From owner-svn-src-all@FreeBSD.ORG Thu Oct 15 15:50:51 2009 Return-Path: Delivered-To: svn-src-all@freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:4f8:fff6::34]) by hub.freebsd.org (Postfix) with ESMTP id 3C96E106568B; Thu, 15 Oct 2009 15:50:51 +0000 (UTC) (envelope-from rwatson@FreeBSD.org) Received: from cyrus.watson.org (cyrus.watson.org [65.122.17.42]) by mx1.freebsd.org (Postfix) with ESMTP id E6E318FC1B; Thu, 15 Oct 2009 15:50:50 +0000 (UTC) Received: from fledge.watson.org (fledge.watson.org [65.122.17.41]) by cyrus.watson.org (Postfix) with ESMTPS id 827F746B03; Thu, 15 Oct 2009 11:50:50 -0400 (EDT) Date: Thu, 15 Oct 2009 16:50:50 +0100 (BST) From: Robert Watson X-X-Sender: robert@fledge.watson.org To: Luigi Rizzo In-Reply-To: <200910151530.n9FFUfxI077369@svn.freebsd.org> Message-ID: References: <200910151530.n9FFUfxI077369@svn.freebsd.org> User-Agent: Alpine 2.00 (BSF 1167 2008-08-23) MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed Cc: svn-src-head@freebsd.org, svn-src-all@freebsd.org, src-committers@freebsd.org Subject: Re: svn commit: r198136 - head/tools/tools/netrate/netsend X-BeenThere: svn-src-all@freebsd.org X-Mailman-Version: 2.1.5 Precedence: list List-Id: "SVN commit messages for the entire src tree \(except for " user" and " projects" \)" List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 15 Oct 2009 15:50:51 -0000 On Thu, 15 Oct 2009, Luigi Rizzo wrote: > Support the specification of a range of destination ports e.g. > > netsend 127.0.0.1 6666-7777 [payloadsize] [packet_rate] [duration] All sounds very useful -- consider making some of these changes (where appropriate) to netblast as well? Also, I can't remember if netsend / netblast know about calling bind() before commencing, but if not they should be taught. That should (a) eliminate significant overhead and (b) use weaker locking so that greater concurrency can be exploited by multiple instances. The connect(2) case should force binding, but in absence of connect(2), a per-packet "source address selection" decision needs to be made. Robert > > This is useful to test the behaviour of systems that do some kind > of flow classifications and so exhibit different behaviour depending > on the number of flows that hit them. > I plan to add a similar extension to sweep on a range of IP addresses, > so we can issue a single command to flood (obviously, for testing > purposes!) a number of different destinations. > > When there is only one destination, we do a preliminary connect() > of the socket so we can use send() instead of sendto(). > When we have multiple ports, the socket is not connect()'ed and we > do a sendto() instead. There is a performance hit in this case, > as the throughput on the loopback interface (with a firewall rule > that blocks the transmission) goes down from 900kpps to 490kpps on > my test machine. > > If the number of different destinations is limited, one option to > explore is to have multiple connect()ed sockets. > > MFC after: 1 month > > Modified: > head/tools/tools/netrate/netsend/netsend.c > > Modified: head/tools/tools/netrate/netsend/netsend.c > ============================================================================== > --- head/tools/tools/netrate/netsend/netsend.c Thu Oct 15 14:55:11 2009 (r198135) > +++ head/tools/tools/netrate/netsend/netsend.c Thu Oct 15 15:30:41 2009 (r198136) > @@ -39,12 +39,23 @@ > #include > #include > > +/* program arguments */ > +struct _a { > + int s; > + struct timespec interval; > + int port, port_max; > + long duration; > + struct sockaddr_in sin; > + int packet_len; > + void *packet; > +}; > + > static void > usage(void) > { > > fprintf(stderr, > - "netsend [ip] [port] [payloadsize] [rate] [duration]\n"); > + "netsend [ip] [port[-port_max]] [payloadsize] [packet_rate] [duration]\n"); > exit(-1); > } > > @@ -114,10 +125,12 @@ wait_time(struct timespec ts, struct tim > * Calculate a second-aligned starting time for the packet stream. Busy > * wait between our calculated interval and dropping the provided packet > * into the socket. If we hit our duration limit, bail. > + * We sweep the ports from a->port to a->port_max included. > + * If the two ports are the same we connect() the socket upfront, which > + * almost halves the cost of the sendto() call. > */ > static int > -timing_loop(int s, struct timespec interval, long duration, u_char *packet, > - u_int packet_len) > +timing_loop(struct _a *a) > { > struct timespec nexttime, starttime, tmptime; > long long waited; > @@ -127,18 +140,19 @@ timing_loop(int s, struct timespec inter > /* do not call gettimeofday more than every 20us */ > long minres_ns = 20000; > int ic, gettimeofday_cycles; > + int cur_port; > > if (clock_getres(CLOCK_REALTIME, &tmptime) == -1) { > perror("clock_getres"); > return (-1); > } > > - if (timespec_ge(&tmptime, &interval)) > + if (timespec_ge(&tmptime, &a->interval)) > fprintf(stderr, > "warning: interval (%jd.%09ld) less than resolution (%jd.%09ld)\n", > - (intmax_t)interval.tv_sec, interval.tv_nsec, > + (intmax_t)a->interval.tv_sec, a->interval.tv_nsec, > (intmax_t)tmptime.tv_sec, tmptime.tv_nsec); > - if (tmptime.tv_nsec < minres_ns) { > + if (a->interval.tv_nsec < minres_ns) { > gettimeofday_cycles = minres_ns/(tmptime.tv_nsec + 1); > fprintf(stderr, > "calling time every %d cycles\n", gettimeofday_cycles); > @@ -156,14 +170,23 @@ timing_loop(int s, struct timespec inter > if (wait_time(starttime, NULL, NULL) == -1) > return (-1); > nexttime = starttime; > - finishtime = starttime.tv_sec + duration; > + finishtime = starttime.tv_sec + a->duration; > > send_errors = send_calls = 0; > counter = 0; > waited = 0; > ic = gettimeofday_cycles; > + cur_port = a->port; > + if (a->port == a->port_max) { > + if (connect(a->s, (struct sockaddr *)&a->sin, sizeof(a->sin))) { > + perror("connect"); > + return (-1); > + } > + } > while (1) { > - timespec_add(&nexttime, &interval); > + int ret; > + > + timespec_add(&nexttime, &a->interval); > if (--ic <= 0) { > ic = gettimeofday_cycles; > if (wait_time(nexttime, &tmptime, &waited) == -1) > @@ -178,17 +201,28 @@ timing_loop(int s, struct timespec inter > * previous send, the error will turn up the current send > * operation, causing the current sequence number also to be > * skipped. > + * The counter is incremented only on the initial port number, > + * so all destinations will see the same set of packets. > * > * XXXRW: Note alignment assumption. > */ > - if (packet_len >= 4) { > - *((u_int32_t *)packet) = htonl(counter); > + if (cur_port == a->port && a->packet_len >= 4) { > + *((u_int32_t *)a->packet) = htonl(counter); > counter++; > } > - if (send(s, packet, packet_len, 0) < 0) > + if (a->port == a->port_max) { /* socket already bound */ > + ret = send(a->s, a->packet, a->packet_len, 0); > + } else { > + a->sin.sin_port = htons(cur_port++); > + if (cur_port > a->port_max) > + cur_port = a->port; > + ret = sendto(a->s, a->packet, a->packet_len, 0, > + (struct sockaddr *)&a->sin, sizeof(a->sin)); > + } > + if (ret < 0) > send_errors++; > send_calls++; > - if (duration != 0 && tmptime.tv_sec >= finishtime) > + if (a->duration != 0 && tmptime.tv_sec >= finishtime) > goto done; > } > > @@ -205,11 +239,11 @@ done: > tmptime.tv_nsec); > printf("send calls: %ld\n", send_calls); > printf("send errors: %ld\n", send_errors); > - printf("approx send rate: %ld\n", (send_calls - send_errors) / > - duration); > + printf("approx send rate: %ld pps\n", (send_calls - send_errors) / > + a->duration); > printf("approx error rate: %ld\n", (send_errors / send_calls)); > printf("waited: %lld\n", waited); > - printf("approx waits/sec: %lld\n", (long long)(waited / duration)); > + printf("approx waits/sec: %lld\n", (long long)(waited / a->duration)); > printf("approx wait rate: %lld\n", (long long)(waited / send_calls)); > > return (0); > @@ -218,27 +252,35 @@ done: > int > main(int argc, char *argv[]) > { > - long rate, payloadsize, port, duration; > - struct timespec interval; > - struct sockaddr_in sin; > - char *dummy, *packet; > - int s; > + long rate, payloadsize, port; > + char *dummy; > + struct _a a; /* arguments */ > + > + bzero(&a, sizeof(a)); > > if (argc != 6) > usage(); > > - bzero(&sin, sizeof(sin)); > - sin.sin_len = sizeof(sin); > - sin.sin_family = AF_INET; > - if (inet_aton(argv[1], &sin.sin_addr) == 0) { > + a.sin.sin_len = sizeof(a.sin); > + a.sin.sin_family = AF_INET; > + if (inet_aton(argv[1], &a.sin.sin_addr) == 0) { > perror(argv[1]); > return (-1); > } > > port = strtoul(argv[2], &dummy, 10); > - if (port < 1 || port > 65535 || *dummy != '\0') > + if (port < 1 || port > 65535) > usage(); > - sin.sin_port = htons(port); > + if (*dummy != '\0' && *dummy != '-') > + usage(); > + a.sin.sin_port = htons(port); > + a.port = a.port_max = port; > + if (*dummy == '-') { /* set high port */ > + port = strtoul(dummy + 1, &dummy, 10); > + if (port < a.port || port > 65535) > + usage(); > + a.port_max = port; > + } > > payloadsize = strtoul(argv[3], &dummy, 10); > if (payloadsize < 0 || *dummy != '\0') > @@ -247,56 +289,51 @@ main(int argc, char *argv[]) > fprintf(stderr, "payloadsize > 32768\n"); > return (-1); > } > + a.packet_len = payloadsize; > > /* > * Specify an arbitrary limit. It's exactly that, not selected by > * any particular strategy. '0' is a special value meaning "blast", > * and avoids the cost of a timing loop. > - * XXX 0 is not actually implemented. > */ > rate = strtoul(argv[4], &dummy, 10); > - if (rate < 1 || *dummy != '\0') > + if (rate < 0 || *dummy != '\0') > usage(); > if (rate > MAX_RATE) { > - fprintf(stderr, "rate > %d\n", MAX_RATE); > + fprintf(stderr, "packet rate at most %d\n", MAX_RATE); > return (-1); > } > > - duration = strtoul(argv[5], &dummy, 10); > - if (duration < 0 || *dummy != '\0') > + a.duration = strtoul(argv[5], &dummy, 10); > + if (a.duration < 0 || *dummy != '\0') > usage(); > > - packet = malloc(payloadsize); > - if (packet == NULL) { > + a.packet = malloc(payloadsize); > + if (a.packet == NULL) { > perror("malloc"); > return (-1); > } > - bzero(packet, payloadsize); > - > + bzero(a.packet, payloadsize); > if (rate == 0) { > - interval.tv_sec = 0; > - interval.tv_nsec = 0; > + a.interval.tv_sec = 0; > + a.interval.tv_nsec = 0; > } else if (rate == 1) { > - interval.tv_sec = 1; > - interval.tv_nsec = 0; > + a.interval.tv_sec = 1; > + a.interval.tv_nsec = 0; > } else { > - interval.tv_sec = 0; > - interval.tv_nsec = ((1 * 1000000000) / rate); > + a.interval.tv_sec = 0; > + a.interval.tv_nsec = ((1 * 1000000000) / rate); > } > + > printf("Sending packet of payload size %ld every %jd.%09lds for %ld " > - "seconds\n", payloadsize, (intmax_t)interval.tv_sec, > - interval.tv_nsec, duration); > + "seconds\n", payloadsize, (intmax_t)a.interval.tv_sec, > + a.interval.tv_nsec, a.duration); > > - s = socket(PF_INET, SOCK_DGRAM, 0); > - if (s == -1) { > + a.s = socket(PF_INET, SOCK_DGRAM, 0); > + if (a.s == -1) { > perror("socket"); > return (-1); > } > > - if (connect(s, (struct sockaddr *)&sin, sizeof(sin)) < 0) { > - perror("connect"); > - return (-1); > - } > - > - return (timing_loop(s, interval, duration, packet, payloadsize)); > + return (timing_loop(&a)); > } >