
> >>>
> >>> diff --git a/examples/l3fwd/l3fwd_fib.c b/examples/l3fwd/l3fwd_fib.c
> >>> index 82f1739df778..4223540b30ae 100644
> >>> --- a/examples/l3fwd/l3fwd_fib.c
> >>> +++ b/examples/l3fwd/l3fwd_fib.c
> >>> @@ -239,7 +239,7 @@ fib_main_loop(__rte_unused void *dummy)
> >>> portid = qconf->rx_queue_list[i].port_id;
> >>> queueid = qconf->rx_queue_list[i].queue_id;
> >>> nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
> >>> - nb_pkt_per_burst);
> >>> + rx_pkt_burst);
> >>> if (nb_rx == 0)
> >>> continue;
> >>>
> >>> diff --git a/examples/l3fwd/l3fwd_lpm.c b/examples/l3fwd/l3fwd_lpm.c
> >>> index fec0aeb79c6a..bd1307c43e70 100644
> >>> --- a/examples/l3fwd/l3fwd_lpm.c
> >>> +++ b/examples/l3fwd/l3fwd_lpm.c
> >>> @@ -205,7 +205,7 @@ lpm_main_loop(__rte_unused void *dummy)
> >>> portid = qconf->rx_queue_list[i].port_id;
> >>> queueid = qconf->rx_queue_list[i].queue_id;
> >>> nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
> >>> - nb_pkt_per_burst);
> >>> + rx_pkt_burst);
> >>> if (nb_rx == 0)
> >>> continue;
> >>>
> >>> diff --git a/examples/l3fwd/main.c b/examples/l3fwd/main.c
> >>> index 994b7dd8e596..48453faf7c56 100644
> >>> --- a/examples/l3fwd/main.c
> >>> +++ b/examples/l3fwd/main.c
> >>> @@ -57,7 +57,8 @@
> >>> static_assert(MEMPOOL_CACHE_SIZE >= MAX_PKT_BURST, "MAX_PKT_BURST should be at most MEMPOOL_CACHE_SIZE");
> >>> uint16_t nb_rxd = RX_DESC_DEFAULT;
> >>> uint16_t nb_txd = TX_DESC_DEFAULT;
> >>> -uint32_t nb_pkt_per_burst = DEFAULT_PKT_BURST;
> >>> +uint32_t rx_pkt_burst = DEFAULT_PKT_BURST;
> >>> +uint32_t tx_pkt_burst = DEFAULT_PKT_BURST;
> >>> uint32_t mb_mempool_cache_size = MEMPOOL_CACHE_SIZE;
> >>>
> >>> /**< Ports set in promiscuous mode off by default. */
> >>> @@ -399,7 +400,8 @@ print_usage(const char *prgname)
> >>> " --config (port,queue,lcore)[,(port,queue,lcore)]"
> >>> " [--rx-queue-size NPKTS]"
> >>> " [--tx-queue-size NPKTS]"
> >>> - " [--burst NPKTS]"
> >>> + " [--rx-burst NPKTS]"
> >>> + " [--tx-burst NPKTS]"
> >>> " [--mbcache CACHESZ]"
> >>> " [--eth-dest=X,MM:MM:MM:MM:MM:MM]"
> >>> " [--max-pkt-len PKTLEN]"
> >>> @@ -426,8 +428,10 @@ print_usage(const char *prgname)
> >>> " Default: %d\n"
> >>> " --tx-queue-size NPKTS: Tx queue size in decimal\n"
> >>> " Default: %d\n"
> >>> - " --burst NPKTS: Burst size in decimal\n"
> >>> - " Default: %d\n"
> >>> + " --rx-burst NPKTS: RX Burst size in decimal\n"
> >>> + " Default: %d, Maximum: %d\n"
> >>> + " --tx-burst NPKTS: TX Burst size in decimal\n"
> >>> + " Default: %d, Maximum: %d\n"
> >>> " --mbcache CACHESZ: Mbuf cache size in decimal\n"
> >>> " Default: %d\n"
> >>> " --eth-dest=X,MM:MM:MM:MM:MM:MM: Ethernet destination for port X\n"
> >>> @@ -459,7 +463,10 @@ print_usage(const char *prgname)
> >>> " another is route entry at while line leads with character '%c'.\n"
> >>> " --rule_ipv6=FILE: Specify the ipv6 rules entries file.\n"
> >>> " --alg: ACL classify method to use, one of: %s.\n\n",
> >>> - prgname, RX_DESC_DEFAULT, TX_DESC_DEFAULT, DEFAULT_PKT_BURST, MEMPOOL_CACHE_SIZE,
> >>> + prgname, RX_DESC_DEFAULT, TX_DESC_DEFAULT,
> >>> + DEFAULT_PKT_BURST, MAX_PKT_BURST,
> >>> + DEFAULT_PKT_BURST, MAX_PKT_BURST,
> >>> + MEMPOOL_CACHE_SIZE,
> >>> ACL_LEAD_CHAR, ROUTE_LEAD_CHAR, alg);
> >>> }
> >>>
> >>> @@ -693,47 +700,30 @@ parse_mbcache_size(const char *optarg)
> >>> RTE_MEMPOOL_CACHE_MAX_SIZE);
> >>> }
> >>>
> >>> -static void
> >>> -parse_pkt_burst(const char *optarg)
> >>> +static int
> >>> +parse_pkt_burst(const char *optarg, bool is_rx)
> >>> {
> >>> - struct rte_eth_dev_info dev_info;
> >>> unsigned long pkt_burst;
> >>> - uint16_t burst_size;
> >>> char *end = NULL;
> >>> - int ret;
> >>>
> >>> /* parse decimal string */
> >>> pkt_burst = strtoul(optarg, &end, 10);
> >>> - if ((optarg[0] == '\0') || (end == NULL) || (*end != '\0'))
> >>> - return;
> >>> -
> >>> - if (pkt_burst > MAX_PKT_BURST) {
> >>> - RTE_LOG(INFO, L3FWD, "User provided burst must be <= %d. Using default value %d\n",
> >>> - MAX_PKT_BURST, nb_pkt_per_burst);
> >>> - return;
> >>> - } else if (pkt_burst > 0) {
> >>> - nb_pkt_per_burst = (uint32_t)pkt_burst;
> >>> - return;
> >>> + if ((optarg[0] == '\0') || (end == NULL) || (*end != '\0')) {
> >>> + RTE_LOG(ERR, L3FWD, "Invalid %s-burst\n", is_rx ? "rx" : "tx");
> >>> + return -1;
> >>> }
> >>>
> >>> - /* If user gives a value of zero, query the PMD for its recommended Rx burst size. */
> >>> - ret = rte_eth_dev_info_get(0, &dev_info);
> >>> - if (ret != 0)
> >>> - return;
> >>> - burst_size = dev_info.default_rxportconf.burst_size;
> >>> - if (burst_size == 0) {
> >>> - RTE_LOG(INFO, L3FWD, "PMD does not recommend a burst size. Using default value %d. "
> >>> - "User provided value must be in [1, %d]\n",
> >>> - nb_pkt_per_burst, MAX_PKT_BURST);
> >>> - return;
> >>> - } else if (burst_size > MAX_PKT_BURST) {
> >>> - RTE_LOG(INFO, L3FWD, "PMD recommended burst size %d exceeds maximum value %d. "
> >>> - "Using default value %d\n",
> >>> - burst_size, MAX_PKT_BURST, nb_pkt_per_burst);
> >>> - return;
> >>> + if (pkt_burst > MAX_PKT_BURST || pkt_burst == 0) {
> >>> + RTE_LOG(ERR, L3FWD, "User-provided %s burst must be in (0, %d].\n",
> >>> + is_rx ? "Rx" : "Tx", MAX_PKT_BURST);
> >>> + return -1;
> >>> + } else if (pkt_burst > 0) {
> >>> + if (is_rx)
> >>> + rx_pkt_burst = (uint16_t)pkt_burst;
> >>> + else
> >>> + tx_pkt_burst = (uint16_t)pkt_burst;
> >>> }
> >>> - nb_pkt_per_burst = burst_size;
> >>> - RTE_LOG(INFO, L3FWD, "Using PMD-provided burst value %d\n", burst_size);
> >>> + return 0;
> >>> }
> >>>
> >>> #define MAX_JUMBO_PKT_LEN 9600
> >>> @@ -767,7 +757,8 @@ static const char short_options[] =
> >>> #define CMD_LINE_OPT_RULE_IPV4 "rule_ipv4"
> >>> #define CMD_LINE_OPT_RULE_IPV6 "rule_ipv6"
> >>> #define CMD_LINE_OPT_ALG "alg"
> >>> -#define CMD_LINE_OPT_PKT_BURST "burst"
> >>> +#define CMD_LINE_OPT_PKT_RX_BURST "rx-burst"
> >>> +#define CMD_LINE_OPT_PKT_TX_BURST "tx-burst"
> >>> #define CMD_LINE_OPT_MB_CACHE_SIZE "mbcache"
> >>>
> >>> enum {
> >>> @@ -798,7 +789,8 @@ enum {
> >>> CMD_LINE_OPT_ENABLE_VECTOR_NUM,
> >>> CMD_LINE_OPT_VECTOR_SIZE_NUM,
> >>> CMD_LINE_OPT_VECTOR_TMO_NS_NUM,
> >>> - CMD_LINE_OPT_PKT_BURST_NUM,
> >>> + CMD_LINE_OPT_RX_PKT_BURST_NUM,
> >>> + CMD_LINE_OPT_TX_PKT_BURST_NUM,
> >>> CMD_LINE_OPT_MB_CACHE_SIZE_NUM,
> >>> };
> >>>
> >>> @@ -826,7 +818,8 @@ static const struct option lgopts[] = {
> >>> {CMD_LINE_OPT_RULE_IPV4, 1, 0, CMD_LINE_OPT_RULE_IPV4_NUM},
> >>> {CMD_LINE_OPT_RULE_IPV6, 1, 0, CMD_LINE_OPT_RULE_IPV6_NUM},
> >>> {CMD_LINE_OPT_ALG, 1, 0, CMD_LINE_OPT_ALG_NUM},
> >>> - {CMD_LINE_OPT_PKT_BURST, 1, 0, CMD_LINE_OPT_PKT_BURST_NUM},
> >>> + {CMD_LINE_OPT_PKT_RX_BURST, 1, 0, CMD_LINE_OPT_RX_PKT_BURST_NUM},
> >>> + {CMD_LINE_OPT_PKT_TX_BURST, 1, 0, CMD_LINE_OPT_TX_PKT_BURST_NUM},
> >>> {CMD_LINE_OPT_MB_CACHE_SIZE, 1, 0, CMD_LINE_OPT_MB_CACHE_SIZE_NUM},
> >>> {NULL, 0, 0, 0}
> >>> };
> >>> @@ -916,8 +909,20 @@ parse_args(int argc, char **argv)
> >>> parse_queue_size(optarg, &nb_txd, 0);
> >>> break;
> >>>
> >>> - case CMD_LINE_OPT_PKT_BURST_NUM:
> >>> - parse_pkt_burst(optarg);
> >>> + case CMD_LINE_OPT_TX_PKT_BURST_NUM:
> >>> + ret = parse_pkt_burst(optarg, false);
> >>> + if (ret) {
> >>> + print_usage(prgname);
> >>> + return -1;
> >>> + }
> >>> + break;
> >>> +
> >>> + case CMD_LINE_OPT_RX_PKT_BURST_NUM:
> >>> + ret = parse_pkt_burst(optarg, true);
> >>> + if (ret) {
> >>> + print_usage(prgname);
> >>> + return -1;
> >>> + }
> >>> break;
> >>>
> >>> case CMD_LINE_OPT_MB_CACHE_SIZE_NUM:
next prev parent reply other threads:[~2025-02-21 8:45 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-11-22 7:13 [PATCH] examples/l3fwd: fix Tx performance deteriorate Jie Hai
2024-11-22 9:14 ` lihuisong (C)
2024-11-22 11:01 ` Konstantin Ananyev
2024-11-26 7:39 ` Jie Hai
2024-11-25 19:44 ` Stephen Hemminger
2024-12-04 2:06 ` [PATCH v2] examples/l3fwd: add option to set Tx burst size Jie Hai
2025-02-14 2:11 ` Jie Hai
2025-02-19 17:01 ` Konstantin Ananyev
2025-02-21 3:42 ` Jie Hai
2025-02-21 8:45 ` Konstantin Ananyev [this message]
2025-02-14 8:35 ` fengchengwen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=64b206a0090347a8b321928eb21544e7@huawei.com \
--to=konstantin.ananyev@huawei.com \
--cc=dev@dpdk.org \
--cc=fengchengwen@huawei.com \
--cc=ferruh.yigit@amd.com \
--cc=haijie1@huawei.com \
--cc=huangdengdui@huawei.com \
--cc=lihuisong@huawei.com \
--cc=mb@smartsharesystems.com \
--cc=stephen@networkplumber.org \
--cc=thomas@monjalon.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).