2 * Copyright (C) 2016 Universita` di Pisa. All rights reserved.
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * This program implements NMREPLAY, a program to replay a pcap file
31 * enforcing the output rate and possibly random losses and delay
33 * It is meant to be run from the command line and implemented with a main
34 * control thread for monitoring, plus a thread to push packets out.
36 * The control thread parses command line arguments, prepares a
37 * schedule for transmission in a memory buffer and then sits
38 * in a loop where it periodically reads traffic statistics from
39 * the other threads and prints them out on the console.
41 * The transmit buffer contains headers and packets. Each header
42 * includes a timestamp that determines when the packet should be sent out.
43 * A "consumer" thread cons() reads from the queue and transmits packets
44 * on the output netmap port when their time has come.
46 * The program does CPU pinning and sets the scheduler and priority
47 * for the "cons" threads. Externally one should do the
48 * assignment of other threads (e.g. interrupt handlers) and
49 * make sure that network interfaces are configured properly.
51 * --- Main functions of the program ---
52 * within each function, q is used as a pointer to the queue holding
53 * packets and parameters.
57 * reads from the pcap file and prepares packets to transmit.
58 * After reading a packet from the pcap file, the following information
59 * are extracted which can be used to determine the schedule:
61 * q->cur_pkt points to the buffer containing the packet
62 * q->cur_len packet length, excluding CRC
63 * q->cur_caplen available packet length (may be shorter than cur_len)
64 * q->cur_tt transmission time for the packet, computed from the trace.
66 * The following functions are then called in sequence:
68 * q->c_loss (set with the -L command line option) decides
69 * whether the packet should be dropped before even queuing.
70 * This is generally useful to emulate random loss.
71 * The function is supposed to set q->c_drop = 1 if the
72 * packet should be dropped, or leave it to 0 otherwise.
74 * q->c_bw (set with the -B command line option) is used to
75 * enforce the transmit bandwidth. The function must store
76 * in q->cur_tt the transmission time (in nanoseconds) of
77 * the packet, which is typically proportional to the length
78 * of the packet, i.e. q->cur_tt = q->cur_len / <bandwidth>
79 * Variants are possible, eg. to account for constant framing
80 * bits as on the ethernet, or variable channel acquisition times,
82 * This mechanism can also be used to simulate variable queueing
83 * delay e.g. due to the presence of cross traffic.
85 * q->c_delay (set with the -D option) implements delay emulation.
86 * The function should set q->cur_delay to the additional
87 * delay the packet is subject to. The framework will take care of
88 * computing the actual exit time of a packet so that there is no
93 #define NED(_fmt, ...) do {} while (0)
94 #define ED(_fmt, ...) \
97 gettimeofday(&_t0, NULL); \
98 fprintf(stderr, "%03d.%03d %-10.10s [%5d] \t" _fmt "\n", \
99 (int)(_t0.tv_sec % 1000), (int)_t0.tv_usec/1000, \
100 __FUNCTION__, __LINE__, ##__VA_ARGS__); \
103 /* WWW is for warnings, EEE is for errors */
104 #define WWW(_fmt, ...) ED("--WWW-- " _fmt, ##__VA_ARGS__)
105 #define EEE(_fmt, ...) ED("--EEE-- " _fmt, ##__VA_ARGS__)
106 #define DDD(_fmt, ...) ED("--DDD-- " _fmt, ##__VA_ARGS__)
108 #define _GNU_SOURCE // for CPU_SET() etc
110 #define NETMAP_WITH_LIBS
111 #include <net/netmap_user.h>
112 #include <sys/poll.h>
117 * A packet in the queue is q_pkt plus the payload.
119 * For the packet descriptor we need the following:
121 * - position of next packet in the queue (can go backwards).
122 * We can reduce to 32 bits if we consider alignments,
123 * or we just store the length to be added to the current
124 * value and assume 0 as a special index.
125 * - actual packet length (16 bits may be ok)
126 * - queue output time, in nanoseconds (64 bits)
127 * - delay line output time, in nanoseconds
128 * One of the two can be packed to a 32bit value
130 * A convenient coding uses 32 bytes per packet.
134 uint64_t next; /* buffer index for next packet */
135 uint64_t pktlen; /* actual packet len */
136 uint64_t pt_qout; /* time of output from queue */
137 uint64_t pt_tx; /* transmit time */
142 * The header for a pcap file
144 struct pcap_file_header {
146 /*used to detect the file format itself and the byte
147 ordering. The writing application writes 0xa1b2c3d4 with it's native byte
148 ordering format into this field. The reading application will read either
149 0xa1b2c3d4 (identical) or 0xd4c3b2a1 (swapped). If the reading application
150 reads the swapped 0xd4c3b2a1 value, it knows that all the following fields
151 will have to be swapped too. For nanosecond-resolution files, the writing
152 application writes 0xa1b23c4d, with the two nibbles of the two lower-order
153 bytes swapped, and the reading application will read either 0xa1b23c4d
154 (identical) or 0x4d3cb2a1 (swapped)*/
155 uint16_t version_major;
156 uint16_t version_minor; /*the version number of this file format */
158 /*the correction time in seconds between GMT (UTC) and the
159 local timezone of the following packet header timestamps. Examples: If the
160 timestamps are in GMT (UTC), thiszone is simply 0. If the timestamps are in
161 Central European time (Amsterdam, Berlin, ...) which is GMT + 1:00, thiszone
163 uint32_t stampacc; /*the accuracy of time stamps in the capture*/
165 /*the "snapshot length" for the capture (typically 65535
166 or even more, but might be limited by the user)*/
168 /*link-layer header type, specifying the type of headers
169 at the beginning of the packet (e.g. 1 for Ethernet); this can be various
170 types such as 802.11, 802.11 with various radio information, PPP, Token
174 #if 0 /* from pcap.h */
175 struct pcap_file_header {
177 u_short version_major;
178 u_short version_minor;
179 bpf_int32 thiszone; /* gmt to local correction */
180 bpf_u_int32 sigfigs; /* accuracy of timestamps */
181 bpf_u_int32 snaplen; /* max length saved portion of each pkt */
182 bpf_u_int32 linktype; /* data link type (LINKTYPE_*) */
186 struct timeval ts; /* time stamp */
187 bpf_u_int32 caplen; /* length of portion present */
188 bpf_u_int32 len; /* length this packet (off wire) */
190 #endif /* from pcap.h */
193 uint32_t ts_sec; /* seconds from epoch */
194 uint32_t ts_frac; /* microseconds or nanoseconds depending on sigfigs */
196 /*the number of bytes of packet data actually captured
197 and saved in the file. This value should never become larger than orig_len
198 or the snaplen value of the global header*/
199 uint32_t len; /* wire length */
203 #define PKT_PAD (32) /* padding on packets */
205 static inline int pad(int x)
207 return ((x) + PKT_PAD - 1) & ~(PKT_PAD - 1) ;
213 * wrapper around the pcap file.
214 * We mmap the file so it is easy to do multiple passes through it.
216 struct nm_pcap_file {
219 const char *data; /* mmapped file */
223 uint64_t tot_bytes_rounded; /* need hdr + pad(len) */
224 uint32_t resolution; /* 1000 for us, 1 for ns */
225 int swap; /* need to swap fields ? */
228 uint64_t total_tx_time;
230 * total_tx_time is computed as last_ts - first_ts, plus the
231 * transmission time for the first packet which in turn is
232 * computed according to the average bandwidth
236 const char *cur; /* running pointer */
237 const char *lim; /* data + file_len */
241 static struct nm_pcap_file *readpcap(const char *fn);
242 static void destroy_pcap(struct nm_pcap_file *file);
250 #include <string.h> /* memcpy */
252 #include <sys/mman.h>
254 #define NS_SCALE 1000000000UL /* nanoseconds in 1s */
256 static void destroy_pcap(struct nm_pcap_file *pf)
261 munmap((void *)(uintptr_t)pf->data, pf->filesize);
263 bzero(pf, sizeof(*pf));
268 // convert a field of given size if swap is needed.
270 cvt(const void *src, int size, char swap)
273 if (size != 2 && size != 4) {
274 EEE("Invalid size %d\n", size);
277 memcpy(&ret, src, size);
279 unsigned char tmp, *data = (unsigned char *)&ret;
281 for (i = 0; i < size / 2; i++) {
283 data[i] = data[size - (1 + i)];
284 data[size - (1 + i)] = tmp;
291 read_next_info(struct nm_pcap_file *pf, int size)
293 const char *end = pf->cur + size;
299 ret = cvt(pf->cur, size, pf->swap);
306 * mmap the file, make sure timestamps are sorted, and count
308 * Timestamps represent the receive time of the packets.
309 * We need to compute also the 'first_ts' which refers to a hypotetical
310 * packet right before the first one, see the code for details.
312 static struct nm_pcap_file *
313 readpcap(const char *fn)
315 struct nm_pcap_file _f, *pf = &_f;
316 uint64_t prev_ts, first_pkt_time;
317 uint32_t magic, first_len = 0;
319 bzero(pf, sizeof(*pf));
320 pf->fd = open(fn, O_RDONLY);
322 EEE("cannot open file %s", fn);
326 pf->filesize = lseek(pf->fd, 0, SEEK_END);
327 lseek(pf->fd, 0, SEEK_SET);
328 ED("filesize is %lu", (u_long)(pf->filesize));
329 if (pf->filesize < sizeof(struct pcap_file_header)) {
330 EEE("file too short %s", fn);
334 pf->data = mmap(NULL, pf->filesize, PROT_READ, MAP_SHARED, pf->fd, 0);
335 if (pf->data == MAP_FAILED) {
336 EEE("cannot mmap file %s", fn);
341 pf->lim = pf->data + pf->filesize;
343 pf->swap = 0; /* default, same endianness when read magic */
345 magic = read_next_info(pf, 4);
346 ED("magic is 0x%x", magic);
348 case 0xa1b2c3d4: /* native, us resolution */
350 pf->resolution = 1000;
352 case 0xd4c3b2a1: /* swapped, us resolution */
354 pf->resolution = 1000;
356 case 0xa1b23c4d: /* native, ns resolution */
358 pf->resolution = 1; /* nanoseconds */
360 case 0x4d3cb2a1: /* swapped, ns resolution */
362 pf->resolution = 1; /* nanoseconds */
365 EEE("unknown magic 0x%x", magic);
369 ED("swap %d res %d\n", pf->swap, pf->resolution);
370 pf->cur = pf->data + sizeof(struct pcap_file_header);
371 pf->lim = pf->data + pf->filesize;
374 while (pf->cur < pf->lim && pf->err == 0) {
375 uint32_t base = pf->cur - pf->data;
376 uint64_t cur_ts = read_next_info(pf, 4) * NS_SCALE +
377 read_next_info(pf, 4) * pf->resolution;
378 uint32_t caplen = read_next_info(pf, 4);
379 uint32_t len = read_next_info(pf, 4);
382 WWW("end of pcap file after %d packets\n",
386 if (cur_ts < prev_ts) {
387 WWW("reordered packet %d\n",
392 if (pf->tot_pkt == 0) {
393 pf->first_ts = cur_ts;
397 pf->tot_bytes += len;
398 pf->tot_bytes_rounded += pad(len) + sizeof(struct q_pkt);
401 pf->total_tx_time = prev_ts - pf->first_ts; /* excluding first packet */
402 ED("tot_pkt %lu tot_bytes %lu tx_time %.6f s first_len %lu",
403 (u_long)pf->tot_pkt, (u_long)pf->tot_bytes,
404 1e-9*pf->total_tx_time, (u_long)first_len);
406 * We determine that based on the
407 * average bandwidth of the trace, as follows
408 * first_pkt_ts = p[0].len / avg_bw
409 * In turn avg_bw = (total_len - p[0].len)/(p[n-1].ts - p[0].ts)
411 * first_ts = p[0].ts - p[0].len * (p[n-1].ts - p[0].ts) / (total_len - p[0].len)
413 if (pf->tot_bytes == first_len) {
414 /* cannot estimate bandwidth, so force 1 Gbit */
415 first_pkt_time = first_len * 8; /* * 10^9 / bw */
417 first_pkt_time = pf->total_tx_time * first_len / (pf->tot_bytes - first_len);
419 ED("first_pkt_time %.6f s", 1e-9*first_pkt_time);
420 pf->total_tx_time += first_pkt_time;
421 pf->first_ts -= first_pkt_time;
423 /* all correct, allocate a record and copy */
424 pf = calloc(1, sizeof(*pf));
426 /* reset pointer to start */
427 pf->cur = pf->data + sizeof(struct pcap_file_header);
432 enum my_pcap_mode { PM_NONE, PM_FAST, PM_FIXED, PM_REAL };
436 static int do_abort = 0;
441 #include <sys/time.h>
443 #include <sys/resource.h> // setpriority
446 #include <pthread_np.h> /* pthread w/ affinity */
447 #include <sys/cpuset.h> /* cpu_set */
448 #endif /* __FreeBSD__ */
451 #define cpuset_t cpu_set_t
455 #define cpuset_t uint64_t // XXX
456 static inline void CPU_ZERO(cpuset_t *p)
461 static inline void CPU_SET(uint32_t i, cpuset_t *p)
463 *p |= 1<< (i & 0x3f);
466 #define pthread_setaffinity_np(a, b, c) ((void)a, 0)
467 #define sched_setscheduler(a, b, c) (1) /* error */
468 #define clock_gettime(a,b) \
469 do {struct timespec t0 = {0,0}; *(b) = t0; } while (0)
471 #define _P64 unsigned long
476 /* we use uint64_t widely, but printf gives trouble on different
477 * platforms so we use _P64 as a cast
479 #define _P64 uint64_t
480 #endif /* print stuff */
483 struct _qs; /* forward */
485 * descriptor of a configuration entry.
486 * Each handler has a parse function which takes ac/av[] and returns
487 * true if successful. Any allocated space is stored into struct _cfg *
488 * that is passed as argument.
489 * arg and arg_len are included for convenience.
492 int (*parse)(struct _qs *, struct _cfg *, int ac, char *av[]); /* 0 ok, 1 on error */
493 int (*run)(struct _qs *, struct _cfg *arg); /* 0 Ok, 1 on error */
494 // int close(struct _qs *, void *arg); /* 0 Ok, 1 on error */
496 const char *optarg; /* command line argument. Initial value is the error message */
497 /* placeholders for common values */
498 void *arg; /* allocated memory if any */
499 int arg_len; /* size of *arg in case a realloc is needed */
500 uint64_t d[16]; /* static storage for simple cases */
501 double f[4]; /* static storage for simple cases */
506 * communication occurs through this data structure, with fields
507 * cache-aligned according to who are the readers/writers.
510 The queue is an array of memory (buf) of size buflen (does not change).
512 The producer uses 'tail' as an index in the queue to indicate
513 the first empty location (ie. after the last byte of data),
514 the consumer uses head to indicate the next byte to consume.
516 For best performance we should align buffers and packets
517 to multiples of cacheline, but this would explode memory too much.
518 Worst case memory explosion is with 65 byte packets.
519 Memory usage as shown below:
522 size 32-16 32-32 32-64 64-64
528 An empty queue has head == tail, a full queue will have free space
529 below a threshold. In our case the queue is large enough and we
530 are non blocking so we can simply drop traffic when the queue
531 approaches a full state.
533 To simulate bandwidth limitations efficiently, the producer has a second
534 pointer, prod_tail_1, used to check for expired packets. This is done lazily.
538 * When sizing the buffer, we must assume some value for the bandwidth.
539 * INFINITE_BW is supposed to be faster than what we support
541 #define INFINITE_BW (200ULL*1000000*1000)
542 #define MY_CACHELINE (128ULL)
543 #define MAX_PKT (9200) /* max packet size */
545 #define ALIGN_CACHE __attribute__ ((aligned (MY_CACHELINE)))
547 struct _qs { /* shared queue */
548 uint64_t t0; /* start of times */
550 uint64_t buflen; /* queue length */
553 /* handlers for various options */
558 /* producer's fields */
559 uint64_t tx ALIGN_CACHE; /* tx counter */
560 uint64_t prod_tail_1; /* head of queue */
561 uint64_t prod_head; /* cached copy */
562 uint64_t prod_tail; /* cached copy */
563 uint64_t prod_drop; /* drop packet count */
564 uint64_t prod_max_gap; /* rx round duration */
566 struct nm_pcap_file *pcap; /* the pcap struct */
568 /* parameters for reading from the netmap port */
569 struct nm_desc *src_port; /* netmap descriptor */
570 const char * prod_ifname; /* interface name or pcap file */
571 struct netmap_ring *rxring; /* current ring being handled */
572 uint32_t si; /* ring index */
574 uint32_t rx_qmax; /* stats on max queued */
576 uint64_t qt_qout; /* queue exit time for last packet */
578 * when doing shaping, the software computes and stores here
579 * the time when the most recently queued packet will exit from
583 uint64_t qt_tx; /* delay line exit time for last packet */
585 * The software computes the time at which the most recently
586 * queued packet exits from the queue.
587 * To avoid reordering, the next packet should exit at least
591 /* producer's fields controlling the queueing */
592 const char * cur_pkt; /* current packet being analysed */
593 uint32_t cur_len; /* length of current packet */
594 uint32_t cur_caplen; /* captured length of current packet */
596 int cur_drop; /* 1 if current packet should be dropped. */
598 * cur_drop can be set as a result of the loss emulation,
599 * and may need to use the packet size, current time, etc.
602 uint64_t cur_tt; /* transmission time (ns) for current packet */
604 * The transmission time is how much link time the packet will consume.
605 * should be set by the function that does the bandwidth emulation,
606 * but could also be the result of a function that emulates the
607 * presence of competing traffic, MAC protocols etc.
608 * cur_tt is 0 for links with infinite bandwidth.
611 uint64_t cur_delay; /* delay (ns) for current packet from c_delay.run() */
613 * this should be set by the function that computes the extra delay
614 * applied to the packet.
615 * The code makes sure that there is no reordering and possibly
616 * bumps the output time as needed.
620 /* consumer's fields */
621 const char * cons_ifname;
622 uint64_t rx ALIGN_CACHE; /* rx counter */
623 uint64_t cons_head; /* cached copy */
624 uint64_t cons_tail; /* cached copy */
625 uint64_t cons_now; /* most recent producer timestamp */
626 uint64_t rx_wait; /* stats */
629 volatile uint64_t _tail ALIGN_CACHE ; /* producer writes here */
630 volatile uint64_t _head ALIGN_CACHE ; /* consumer reads from here */
636 pthread_t cons_tid; /* main thread */
637 pthread_t prod_tid; /* producer thread */
640 int cons_core; /* core for cons() */
641 int prod_core; /* core for prod() */
643 struct nm_desc *pa; /* netmap descriptor */
649 #define NS_IN_S (1000000000ULL) // nanoseconds
650 #define TIME_UNITS NS_IN_S
651 /* set the thread affinity. */
656 struct sched_param p;
661 /* Set thread affinity affinity.*/
663 CPU_SET(i, &cpumask);
665 if (pthread_setaffinity_np(pthread_self(), sizeof(cpuset_t), &cpumask) != 0) {
666 WWW("Unable to set affinity: %s", strerror(errno));
668 if (setpriority(PRIO_PROCESS, 0, -10)) {; // XXX not meaningful
669 WWW("Unable to set priority: %s", strerror(errno));
671 bzero(&p, sizeof(p));
672 p.sched_priority = 10; // 99 on linux ?
673 // use SCHED_RR or SCHED_FIFO
674 if (sched_setscheduler(0, SCHED_RR, &p)) {
675 WWW("Unable to set scheduler: %s", strerror(errno));
682 * set the timestamp from the clock, subtract t0
685 set_tns_now(uint64_t *now, uint64_t t0)
689 clock_gettime(CLOCK_REALTIME, &t); // XXX precise on FreeBSD ?
690 *now = (uint64_t)(t.tv_nsec + NS_IN_S * t.tv_sec);
696 /* compare two timestamps */
697 static inline int64_t
698 ts_cmp(uint64_t a, uint64_t b)
700 return (int64_t)(a - b);
703 /* create a packet descriptor */
704 static inline struct q_pkt *
705 pkt_at(struct _qs *q, uint64_t ofs)
707 return (struct q_pkt *)(q->buf + ofs);
712 * we have already checked for room and prepared p->next
717 struct q_pkt *p = pkt_at(q, q->prod_tail);
719 /* hopefully prefetch has been done ahead */
720 nm_pkt_copy(q->cur_pkt, (char *)(p+1), q->cur_caplen);
721 p->pktlen = q->cur_len;
722 p->pt_qout = q->qt_qout;
724 p->next = q->prod_tail + pad(q->cur_len) + sizeof(struct q_pkt);
725 ND("enqueue len %d at %d new tail %ld qout %.6f tx %.6f",
726 q->cur_len, (int)q->prod_tail, p->next,
727 1e-9*p->pt_qout, 1e-9*p->pt_tx);
728 q->prod_tail = p->next;
734 * simple handler for parameters not supplied
737 null_run_fn(struct _qs *q, struct _cfg *cfg)
747 * put packet data into the buffer.
748 * We read from the mmapped pcap file, construct header, copy
749 * the captured length of the packet and pad with zeroes.
754 struct pipe_args *pa = _pa;
755 struct _qs *q = &pa->q;
756 struct nm_pcap_file *pf = q->pcap; /* already opened by readpcap */
757 uint32_t loops, i, tot_pkts;
759 /* data plus the loop record */
761 uint64_t t_tx, tt, last_ts; /* last timestamp from trace */
764 * For speed we make sure the trace is at least some 1000 packets,
765 * so we may need to loop the trace more than once (for short traces)
767 loops = (1 + 10000 / pf->tot_pkt);
768 tot_pkts = loops * pf->tot_pkt;
769 need = loops * pf->tot_bytes_rounded + sizeof(struct q_pkt);
770 q->buf = calloc(1, need);
771 if (q->buf == NULL) {
772 D("alloc %lld bytes for queue failed, exiting",(long long)need);
775 q->prod_head = q->prod_tail = 0;
778 pf->cur = pf->data + sizeof(struct pcap_file_header);
781 ED("--- start create %lu packets at tail %d",
782 (u_long)tot_pkts, (int)q->prod_tail);
783 last_ts = pf->first_ts; /* beginning of the trace */
785 q->qt_qout = 0; /* first packet out of the queue */
787 for (loops = 0, i = 0; i < tot_pkts && !do_abort; i++) {
788 const char *next_pkt; /* in the pcap buffer */
791 /* read values from the pcap buffer */
792 cur_ts = read_next_info(pf, 4) * NS_SCALE +
793 read_next_info(pf, 4) * pf->resolution;
794 q->cur_caplen = read_next_info(pf, 4);
795 q->cur_len = read_next_info(pf, 4);
796 next_pkt = pf->cur + q->cur_caplen;
798 /* prepare fields in q for the generator */
799 q->cur_pkt = pf->cur;
800 /* initial estimate of tx time */
801 q->cur_tt = cur_ts - last_ts;
802 // -pf->first_ts + loops * pf->total_tx_time - last_ts;
804 if ((i % pf->tot_pkt) == 0)
805 ED("insert %5d len %lu cur_tt %.6f",
806 i, (u_long)q->cur_len, 1e-9*q->cur_tt);
808 /* prepare for next iteration */
811 if (next_pkt == pf->lim) { //last pkt
812 pf->cur = pf->data + sizeof(struct pcap_file_header);
813 last_ts = pf->first_ts; /* beginning of the trace */
817 q->c_loss.run(q, &q->c_loss);
820 q->c_bw.run(q, &q->c_bw);
827 q->c_delay.run(q, &q->c_delay); /* compute delay */
828 t_tx = q->qt_qout + q->cur_delay;
829 ND(5, "tt %ld qout %ld tx %ld qt_tx %ld", tt, q->qt_qout, t_tx, q->qt_tx);
830 /* insure no reordering and spacing by transmission time */
831 q->qt_tx = (t_tx >= q->qt_tx + tt) ? t_tx : q->qt_tx + tt;
835 ND("ins %d q->prod_tail = %lu", (int)insert, (unsigned long)q->prod_tail);
838 ED("done q->prod_tail:%d",(int)q->prod_tail);
839 q->_tail = q->prod_tail; /* publish */
843 if (q->buf != NULL) {
852 * the consumer reads from the queue using head,
853 * advances it every now and then.
858 struct pipe_args *pa = _pa;
859 struct _qs *q = &pa->q;
861 uint64_t last_ts = 0;
863 /* read the start of times in q->t0 */
864 set_tns_now(&q->t0, 0);
865 /* set the time (cons_now) to clock - q->t0 */
866 set_tns_now(&q->cons_now, q->t0);
867 q->cons_head = q->_head;
868 q->cons_tail = q->_tail;
869 while (!do_abort) { /* consumer, infinite */
870 struct q_pkt *p = pkt_at(q, q->cons_head);
872 __builtin_prefetch (q->buf + p->next);
874 if (q->cons_head == q->cons_tail) { //reset record
875 ND("Transmission restarted");
877 * add to q->t0 the time for the last packet
880 set_tns_now(&q->cons_now, q->t0);
881 q->cons_head = 0; //restart from beginning of the queue
885 if (ts_cmp(p->pt_tx, q->cons_now) > 0) {
888 /* the ioctl should be conditional */
889 ioctl(pa->pb->fd, NIOCTXSYNC, 0); // XXX just in case
892 set_tns_now(&q->cons_now, q->t0);
895 /* XXX copy is inefficient but simple */
896 if (nm_inject(pa->pb, (char *)(p + 1), p->pktlen) == 0) {
897 RD(1, "inject failed len %d now %ld tx %ld h %ld t %ld next %ld",
898 (int)p->pktlen, (u_long)q->cons_now, (u_long)p->pt_tx,
899 (u_long)q->_head, (u_long)q->_tail, (u_long)p->next);
900 ioctl(pa->pb->fd, NIOCTXSYNC, 0);
905 if (pending > q->burst) {
906 ioctl(pa->pb->fd, NIOCTXSYNC, 0);
910 q->cons_head = p->next;
911 /* drain packets from the queue */
914 D("exiting on abort");
919 * In case of pcap file as input, the program acts in 2 different
920 * phases. It first fill the queue and then starts the cons()
923 nmreplay_main(void *_a)
925 struct pipe_args *a = _a;
926 struct _qs *q = &a->q;
927 const char *cap_fname = q->prod_ifname;
929 setaffinity(a->cons_core);
930 set_tns_now(&q->t0, 0); /* starting reference */
931 if (cap_fname == NULL) {
934 q->pcap = readpcap(cap_fname);
935 if (q->pcap == NULL) {
936 EEE("unable to read file %s", cap_fname);
940 destroy_pcap(q->pcap);
942 a->pb = nm_open(q->cons_ifname, NULL, 0, NULL);
944 EEE("cannot open netmap on %s", q->cons_ifname);
945 do_abort = 1; // XXX any better way ?
948 /* continue as cons() */
949 WWW("prepare to send packets");
952 EEE("exiting on abort");
954 if (q->pcap != NULL) {
955 destroy_pcap(q->pcap);
965 (void)sig; /* UNUSED */
967 signal(SIGINT, SIG_DFL);
976 "usage: nmreplay [-v] [-D delay] [-B {[constant,]bps|ether,bps|real,speedup}] [-L loss]\n"
977 "\t[-b burst] -f pcap-file -i <netmap:ifname|valeSSS:PPP>\n");
982 /*---- configuration handling ---- */
984 * support routine: split argument, returns ac and *av.
985 * av contains two extra entries, a NULL and a pointer
986 * to the entire string.
989 split_arg(const char *src, int *_ac)
991 char *my = NULL, **av = NULL, *seps = " \t\r\n,";
992 int l, i, ac; /* number of entries */
997 /* in the first pass we count fields, in the second pass
998 * we allocate the av[] array and a copy of the string
999 * and fill av[]. av[ac] = NULL, av[ac+1]
1003 ND("start pass %d: <%s>", av ? 1 : 0, my);
1005 /* trim leading separator */
1006 while (i <l && strchr(seps, src[i]))
1010 ND(" pass %d arg %d: <%s>", av ? 1 : 0, ac, src+i);
1011 if (av) /* in the second pass, set the result */
1015 while (i <l && !strchr(seps, src[i])) i++;
1017 my[i] = '\0'; /* write marker */
1019 if (!av) { /* end of first pass */
1021 av = calloc(1, (l+1) + (ac + 2)*sizeof(char *));
1022 my = (char *)&(av[ac+2]);
1028 for (i = 0; i < ac; i++) {
1029 NED("%d: <%s>", i, av[i]);
1039 * apply a command against a set of functions,
1040 * install a handler in *dst
1043 cmd_apply(const struct _cfg *a, const char *arg, struct _qs *q, struct _cfg *dst)
1049 if (arg == NULL || *arg == '\0')
1050 return 1; /* no argument may be ok */
1051 if (a == NULL || dst == NULL) {
1052 ED("program error - invalid arguments");
1055 av = split_arg(arg, &ac);
1057 return 1; /* error */
1058 for (i = 0; a[i].parse; i++) {
1059 struct _cfg x = a[i];
1060 const char *errmsg = x.optarg;
1065 bzero(&x.d, sizeof(x.d));
1066 ND("apply %s to %s", av[0], errmsg);
1067 ret = x.parse(q, &x, ac, av);
1068 if (ret == 2) /* not recognised */
1071 ED("invalid arguments: need '%s' have '%s'",
1079 ED("arguments %s not recognised", arg);
1084 static struct _cfg delay_cfg[];
1085 static struct _cfg bw_cfg[];
1086 static struct _cfg loss_cfg[];
1088 static uint64_t parse_bw(const char *arg);
1091 * prodcons [options]
1092 * accept separate sets of arguments for the two directions
1097 add_to(const char ** v, int l, const char *arg, const char *msg)
1099 for (; l > 0 && *v != NULL ; l--, v++);
1101 ED("%s %s", msg, arg);
1108 main(int argc, char **argv)
1113 struct pipe_args bp[N_OPTS];
1114 const char *d[N_OPTS], *b[N_OPTS], *l[N_OPTS], *q[N_OPTS], *ifname[N_OPTS], *m[N_OPTS];
1115 const char *pcap_file[N_OPTS];
1116 int cores[4] = { 2, 8, 4, 10 }; /* default values */
1118 bzero(&bp, sizeof(bp)); /* all data initially go here */
1119 bzero(d, sizeof(d));
1120 bzero(b, sizeof(b));
1121 bzero(l, sizeof(l));
1122 bzero(q, sizeof(q));
1123 bzero(m, sizeof(m));
1124 bzero(ifname, sizeof(ifname));
1125 bzero(pcap_file, sizeof(pcap_file));
1128 /* set default values */
1129 for (i = 0; i < N_OPTS; i++) {
1130 struct _qs *q = &bp[i].q;
1133 q->c_delay.optarg = "0";
1134 q->c_delay.run = null_run_fn;
1135 q->c_loss.optarg = "0";
1136 q->c_loss.run = null_run_fn;
1137 q->c_bw.optarg = "0";
1138 q->c_bw.run = null_run_fn;
1142 // B bandwidth in bps
1143 // D delay in seconds
1144 // L loss probability
1152 while ( (ch = getopt(argc, argv, "B:C:D:L:b:f:i:vw:")) != -1) {
1155 D("bad option %c %s", ch, optarg);
1159 case 'C': /* CPU placement, up to 4 arguments */
1162 char **av = split_arg(optarg, &ac);
1163 if (ac == 1) { /* sequential after the first */
1164 cores[0] = atoi(av[0]);
1165 cores[1] = cores[0] + 1;
1166 cores[2] = cores[1] + 1;
1167 cores[3] = cores[2] + 1;
1168 } else if (ac == 2) { /* two sequential pairs */
1169 cores[0] = atoi(av[0]);
1170 cores[1] = cores[0] + 1;
1171 cores[2] = atoi(av[1]);
1172 cores[3] = cores[2] + 1;
1173 } else if (ac == 4) { /* four values */
1174 cores[0] = atoi(av[0]);
1175 cores[1] = atoi(av[1]);
1176 cores[2] = atoi(av[2]);
1177 cores[3] = atoi(av[3]);
1179 ED(" -C accepts 1, 2 or 4 comma separated arguments");
1187 case 'B': /* bandwidth in bps */
1188 add_to(b, N_OPTS, optarg, "-B too many times");
1191 case 'D': /* delay in seconds (float) */
1192 add_to(d, N_OPTS, optarg, "-D too many times");
1195 case 'L': /* loss probability */
1196 add_to(l, N_OPTS, optarg, "-L too many times");
1199 case 'b': /* burst */
1200 bp[0].q.burst = atoi(optarg);
1203 case 'f': /* pcap_file */
1204 add_to(pcap_file, N_OPTS, optarg, "-f too many times");
1206 case 'i': /* interface */
1207 add_to(ifname, N_OPTS, optarg, "-i too many times");
1213 bp[0].wait_link = atoi(optarg);
1223 * consistency checks for common arguments
1224 * if pcap file has been provided we need just one interface, two otherwise
1226 if (!pcap_file[0]) {
1227 ED("missing pcap file");
1231 ED("missing interface");
1234 if (bp[0].q.burst < 1 || bp[0].q.burst > 8192) {
1235 WWW("invalid burst %d, set to 1024", bp[0].q.burst);
1236 bp[0].q.burst = 1024; // XXX 128 is probably better
1238 if (bp[0].wait_link > 100) {
1239 ED("invalid wait_link %d, set to 4", bp[0].wait_link);
1240 bp[0].wait_link = 4;
1243 bp[0].q.prod_ifname = pcap_file[0];
1244 bp[0].q.cons_ifname = ifname[0];
1246 /* assign cores. prod and cons work better if on the same HT */
1247 bp[0].cons_core = cores[0];
1248 bp[0].prod_core = cores[1];
1249 ED("running on cores %d %d %d %d", cores[0], cores[1], cores[2], cores[3]);
1251 /* apply commands */
1252 for (i = 0; i < N_OPTS; i++) { /* once per queue */
1253 struct _qs *q = &bp[i].q;
1254 err += cmd_apply(delay_cfg, d[i], q, &q->c_delay);
1255 err += cmd_apply(bw_cfg, b[i], q, &q->c_bw);
1256 err += cmd_apply(loss_cfg, l[i], q, &q->c_loss);
1259 pthread_create(&bp[0].cons_tid, NULL, nmreplay_main, (void*)&bp[0]);
1260 signal(SIGINT, sigint_h);
1263 struct _qs olda = bp[0].q;
1264 struct _qs *q0 = &bp[0].q;
1267 ED("%lld -> %lld maxq %d round %lld",
1268 (long long)(q0->rx - olda.rx), (long long)(q0->tx - olda.tx),
1269 q0->rx_qmax, (long long)q0->prod_max_gap
1271 ED("plr nominal %le actual %le",
1272 (double)(q0->c_loss.d[0])/(1<<24),
1273 q0->c_loss.d[1] == 0 ? 0 :
1274 (double)(q0->c_loss.d[2])/q0->c_loss.d[1]);
1275 bp[0].q.rx_qmax = (bp[0].q.rx_qmax * 7)/8; // ewma
1276 bp[0].q.prod_max_gap = (bp[0].q.prod_max_gap * 7)/8; // ewma
1278 D("exiting on abort");
1284 /* conversion factor for numbers.
1285 * Each entry has a set of characters and conversion factor,
1286 * the first entry should have an empty string and default factor,
1287 * the final entry has s = NULL.
1289 struct _sm { /* string and multiplier */
1295 * parse a generic value
1298 parse_gen(const char *arg, const struct _sm *conv, int *err)
1309 d = strtod(arg, &ep);
1310 if (ep == arg) { /* no value */
1311 ED("bad argument %s", arg);
1314 /* special case, no conversion */
1315 if (conv == NULL && *ep == '\0')
1317 ND("checking %s [%s]", arg, ep);
1318 for (;conv->s; conv++) {
1319 if (strchr(conv->s, *ep))
1323 *err = 1; /* unrecognised */
1328 ND("scale is %s %lf", conv->s, conv->m);
1329 d *= conv->m; /* apply default conversion */
1331 ND("returning %lf", d);
1335 #define U_PARSE_ERR ~(0ULL)
1337 /* returns a value in nanoseconds */
1339 parse_time(const char *arg)
1342 {"", 1000000000 /* seconds */},
1343 {"n", 1 /* nanoseconds */}, {"u", 1000 /* microseconds */},
1344 {"m", 1000000 /* milliseconds */}, {"s", 1000000000 /* seconds */},
1345 {NULL, 0 /* seconds */}
1348 uint64_t ret = (uint64_t)parse_gen(arg, a, &err);
1349 return err ? U_PARSE_ERR : ret;
1354 * parse a bandwidth, returns value in bps or U_PARSE_ERR if error.
1357 parse_bw(const char *arg)
1360 {"", 1}, {"kK", 1000}, {"mM", 1000000}, {"gG", 1000000000}, {NULL, 0}
1363 uint64_t ret = (uint64_t)parse_gen(arg, a, &err);
1364 return err ? U_PARSE_ERR : ret;
1369 * For some function we need random bits.
1370 * This is a wrapper to whatever function you want that returns
1371 * 24 useful random bits.
1374 #include <math.h> /* log, exp etc. */
1375 static inline uint64_t
1376 my_random24(void) /* 24 useful bits */
1378 return random() & ((1<<24) - 1);
1382 /*-------------- user-configuration -----------------*/
1384 #if 0 /* start of comment block */
1386 Here we place the functions to implement the various features
1387 of the system. For each feature one should define a struct _cfg
1388 (see at the beginning for definition) that refers a *_parse() function
1389 to extract values from the command line, and a *_run() function
1390 that is invoked on each packet to implement the desired function.
1392 Examples of the two functions are below. In general:
1394 - the *_parse() function takes argc/argv[], matches the function
1395 name in argv[0], extracts the operating parameters, allocates memory
1396 if needed, and stores them in the struct _cfg.
1397 Return value is 2 if argv[0] is not recosnised, 1 if there is an
1398 error in the arguments, 0 if all ok.
1400 On the command line, argv[] is a single, comma separated argument
1401 that follow the specific option eg -D constant,20ms
1403 struct _cfg has some preallocated space (e.g an array of uint64_t) so simple
1404 function can use that without having to allocate memory.
1406 - the *_run() function takes struct _q *q and struct _cfg *cfg as arguments.
1407 *q contains all the informatio that may be possibly needed, including
1408 those on the packet currently under processing.
1409 The basic values are the following:
1411 char * cur_pkt points to the current packet (linear buffer)
1412 uint32_t cur_len; length of the current packet
1413 the functions are not supposed to modify these values
1415 int cur_drop; true if current packet must be dropped.
1416 Must be set to non-zero by the loss emulation function
1418 uint64_t cur_delay; delay in nanoseconds for the current packet
1419 Must be set by the delay emulation function
1421 More sophisticated functions may need to access other fields in *q,
1422 see the structure description for that.
1424 When implementing a new function for a feature (e.g. for delay,
1425 bandwidth, loss...) the struct _cfg should be added to the array
1426 that contains all possible options.
1428 --- Specific notes ---
1430 DELAY emulation -D option_arguments
1432 If the option is not supplied, the system applies 0 extra delay
1434 The resolution for times is 1ns, the precision is load dependent and
1435 generally in the order of 20-50us.
1436 Times are in nanoseconds, can be followed by a character specifying
1437 a different unit e.g.
1444 Currently implemented options:
1446 constant,t constant delay equal to t
1448 uniform,tmin,tmax uniform delay between tmin and tmax
1450 exp,tavg,tmin exponential distribution with average tavg
1451 and minimum tmin (corresponds to an exponential
1452 distribution with argument 1/(tavg-tmin) )
1455 LOSS emulation -L option_arguments
1457 Loss is expressed as packet or bit error rate, which is an absolute
1458 number between 0 and 1 (typically small).
1460 Currently implemented options
1462 plr,p uniform packet loss rate p, independent
1465 burst,p,lmin,lmax burst loss with burst probability p and
1466 burst length uniformly distributed between
1469 ber,p uniformly distributed bit error rate p,
1470 so actual loss prob. depends on size.
1472 BANDWIDTH emulation -B option_arguments
1474 Bandwidths are expressed in bits per second, can be followed by a
1475 character specifying a different unit e.g.
1478 k/K kbits/s (10^3 bits/s)
1479 m/M mbits/s (10^6 bits/s)
1480 g/G gbits/s (10^9 bits/s)
1482 Currently implemented options
1484 const,b constant bw, excluding mac framing
1485 ether,b constant bw, including ethernet framing
1486 (20 bytes framing + 4 bytes crc)
1487 real,[scale] use real time, optionally with a scaling factor
1489 #endif /* end of comment block */
1492 * Configuration options for delay
1495 /* constant delay, also accepts just a number */
1497 const_delay_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1502 if (strncmp(av[0], "const", 5) != 0 && ac > 1)
1503 return 2; /* unrecognised */
1505 return 1; /* error */
1506 delay = parse_time(av[ac - 1]);
1507 if (delay == U_PARSE_ERR)
1508 return 1; /* error */
1510 return 0; /* success */
1513 /* runtime function, store the delay into q->cur_delay */
1515 const_delay_run(struct _qs *q, struct _cfg *arg)
1517 q->cur_delay = arg->d[0]; /* the delay */
1522 uniform_delay_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1524 uint64_t dmin, dmax;
1527 if (strcmp(av[0], "uniform") != 0)
1528 return 2; /* not recognised */
1530 return 1; /* error */
1531 dmin = parse_time(av[1]);
1532 dmax = parse_time(av[2]);
1533 if (dmin == U_PARSE_ERR || dmax == U_PARSE_ERR || dmin > dmax)
1535 D("dmin %lld dmax %lld", (long long)dmin, (long long)dmax);
1538 dst->d[2] = dmax - dmin;
1543 uniform_delay_run(struct _qs *q, struct _cfg *arg)
1545 uint64_t x = my_random24();
1546 q->cur_delay = arg->d[0] + ((arg->d[2] * x) >> 24);
1547 #if 0 /* COMPUTE_STATS */
1548 #endif /* COMPUTE_STATS */
1553 * exponential delay: Prob(delay = x) = exp(-x/d_av)
1554 * gives a delay between 0 and infinity with average d_av
1555 * The cumulative function is 1 - d_av exp(-x/d_av)
1557 * The inverse function generates a uniform random number p in 0..1
1558 * and generates delay = (d_av-d_min) * -ln(1-p) + d_min
1560 * To speed up behaviour at runtime we tabulate the values
1564 exp_delay_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1566 #define PTS_D_EXP 512
1567 uint64_t i, d_av, d_min, *t; /*table of values */
1570 if (strcmp(av[0], "exp") != 0)
1571 return 2; /* not recognised */
1573 return 1; /* error */
1574 d_av = parse_time(av[1]);
1575 d_min = parse_time(av[2]);
1576 if (d_av == U_PARSE_ERR || d_min == U_PARSE_ERR || d_av < d_min)
1577 return 1; /* error */
1579 dst->arg_len = PTS_D_EXP * sizeof(uint64_t);
1580 dst->arg = calloc(1, dst->arg_len);
1581 if (dst->arg == NULL)
1582 return 1; /* no memory */
1583 t = (uint64_t *)dst->arg;
1584 /* tabulate -ln(1-n)*delay for n in 0..1 */
1585 for (i = 0; i < PTS_D_EXP; i++) {
1586 double d = -log2 ((double)(PTS_D_EXP - i) / PTS_D_EXP) * d_av + d_min;
1588 ND(5, "%ld: %le", i, d);
1594 exp_delay_run(struct _qs *q, struct _cfg *arg)
1596 uint64_t *t = (uint64_t *)arg->arg;
1597 q->cur_delay = t[my_random24() & (PTS_D_EXP - 1)];
1598 RD(5, "delay %llu", (unsigned long long)q->cur_delay);
1603 /* unused arguments in configuration */
1604 #define TLEM_CFG_END NULL, 0, {0}, {0}
1606 static struct _cfg delay_cfg[] = {
1607 { const_delay_parse, const_delay_run,
1608 "constant,delay", TLEM_CFG_END },
1609 { uniform_delay_parse, uniform_delay_run,
1610 "uniform,dmin,dmax # dmin <= dmax", TLEM_CFG_END },
1611 { exp_delay_parse, exp_delay_run,
1612 "exp,dmin,davg # dmin <= davg", TLEM_CFG_END },
1613 { NULL, NULL, NULL, TLEM_CFG_END }
1616 /* standard bandwidth, also accepts just a number */
1618 const_bw_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1623 if (strncmp(av[0], "const", 5) != 0 && ac > 1)
1624 return 2; /* unrecognised */
1626 return 1; /* error */
1627 bw = parse_bw(av[ac - 1]);
1628 if (bw == U_PARSE_ERR) {
1629 return (ac == 2) ? 1 /* error */ : 2 /* unrecognised */;
1632 return 0; /* success */
1636 /* runtime function, store the delay into q->cur_delay */
1638 const_bw_run(struct _qs *q, struct _cfg *arg)
1640 uint64_t bps = arg->d[0];
1641 q->cur_tt = bps ? 8ULL* TIME_UNITS * q->cur_len / bps : 0 ;
1645 /* ethernet bandwidth, add 672 bits per packet */
1647 ether_bw_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1652 if (strcmp(av[0], "ether") != 0)
1653 return 2; /* unrecognised */
1655 return 1; /* error */
1656 bw = parse_bw(av[ac - 1]);
1657 if (bw == U_PARSE_ERR)
1658 return 1; /* error */
1660 return 0; /* success */
1664 /* runtime function, add 20 bytes (framing) + 4 bytes (crc) */
1666 ether_bw_run(struct _qs *q, struct _cfg *arg)
1668 uint64_t bps = arg->d[0];
1669 q->cur_tt = bps ? 8ULL * TIME_UNITS * (q->cur_len + 24) / bps : 0 ;
1673 /* real bandwidth, plus scaling factor */
1675 real_bw_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1680 if (strcmp(av[0], "real") != 0)
1681 return 2; /* unrecognised */
1682 if (ac > 2) { /* second argument is optional */
1683 return 1; /* error */
1684 } else if (ac == 1) {
1688 scale = parse_gen(av[ac-1], NULL, &err);
1689 if (err || scale <= 0 || scale > 1000)
1692 ED("real -> scale is %.6f", scale);
1694 return 0; /* success */
1698 real_bw_run(struct _qs *q, struct _cfg *arg)
1700 q->cur_tt /= arg->f[0];
1704 static struct _cfg bw_cfg[] = {
1705 { const_bw_parse, const_bw_run,
1706 "constant,bps", TLEM_CFG_END },
1707 { ether_bw_parse, ether_bw_run,
1708 "ether,bps", TLEM_CFG_END },
1709 { real_bw_parse, real_bw_run,
1710 "real,scale", TLEM_CFG_END },
1711 { NULL, NULL, NULL, TLEM_CFG_END }
1718 const_plr_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1724 if (strcmp(av[0], "plr") != 0 && ac > 1)
1725 return 2; /* unrecognised */
1727 return 1; /* error */
1728 // XXX to be completed
1729 plr = parse_gen(av[ac-1], NULL, &err);
1730 if (err || plr < 0 || plr > 1)
1732 dst->d[0] = plr * (1<<24); /* scale is 16m */
1733 if (plr != 0 && dst->d[0] == 0)
1734 ED("WWW warning, rounding %le down to 0", plr);
1735 return 0; /* success */
1739 const_plr_run(struct _qs *q, struct _cfg *arg)
1742 uint64_t r = my_random24();
1743 q->cur_drop = r < arg->d[0];
1744 #if 1 /* keep stats */
1746 arg->d[2] += q->cur_drop;
1753 * For BER the loss is 1- (1-ber)**bit_len
1754 * The linear approximation is only good for small values, so we
1755 * tabulate (1-ber)**len for various sizes in bytes
1758 const_ber_parse(struct _qs *q, struct _cfg *dst, int ac, char *av[])
1760 double ber, ber8, cur;
1763 const uint32_t mask = (1<<24) - 1;
1766 if (strcmp(av[0], "ber") != 0)
1767 return 2; /* unrecognised */
1769 return 1; /* error */
1770 ber = parse_gen(av[ac-1], NULL, &err);
1771 if (err || ber < 0 || ber > 1)
1773 dst->arg_len = MAX_PKT * sizeof(uint32_t);
1774 plr = calloc(1, dst->arg_len);
1776 return 1; /* no memory */
1779 ber8 *= ber8; /* **2 */
1780 ber8 *= ber8; /* **4 */
1781 ber8 *= ber8; /* **8 */
1783 for (i=0; i < MAX_PKT; i++, cur *= ber8) {
1784 plr[i] = (mask + 1)*(1 - cur);
1788 if (i>= 60) // && plr[i] < mask/2)
1789 RD(50,"%4d: %le %ld", i, 1.0 - cur, (_P64)plr[i]);
1792 dst->d[0] = ber * (mask + 1);
1793 return 0; /* success */
1797 const_ber_run(struct _qs *q, struct _cfg *arg)
1800 uint64_t r = my_random24();
1801 uint32_t *plr = arg->arg;
1804 RD(5, "pkt len %d too large, trim to %d", l, MAX_PKT-1);
1807 q->cur_drop = r < plr[l];
1808 #if 1 /* keep stats */
1810 arg->d[2] += q->cur_drop;
1815 static struct _cfg loss_cfg[] = {
1816 { const_plr_parse, const_plr_run,
1817 "plr,prob # 0 <= prob <= 1", TLEM_CFG_END },
1818 { const_ber_parse, const_ber_run,
1819 "ber,prob # 0 <= prob <= 1", TLEM_CFG_END },
1820 { NULL, NULL, NULL, TLEM_CFG_END }