public inbox for netdev@vger.kernel.org
 help / color / mirror / Atom feed
From: Eric Dumazet <eric.dumazet@gmail.com>
To: hadi@cyberus.ca
Cc: Changli Gao <xiaosuo@gmail.com>,
	David Miller <davem@davemloft.net>,
	therbert@google.com, shemminger@vyatta.com,
	netdev@vger.kernel.org, Eilon Greenstein <eilong@broadcom.com>,
	Brian Bloniarz <bmb@athenacr.com>
Subject: Re: [PATCH net-next-2.6] net: speedup udp receive path
Date: Sat, 01 May 2010 07:57:04 +0200	[thread overview]
Message-ID: <1272693424.2230.75.camel@edumazet-laptop> (raw)
In-Reply-To: <1272672394.14499.1.camel@bigi>

Le vendredi 30 avril 2010 à 20:06 -0400, jamal a écrit :

> Yes, Nehalem. 
> RPS off is better (~700Kpp) than RPS on(~650kpps). Are you seeing the
> same trend on the old hardware?
> 

Of course not ! Or else RPS would be useless :(

I changed your program a bit to use EV_PERSIST, (to avoid epoll_ctl()
overhead for each packet...)

RPS off : 220.000 pps 

RPS on (ee mask) : 700.000 pps  (with a slightly modified tg3 driver)
96% of delivered packets

This is on tg3 adapter, and tg3 has copybreak feature : small packets
are copied into skb of the right size.

define TG3_RX_COPY_THRESHOLD       256 -> 40 ...

We really should disable this feature for RPS workload,
unfortunatly ethtool cannot tweak this.

So profile of cpu 0 (RPS ON) looks like :

------------------------------------------------------------------------------------------------------------------------
   PerfTop:    1001 irqs/sec  kernel:99.7% [1000Hz cycles],  (all, cpu: 0)
------------------------------------------------------------------------------------------------------------------------

             samples  pcnt function               DSO
             _______ _____ ______________________ _______

              819.00 12.6% __alloc_skb            vmlinux
              592.00  9.1% eth_type_trans         vmlinux
              509.00  7.8% _raw_spin_lock         vmlinux
              475.00  7.3% __kmalloc_track_caller vmlinux
              358.00  5.5% tg3_read32             vmlinux
              345.00  5.3% __netdev_alloc_skb     vmlinux
              329.00  5.0% kmem_cache_alloc       vmlinux
              307.00  4.7% _raw_spin_lock_irqsave vmlinux
              284.00  4.4% bnx2_interrupt         vmlinux
              277.00  4.2% skb_pull               vmlinux
              248.00  3.8% tg3_poll_work          vmlinux
              202.00  3.1% __slab_alloc           vmlinux
              197.00  3.0% get_rps_cpu            vmlinux
              106.00  1.6% enqueue_to_backlog     vmlinux
               87.00  1.3% _raw_spin_lock_bh      vmlinux
               80.00  1.2% __copy_to_user_ll      vmlinux
               77.00  1.2% nommu_map_page         vmlinux
               77.00  1.2% __napi_gro_receive     vmlinux
               65.00  1.0% tg3_alloc_rx_skb       vmlinux
               60.00  0.9% skb_gro_reset_offset   vmlinux
               57.00  0.9% skb_put                vmlinux
               57.00  0.9% __slab_free            vmlinux


/*
 *  Usage: udpsnkfrk [ -p baseport] nbports
*/
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <string.h>
#include <stdio.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h>
#include <event.h>

struct worker_data {
	struct event *snk_ev;
	struct event_base *base;
	struct timeval t;
	unsigned long pack_count;
	unsigned long bytes_count;
	unsigned long tout;
	int fd;			/* move to avoid hole on 64-bit */
	int pad1;	
	unsigned long _padd[99]; /* avoid false sharing */
};

void usage(int code)
{
	fprintf(stderr, "Usage: udpsink [-p baseport] nbports\n");
	exit(code);
}

void process_recv(int fd, short ev, void *arg)
{
	char buffer[4096];
	struct sockaddr_in addr;
	socklen_t len = sizeof(addr);
	struct worker_data *wdata = (struct worker_data *)arg;
	int lu = 0;


	if (ev == EV_TIMEOUT) {
		wdata->tout++;
		if ((event_add(wdata->snk_ev, &wdata->t)) < 0) {
			perror("cb event_add");
			return;
		}
	} else {
		do {
			lu = recvfrom(wdata->fd, buffer, sizeof(buffer), 0,
			      (struct sockaddr *)&addr, &len);
			if (lu > 0) {
				wdata->pack_count++;
				wdata->bytes_count += lu;
			}
		} while (lu > 0);
	}
}

int prep_thread(struct worker_data *wdata)
{
	wdata->t.tv_sec = 1;
	wdata->t.tv_usec = random() % 50000L;

	wdata->base = event_init();
	event_set(wdata->snk_ev, wdata->fd, EV_READ|EV_PERSIST, process_recv, wdata);
	event_base_set(wdata->base, wdata->snk_ev);
	if ((event_add(wdata->snk_ev, &wdata->t)) < 0) {
		perror("event_add");
		return -1;
	}
	return 0;
}

void *worker_func(void *arg)
{
	struct worker_data *wdata = (struct worker_data *)arg;

	return (void *)event_base_loop(wdata->base, 0);
}

int main(int argc, char *argv[])
{
	int c;
	int baseport = 4000;
	int nbthreads;
	struct worker_data *wdata;
	unsigned long ototal = 0;
	int concurrent = 0;
	int verbose = 0;
	int i;
	while ((c = getopt(argc, argv, "cvp:")) != -1) {
		if (c == 'p')
			baseport = atoi(optarg);
		else if (c == 'c')
			concurrent = 1;
		else if (c == 'v')
			verbose++;
		else
			usage(1);
	}
	if (optind == argc)
		usage(1);
	nbthreads = atoi(argv[optind]);
	wdata = calloc(sizeof(struct worker_data), nbthreads);
	if (!wdata) {
		perror("calloc");
		return 1;
	}

	for (i = 0; i < nbthreads; i++) {
		struct sockaddr_in addr;
		pthread_t tid;

		if (i && concurrent) {
			wdata[i].fd = wdata[0].fd;
		} else {
			wdata[i].snk_ev = malloc(sizeof(struct event));
			if (!wdata[i].snk_ev)
				return 1;
			memset(wdata[i].snk_ev, 0, sizeof(struct event));

			wdata[i].fd = socket(PF_INET, SOCK_DGRAM, 0);
			if (wdata[i].fd == -1) {
				free(wdata[i].snk_ev);
				perror("socket");
				return 1;
			}
			memset(&addr, 0, sizeof(addr));
			addr.sin_family = AF_INET;
//                      addr.sin_addr.s_addr = inet_addr(argv[optind]);
			addr.sin_port = htons(baseport + i);
			if (bind
			    (wdata[i].fd, (struct sockaddr *)&addr,
			     sizeof(addr)) < 0) {
				free(wdata[i].snk_ev);
				perror("bind");
				return 1;
			}
                      fcntl(wdata[i].fd, F_SETFL, O_NDELAY);
		}
		if (prep_thread(wdata + i)) {
			printf("failed to allocate thread %d, exit\n", i);
			exit(0);
		}
		pthread_create(&tid, NULL, worker_func, wdata + i);
	}

	for (;;) {
		unsigned long total;
		long delta;

		sleep(1);
		total = 0;
		for (i = 0; i < nbthreads; i++) {
			total += wdata[i].pack_count;
		}
		delta = total - ototal;
		if (delta) {
			printf("%lu pps (%lu", delta, total);
			if (verbose) {
				for (i = 0; i < nbthreads; i++) {
					if (wdata[i].pack_count)
						printf(" %d:%lu", i,
						       wdata[i].pack_count);
				}
			}
			printf(")\n");
		}
		ototal = total;
	}
}




  reply	other threads:[~2010-05-01  5:57 UTC|newest]

Thread overview: 108+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2010-04-23  8:12 [PATCH v6] net: batch skb dequeueing from softnet input_pkt_queue Changli Gao
2010-04-23  9:27 ` Eric Dumazet
2010-04-23 22:02   ` jamal
2010-04-24 14:10     ` jamal
2010-04-26 14:03       ` Eric Dumazet
2010-04-26 14:55         ` Eric Dumazet
2010-04-26 21:06           ` jamal
     [not found]           ` <20100429174056.GA8044@gargoyle.fritz.box>
2010-04-29 17:56             ` Eric Dumazet
2010-04-29 18:10               ` OFT - reserving CPU's for networking Stephen Hemminger
2010-04-29 19:19                 ` Thomas Gleixner
2010-04-29 20:02                   ` Eric Dumazet
2010-04-30 18:15                     ` Brian Bloniarz
2010-04-30 18:57                   ` David Miller
2010-04-30 19:58                     ` Thomas Gleixner
2010-04-30 21:01                     ` Andi Kleen
2010-04-30 22:30                       ` David Miller
2010-05-01 10:53                         ` Andi Kleen
2010-05-01 22:03                           ` David Miller
2010-05-01 22:58                             ` Andi Kleen
2010-05-01 23:29                               ` David Miller
2010-05-01 23:44                             ` Ben Hutchings
2010-05-01 20:31                     ` Martin Josefsson
2010-05-01 22:13                       ` David Miller
     [not found]               ` <20100429182347.GA8512@gargoyle.fritz.box>
2010-04-29 19:12                 ` [PATCH v6] net: batch skb dequeueing from softnet input_pkt_queue Eric Dumazet
     [not found]                   ` <20100429214144.GA10663@gargoyle.fritz.box>
2010-04-30  5:25                     ` Eric Dumazet
2010-04-30 23:38                     ` David Miller
2010-05-01 11:00                       ` Andi Kleen
2010-05-02  6:56                         ` Eric Dumazet
2010-05-02  9:20                           ` Andi Kleen
2010-05-02 10:54                             ` Eric Dumazet
2010-05-02 14:13                               ` Arjan van de Ven
2010-05-02 14:27                                 ` Eric Dumazet
2010-05-02 15:32                                   ` Eric Dumazet
2010-05-02 17:54                                   ` Arjan van de Ven
2010-05-02 19:22                                     ` Eric Dumazet
2010-05-02 22:06                                       ` Andi Kleen
2010-05-03  3:50                                       ` Arjan van de Ven
2010-05-03  5:17                                         ` Eric Dumazet
2010-05-03 10:22                                           ` Arjan van de Ven
2010-05-03 10:34                                             ` Andi Kleen
2010-05-03 14:09                                               ` Arjan van de Ven
2010-05-03 14:45                                                 ` Brian Bloniarz
2010-05-04  1:10                                                   ` Arjan van de Ven
2010-05-03 15:52                                                 ` Andi Kleen
2010-05-04  1:11                                                   ` Arjan van de Ven
2010-05-02 21:30                                     ` Andi Kleen
2010-05-02 15:46                               ` Andi Kleen
2010-05-02 16:35                                 ` Eric Dumazet
2010-05-02 17:43                                   ` Arjan van de Ven
2010-05-02 17:47                                     ` Eric Dumazet
2010-05-02 21:25                                   ` Andi Kleen
2010-05-02 21:45                                     ` Eric Dumazet
2010-05-02 21:54                                       ` Andi Kleen
2010-05-02 22:08                                         ` Eric Dumazet
2010-05-03 20:15                                           ` jamal
2010-04-26 21:03         ` jamal
2010-04-23 10:26 ` Eric Dumazet
2010-04-27 22:08   ` David Miller
2010-04-27 22:18     ` [PATCH net-next-2.6] bnx2x: Remove two prefetch() Eric Dumazet
2010-04-27 22:19       ` David Miller
2010-04-28 13:14         ` Eilon Greenstein
2010-04-28 15:44           ` Eliezer Tamir
2010-04-28 16:53           ` David Miller
     [not found]           ` <w2ue8f3c3211004280842r9f2589e8qb8fd4b7933cd9756@mail.gmail.com>
2010-04-28 16:55             ` David Miller
2010-04-28 11:33       ` jamal
2010-04-28 12:33         ` Eric Dumazet
2010-04-28 12:36           ` jamal
2010-04-28 14:06             ` [PATCH net-next-2.6] net: speedup udp receive path Eric Dumazet
2010-04-28 14:19               ` Eric Dumazet
2010-04-28 14:34                 ` Eric Dumazet
2010-04-28 21:36               ` David Miller
2010-04-28 22:22                 ` [PATCH net-next-2.6] net: ip_queue_rcv_skb() helper Eric Dumazet
2010-04-28 22:39                   ` David Miller
2010-04-28 23:44               ` [PATCH net-next-2.6] net: speedup udp receive path jamal
2010-04-29  0:00                 ` jamal
2010-04-29  4:09                 ` Eric Dumazet
2010-04-29 11:35                   ` jamal
2010-04-29 12:12                     ` Changli Gao
2010-04-29 12:45                       ` Eric Dumazet
2010-04-29 13:17                         ` jamal
2010-04-29 13:21                           ` Eric Dumazet
2010-04-29 13:37                             ` jamal
2010-04-29 13:49                               ` Eric Dumazet
2010-04-29 13:56                                 ` jamal
2010-04-29 20:36                                   ` jamal
2010-04-29 21:01                                     ` [PATCH net-next-2.6] net: sock_def_readable() and friends RCU conversion Eric Dumazet
2010-04-30 13:55                                       ` Brian Bloniarz
2010-04-30 17:26                                         ` Eric Dumazet
2010-04-30 23:35                                       ` David Miller
2010-05-01  4:56                                         ` Eric Dumazet
2010-05-01  7:02                                         ` Eric Dumazet
2010-05-01  8:03                                           ` Eric Dumazet
2010-05-01 22:00                                             ` David Miller
2010-04-30 19:30                                     ` [PATCH net-next-2.6] net: speedup udp receive path jamal
2010-04-30 20:40                                       ` Eric Dumazet
2010-05-01  0:06                                         ` jamal
2010-05-01  5:57                                           ` Eric Dumazet [this message]
2010-05-01  6:14                                             ` Eric Dumazet
2010-05-01 10:24                                               ` Changli Gao
2010-05-01 10:47                                                 ` Eric Dumazet
2010-05-01 11:29                                               ` jamal
2010-05-01 11:23                                             ` jamal
2010-05-01 11:42                                               ` Eric Dumazet
2010-05-01 11:56                                                 ` jamal
2010-05-01 13:22                                                   ` Eric Dumazet
2010-05-01 13:49                                                     ` jamal
2010-05-03 20:10                                                   ` jamal
2010-04-29 23:07                         ` Changli Gao

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=1272693424.2230.75.camel@edumazet-laptop \
    --to=eric.dumazet@gmail.com \
    --cc=bmb@athenacr.com \
    --cc=davem@davemloft.net \
    --cc=eilong@broadcom.com \
    --cc=hadi@cyberus.ca \
    --cc=netdev@vger.kernel.org \
    --cc=shemminger@vyatta.com \
    --cc=therbert@google.com \
    --cc=xiaosuo@gmail.com \
    /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