1 /* Copyright (c) 2010-2014. The SimGrid Team.
2 * All rights reserved. */
4 /* This program is free software; you can redistribute it and/or modify it
5 * under the terms of the license (GNU LGPL) which comes with this package. */
8 #include "simgrid/msg.h"
10 #include "xbt/asserts.h"
11 #include "simgrid/modelchecker.h"
12 #include <xbt/RngStream.h>
14 /** @addtogroup MSG_examples
16 * - <b>chord/chord.c: Classical Chord P2P protocol</b>
17 * This example implements the well known Chord P2P protocol. Its
18 * main advantage is that it constitute a fully working non-trivial
19 * example. In addition, its implementation is rather efficient, as
20 * demonstrated in http://hal.inria.fr/inria-00602216/
24 XBT_LOG_NEW_DEFAULT_CATEGORY(msg_chord,
25 "Messages specific for this msg example");
29 #define MAILBOX_NAME_SIZE 10
31 static int nb_bits = 24;
32 static int nb_keys = 0;
33 static int timeout = 50;
34 static int max_simulation_time = 1000;
35 static int periodic_stabilize_delay = 20;
36 static int periodic_fix_fingers_delay = 120;
37 static int periodic_check_predecessor_delay = 120;
38 static int periodic_lookup_delay = 10;
40 static const double sleep_delay = 4.9999;
42 extern long int smx_total_comms;
47 typedef struct s_finger {
49 char mailbox[MAILBOX_NAME_SIZE]; // string representation of the id
50 } s_finger_t, *finger_t;
55 typedef struct s_node {
57 char mailbox[MAILBOX_NAME_SIZE]; // my mailbox name (string representation of the id)
58 s_finger_t *fingers; // finger table, of size nb_bits (fingers[0] is my successor)
59 int pred_id; // predecessor id
60 char pred_mailbox[MAILBOX_NAME_SIZE]; // predecessor's mailbox name
61 int next_finger_to_fix; // index of the next finger to fix in fix_fingers()
62 msg_comm_t comm_receive; // current communication to receive
63 double last_change_date; // last time I changed a finger or my predecessor
64 RngStream stream; //RngStream for
68 * Types of tasks exchanged between nodes.
72 TASK_FIND_SUCCESSOR_ANSWER,
74 TASK_GET_PREDECESSOR_ANSWER,
76 TASK_SUCCESSOR_LEAVING,
77 TASK_PREDECESSOR_LEAVING,
78 TASK_PREDECESSOR_ALIVE,
79 TASK_PREDECESSOR_ALIVE_ANSWER
83 * Data attached with the tasks sent and received
85 typedef struct s_task_data {
86 e_task_type_t type; // type of task
87 int request_id; // id paramater (used by some types of tasks)
88 int request_finger; // finger parameter (used by some types of tasks)
89 int answer_id; // answer (used by some types of tasks)
90 char answer_to[MAILBOX_NAME_SIZE]; // mailbox to send an answer to (if any)
91 const char* issuer_host_name; // used for logging
92 } s_task_data_t, *task_data_t;
95 static xbt_dynar_t host_list;
98 static void chord_initialize(void);
99 static void chord_exit(void);
100 static int normalize(int id);
101 static int is_in_interval(int id, int start, int end);
102 static void get_mailbox(int host_id, char* mailbox);
103 static void task_free(void* task);
104 static void print_finger_table(node_t node);
105 static void set_finger(node_t node, int finger_index, int id);
106 static void set_predecessor(node_t node, int predecessor_id);
109 static int node(int argc, char *argv[]);
110 static void handle_task(node_t node, msg_task_t task);
113 static void create(node_t node);
114 static int join(node_t node, int known_id);
115 static void leave(node_t node);
116 static int find_successor(node_t node, int id);
117 static int remote_find_successor(node_t node, int ask_to_id, int id);
118 static int remote_get_predecessor(node_t node, int ask_to_id);
119 static int closest_preceding_node(node_t node, int id);
120 static void stabilize(node_t node);
121 static void notify(node_t node, int predecessor_candidate_id);
122 static void remote_notify(node_t node, int notify_to, int predecessor_candidate_id);
123 static void fix_fingers(node_t node);
124 static void check_predecessor(node_t node);
125 static void random_lookup(node_t);
126 static void quit_notify(node_t node);
129 * \brief Global initialization of the Chord simulation.
131 static void chord_initialize(void)
133 // compute the powers of 2 once for all
134 powers2 = xbt_new(int, nb_bits);
137 for (i = 0; i < nb_bits; i++) {
142 XBT_DEBUG("Sets nb_keys to %d", nb_keys);
145 host_list = MSG_hosts_as_dynar();
146 xbt_dynar_foreach(host_list, i, host) {
149 snprintf(descr, sizeof descr, "RngSream<%s>", MSG_host_get_name(host));
150 stream = RngStream_CreateStream(descr);
151 MSG_host_set_property_value(host, "stream", (char*)stream, NULL);
155 static void chord_exit(void)
159 xbt_dynar_foreach(host_list, i, host) {
160 RngStream stream = (RngStream)MSG_host_get_property_value(host, "stream");
161 RngStream_DeleteStream(&stream);
163 xbt_dynar_free(&host_list);
169 * \brief Turns an id into an equivalent id in [0, nb_keys).
171 * \return the corresponding normalized id
173 static int normalize(int id)
175 // like id % nb_keys, but works with negatives numbers (and faster)
176 return id & (nb_keys - 1);
180 * \brief Returns whether an id belongs to the interval [start, end].
182 * The parameters are noramlized to make sure they are between 0 and nb_keys - 1).
183 * 1 belongs to [62, 3]
184 * 1 does not belong to [3, 62]
185 * 63 belongs to [62, 3]
186 * 63 does not belong to [3, 62]
187 * 24 belongs to [21, 29]
188 * 24 does not belong to [29, 21]
190 * \param id id to check
191 * \param start lower bound
192 * \param end upper bound
193 * \return a non-zero value if id in in [start, end]
195 static int is_in_interval(int id, int start, int end)
198 start = normalize(start);
199 end = normalize(end);
201 // make sure end >= start and id >= start
214 * \brief Gets the mailbox name of a host given its chord id.
215 * \param node_id id of a node
216 * \param mailbox pointer to where the mailbox name should be written
217 * (there must be enough space)
219 static void get_mailbox(int node_id, char* mailbox)
221 snprintf(mailbox, MAILBOX_NAME_SIZE - 1, "%d", node_id);
225 * \brief Frees the memory used by a task.
226 * \param task the MSG task to destroy
228 static void task_free(void* task)
230 // TODO add a parameter data_free_function to MSG_task_create?
232 xbt_free(MSG_task_get_data(task));
233 MSG_task_destroy(task);
238 * \brief Displays the finger table of a node.
241 static void print_finger_table(node_t node)
243 if (XBT_LOG_ISENABLED(msg_chord, xbt_log_priority_verbose)) {
245 XBT_VERB("My finger table:");
246 XBT_VERB("Start | Succ");
247 for (i = 0; i < nb_bits; i++) {
248 XBT_VERB(" %3d | %3d", (node->id + powers2[i]) % nb_keys, node->fingers[i].id);
250 XBT_VERB("Predecessor: %d", node->pred_id);
255 * \brief Sets a finger of the current node.
256 * \param node the current node
257 * \param finger_index index of the finger to set (0 to nb_bits - 1)
258 * \param id the id to set for this finger
260 static void set_finger(node_t node, int finger_index, int id)
262 if (id != node->fingers[finger_index].id) {
263 node->fingers[finger_index].id = id;
264 get_mailbox(id, node->fingers[finger_index].mailbox);
265 node->last_change_date = MSG_get_clock();
266 XBT_DEBUG("My new finger #%d is %d", finger_index, id);
271 * \brief Sets the predecessor of the current node.
272 * \param node the current node
273 * \param id the id to predecessor, or -1 to unset the predecessor
275 static void set_predecessor(node_t node, int predecessor_id)
277 if (predecessor_id != node->pred_id) {
278 node->pred_id = predecessor_id;
280 if (predecessor_id != -1) {
281 get_mailbox(predecessor_id, node->pred_mailbox);
283 node->last_change_date = MSG_get_clock();
285 XBT_DEBUG("My new predecessor is %d", predecessor_id);
290 * \brief Node Function
293 * - the id of a guy I know in the system (except for the first node)
294 * - the time to sleep before I join (except for the first node)
296 int node(int argc, char *argv[])
299 /* Reduce the run size for the MC */
300 if(MC_is_active() || MC_record_replay_is_active()){
301 periodic_stabilize_delay = 8;
302 periodic_fix_fingers_delay = 8;
303 periodic_check_predecessor_delay = 8;
306 double init_time = MSG_get_clock();
307 msg_task_t task_received = NULL;
309 int join_success = 0;
311 double next_stabilize_date = init_time + periodic_stabilize_delay;
312 double next_fix_fingers_date = init_time + periodic_fix_fingers_delay;
313 double next_check_predecessor_date = init_time + periodic_check_predecessor_delay;
314 double next_lookup_date = init_time + periodic_lookup_delay;
318 int sub_protocol = 0;
320 xbt_assert(argc == 3 || argc == 5, "Wrong number of arguments for this node");
322 // initialize my node
324 node.id = atoi(argv[1]);
326 (RngStream)MSG_host_get_property_value(MSG_host_self(), "stream");
327 get_mailbox(node.id, node.mailbox);
328 node.next_finger_to_fix = 0;
329 node.fingers = xbt_new0(s_finger_t, nb_bits);
330 node.last_change_date = init_time;
332 for (i = 0; i < nb_bits; i++) {
333 node.fingers[i].id = -1;
334 set_finger(&node, i, node.id);
337 if (argc == 3) { // first ring
338 deadline = atof(argv[2]);
343 int known_id = atoi(argv[2]);
344 //double sleep_time = atof(argv[3]);
345 deadline = atof(argv[4]);
348 // sleep before starting
349 XBT_DEBUG("Let's sleep during %f", sleep_time);
350 MSG_process_sleep(sleep_time);
352 XBT_DEBUG("Hey! Let's join the system.");
354 join_success = join(&node, known_id);
358 while (MSG_get_clock() < init_time + deadline
359 // && MSG_get_clock() < node.last_change_date + 1000
360 && MSG_get_clock() < max_simulation_time) {
362 if (node.comm_receive == NULL) {
363 task_received = NULL;
364 node.comm_receive = MSG_task_irecv(&task_received, node.mailbox);
365 // FIXME: do not make MSG_task_irecv() calls from several functions
368 //XBT_INFO("Node %d is ring member : %d", node.id, is_ring_member(known_id, node.id) != -1);
370 if (!MSG_comm_test(node.comm_receive)) {
372 // no task was received: make some periodic calls
374 if(MC_is_active() || MC_record_replay_is_active()){
375 if(MC_is_active() && !MC_visited_reduction() && no_op){
378 if(listen == 0 && (sub_protocol = MC_random(0, 4)) > 0){
379 if(sub_protocol == 1)
381 else if(sub_protocol == 2)
383 else if(sub_protocol == 3)
384 check_predecessor(&node);
386 random_lookup(&node);
389 MSG_process_sleep(sleep_delay);
390 if(!MC_visited_reduction())
394 if (MSG_get_clock() >= next_stabilize_date) {
396 next_stabilize_date = MSG_get_clock() + periodic_stabilize_delay;
397 }else if (MSG_get_clock() >= next_fix_fingers_date) {
399 next_fix_fingers_date = MSG_get_clock() + periodic_fix_fingers_delay;
400 }else if (MSG_get_clock() >= next_check_predecessor_date) {
401 check_predecessor(&node);
402 next_check_predecessor_date = MSG_get_clock() + periodic_check_predecessor_delay;
403 }else if (MSG_get_clock() >= next_lookup_date) {
404 random_lookup(&node);
405 next_lookup_date = MSG_get_clock() + periodic_lookup_delay;
407 // nothing to do: sleep for a while
408 MSG_process_sleep(sleep_delay);
413 // a transfer has occurred
415 msg_error_t status = MSG_comm_get_status(node.comm_receive);
417 if (status != MSG_OK) {
418 XBT_DEBUG("Failed to receive a task. Nevermind.");
419 MSG_comm_destroy(node.comm_receive);
420 node.comm_receive = NULL;
423 // the task was successfully received
424 MSG_comm_destroy(node.comm_receive);
425 node.comm_receive = NULL;
426 handle_task(&node, task_received);
431 if (node.comm_receive) {
432 /* handle last task if any */
433 if (MSG_comm_wait(node.comm_receive, 0) == MSG_OK)
434 task_free(task_received);
435 MSG_comm_destroy(node.comm_receive);
436 node.comm_receive = NULL;
443 // stop the simulation
444 xbt_free(node.fingers);
449 * \brief This function is called when the current node receives a task.
450 * \param node the current node
451 * \param task the task to handle (don't touch it then:
452 * it will be destroyed, reused or forwarded)
454 static void handle_task(node_t node, msg_task_t task) {
456 XBT_DEBUG("Handling task %p", task);
457 char mailbox[MAILBOX_NAME_SIZE];
458 task_data_t task_data = (task_data_t) MSG_task_get_data(task);
459 e_task_type_t type = task_data->type;
463 case TASK_FIND_SUCCESSOR:
464 XBT_DEBUG("Receiving a 'Find Successor' request from %s for id %d",
465 task_data->issuer_host_name, task_data->request_id);
466 // is my successor the successor?
467 if (is_in_interval(task_data->request_id, node->id + 1, node->fingers[0].id)) {
468 task_data->type = TASK_FIND_SUCCESSOR_ANSWER;
469 task_data->answer_id = node->fingers[0].id;
470 XBT_DEBUG("Sending back a 'Find Successor Answer' to %s (mailbox %s): the successor of %d is %d",
471 task_data->issuer_host_name,
472 task_data->answer_to,
473 task_data->request_id, task_data->answer_id);
474 MSG_task_dsend(task, task_data->answer_to, task_free);
477 // otherwise, forward the request to the closest preceding finger in my table
478 int closest = closest_preceding_node(node, task_data->request_id);
479 XBT_DEBUG("Forwarding the 'Find Successor' request for id %d to my closest preceding finger %d",
480 task_data->request_id, closest);
481 get_mailbox(closest, mailbox);
482 MSG_task_dsend(task, mailbox, task_free);
486 case TASK_GET_PREDECESSOR:
487 XBT_DEBUG("Receiving a 'Get Predecessor' request from %s", task_data->issuer_host_name);
488 task_data->type = TASK_GET_PREDECESSOR_ANSWER;
489 task_data->answer_id = node->pred_id;
490 XBT_DEBUG("Sending back a 'Get Predecessor Answer' to %s via mailbox '%s': my predecessor is %d",
491 task_data->issuer_host_name,
492 task_data->answer_to, task_data->answer_id);
493 MSG_task_dsend(task, task_data->answer_to, task_free);
497 // someone is telling me that he may be my new predecessor
498 XBT_DEBUG("Receiving a 'Notify' request from %s", task_data->issuer_host_name);
499 notify(node, task_data->request_id);
503 case TASK_PREDECESSOR_LEAVING:
504 // my predecessor is about to quit
505 XBT_DEBUG("Receiving a 'Predecessor Leaving' message from %s", task_data->issuer_host_name);
506 // modify my predecessor
507 set_predecessor(node, task_data->request_id);
510 >> notify my new predecessor
511 >> send a notify_predecessors !!
515 case TASK_SUCCESSOR_LEAVING:
516 // my successor is about to quit
517 XBT_DEBUG("Receiving a 'Successor Leaving' message from %s", task_data->issuer_host_name);
518 // modify my successor FIXME : this should be implicit ?
519 set_finger(node, 0, task_data->request_id);
522 >> notify my new successor
523 >> update my table & predecessors table */
526 case TASK_FIND_SUCCESSOR_ANSWER:
527 case TASK_GET_PREDECESSOR_ANSWER:
528 case TASK_PREDECESSOR_ALIVE_ANSWER:
529 XBT_DEBUG("Ignoring unexpected task of type %d (%p)", (int)type, task);
533 case TASK_PREDECESSOR_ALIVE:
534 XBT_DEBUG("Receiving a 'Predecessor Alive' request from %s", task_data->issuer_host_name);
535 task_data->type = TASK_PREDECESSOR_ALIVE_ANSWER;
536 XBT_DEBUG("Sending back a 'Predecessor Alive Answer' to %s (mailbox %s)",
537 task_data->issuer_host_name,
538 task_data->answer_to);
539 MSG_task_dsend(task, task_data->answer_to, task_free);
548 * \brief Initializes the current node as the first one of the system.
549 * \param node the current node
551 static void create(node_t node)
553 XBT_DEBUG("Create a new Chord ring...");
554 set_predecessor(node, -1); // -1 means that I have no predecessor
555 print_finger_table(node);
559 * \brief Makes the current node join the ring, knowing the id of a node
560 * already in the ring
561 * \param node the current node
562 * \param known_id id of a node already in the ring
563 * \return 1 if the join operation succeeded, 0 otherwise
565 static int join(node_t node, int known_id)
567 XBT_INFO("Joining the ring with id %d, knowing node %d", node->id, known_id);
568 set_predecessor(node, -1); // no predecessor (yet)
572 for (i = 0; i < nb_bits; i++) {
573 set_finger(node, i, known_id);
577 int successor_id = remote_find_successor(node, known_id, node->id);
578 if (successor_id == -1) {
579 XBT_INFO("Cannot join the ring.");
582 set_finger(node, 0, successor_id);
583 print_finger_table(node);
586 return successor_id != -1;
590 * \brief Makes the current node quit the system
591 * \param node the current node
593 static void leave(node_t node)
595 XBT_DEBUG("Well Guys! I Think it's time for me to quit ;)");
600 * \brief Notifies the successor and the predecessor of the current node
602 * \param node the current node
604 static void quit_notify(node_t node)
606 char mailbox[MAILBOX_NAME_SIZE];
607 //send the PREDECESSOR_LEAVING to our successor
608 task_data_t req_data = xbt_new0(s_task_data_t,1);
609 req_data->type = TASK_PREDECESSOR_LEAVING;
610 req_data->request_id = node->pred_id;
611 get_mailbox(node->id, req_data->answer_to);
612 req_data->issuer_host_name = MSG_host_get_name(MSG_host_self());
614 msg_task_t task_sent = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data);
615 XBT_DEBUG("Sending a 'PREDECESSOR_LEAVING' to my successor %d",node->fingers[0].id);
616 if (MSG_task_send_with_timeout(task_sent, node->fingers[0].mailbox, timeout)==
618 XBT_DEBUG("Timeout expired when sending a 'PREDECESSOR_LEAVING' to my successor %d",
619 node->fingers[0].id);
620 task_free(task_sent);
623 //send the SUCCESSOR_LEAVING to our predecessor
624 get_mailbox(node->pred_id, mailbox);
625 task_data_t req_data_s = xbt_new0(s_task_data_t,1);
626 req_data_s->type = TASK_SUCCESSOR_LEAVING;
627 req_data_s->request_id = node->fingers[0].id;
628 req_data_s->request_id = node->pred_id;
629 get_mailbox(node->id, req_data_s->answer_to);
630 req_data_s->issuer_host_name = MSG_host_get_name(MSG_host_self());
632 msg_task_t task_sent_s = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data_s);
633 XBT_DEBUG("Sending a 'SUCCESSOR_LEAVING' to my predecessor %d",node->pred_id);
634 if (MSG_task_send_with_timeout(task_sent_s, mailbox, timeout)==
636 XBT_DEBUG("Timeout expired when sending a 'SUCCESSOR_LEAVING' to my predecessor %d",
638 task_free(task_sent_s);
644 * \brief Makes the current node find the successor node of an id.
645 * \param node the current node
646 * \param id the id to find
647 * \return the id of the successor node, or -1 if the request failed
649 static int find_successor(node_t node, int id)
651 // is my successor the successor?
652 if (is_in_interval(id, node->id + 1, node->fingers[0].id)) {
653 return node->fingers[0].id;
656 // otherwise, ask the closest preceding finger in my table
657 int closest = closest_preceding_node(node, id);
658 return remote_find_successor(node, closest, id);
662 * \brief Asks another node the successor node of an id.
663 * \param node the current node
664 * \param ask_to the node to ask to
665 * \param id the id to find
666 * \return the id of the successor node, or -1 if the request failed
668 static int remote_find_successor(node_t node, int ask_to, int id)
672 char mailbox[MAILBOX_NAME_SIZE];
673 get_mailbox(ask_to, mailbox);
674 task_data_t req_data = xbt_new0(s_task_data_t, 1);
675 req_data->type = TASK_FIND_SUCCESSOR;
676 req_data->request_id = id;
677 get_mailbox(node->id, req_data->answer_to);
678 req_data->issuer_host_name = MSG_host_get_name(MSG_host_self());
680 // send a "Find Successor" request to ask_to_id
681 msg_task_t task_sent = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data);
682 XBT_DEBUG("Sending a 'Find Successor' request (task %p) to %d for id %d", task_sent, ask_to, id);
683 msg_error_t res = MSG_task_send_with_timeout(task_sent, mailbox, timeout);
686 XBT_DEBUG("Failed to send the 'Find Successor' request (task %p) to %d for id %d",
687 task_sent, ask_to, id);
688 task_free(task_sent);
692 // receive the answer
693 XBT_DEBUG("Sent a 'Find Successor' request (task %p) to %d for key %d, waiting for the answer",
694 task_sent, ask_to, id);
697 if (node->comm_receive == NULL) {
698 msg_task_t task_received = NULL;
699 node->comm_receive = MSG_task_irecv(&task_received, node->mailbox);
702 res = MSG_comm_wait(node->comm_receive, timeout);
705 XBT_DEBUG("Failed to receive the answer to my 'Find Successor' request (task %p): %d",
706 task_sent, (int)res);
708 MSG_comm_destroy(node->comm_receive);
709 node->comm_receive = NULL;
712 msg_task_t task_received = MSG_comm_get_task(node->comm_receive);
713 XBT_DEBUG("Received a task (%p)", task_received);
714 task_data_t ans_data = MSG_task_get_data(task_received);
716 // Once upon a time, our code assumed that here, task_received != task_sent all the time
718 // This assumption is wrong (as messages from differing round can interleave), leading to a bug in our code.
719 // We failed to find this bug directly, as it only occured on large platforms, leading to hardly usable traces.
720 // Instead, we used the model-checker to track down the issue by adding the following test here in the code:
721 // if (MC_is_active()) {
722 // MC_assert(task_received == task_sent);
724 // That explained the bug in a snap, with a very cool example and everything.
726 // This MC_assert is now desactivated as the case is now properly handled in our code and we don't want the
727 // MC to fail any further under that condition, but this comment is here to as a memorial for this first
728 // brillant victory of the model-checking in the SimGrid community :)
730 if (task_received != task_sent ||
731 ans_data->type != TASK_FIND_SUCCESSOR_ANSWER) {
732 // this is not the expected answer
733 MSG_comm_destroy(node->comm_receive);
734 node->comm_receive = NULL;
735 handle_task(node, task_received);
738 // this is our answer
739 XBT_DEBUG("Received the answer to my 'Find Successor' request for id %d (task %p): the successor of key %d is %d",
740 ans_data->request_id, task_received, id, ans_data->answer_id);
741 successor = ans_data->answer_id;
743 MSG_comm_destroy(node->comm_receive);
744 node->comm_receive = NULL;
745 task_free(task_received);
755 * \brief Asks another node its predecessor.
756 * \param node the current node
757 * \param ask_to the node to ask to
758 * \return the id of its predecessor node, or -1 if the request failed
759 * (or if the node does not know its predecessor)
761 static int remote_get_predecessor(node_t node, int ask_to)
763 int predecessor_id = -1;
765 char mailbox[MAILBOX_NAME_SIZE];
766 get_mailbox(ask_to, mailbox);
767 task_data_t req_data = xbt_new0(s_task_data_t, 1);
768 req_data->type = TASK_GET_PREDECESSOR;
769 get_mailbox(node->id, req_data->answer_to);
770 req_data->issuer_host_name = MSG_host_get_name(MSG_host_self());
772 // send a "Get Predecessor" request to ask_to_id
773 XBT_DEBUG("Sending a 'Get Predecessor' request to %d", ask_to);
774 msg_task_t task_sent = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data);
775 msg_error_t res = MSG_task_send_with_timeout(task_sent, mailbox, timeout);
778 XBT_DEBUG("Failed to send the 'Get Predecessor' request (task %p) to %d",
780 task_free(task_sent);
784 // receive the answer
785 XBT_DEBUG("Sent 'Get Predecessor' request (task %p) to %d, waiting for the answer on my mailbox '%s'",
786 task_sent, ask_to, req_data->answer_to);
789 if (node->comm_receive == NULL) { // FIXME simplify this
790 msg_task_t task_received = NULL;
791 node->comm_receive = MSG_task_irecv(&task_received, node->mailbox);
794 res = MSG_comm_wait(node->comm_receive, timeout);
797 XBT_DEBUG("Failed to receive the answer to my 'Get Predecessor' request (task %p): %d",
798 task_sent, (int)res);
800 MSG_comm_destroy(node->comm_receive);
801 node->comm_receive = NULL;
804 msg_task_t task_received = MSG_comm_get_task(node->comm_receive);
805 task_data_t ans_data = MSG_task_get_data(task_received);
807 /*if (MC_is_active()) {
808 MC_assert(task_received == task_sent);
811 if (task_received != task_sent ||
812 ans_data->type != TASK_GET_PREDECESSOR_ANSWER) {
813 MSG_comm_destroy(node->comm_receive);
814 node->comm_receive = NULL;
815 handle_task(node, task_received);
818 XBT_DEBUG("Received the answer to my 'Get Predecessor' request (task %p): the predecessor of node %d is %d",
819 task_received, ask_to, ans_data->answer_id);
820 predecessor_id = ans_data->answer_id;
822 MSG_comm_destroy(node->comm_receive);
823 node->comm_receive = NULL;
824 task_free(task_received);
830 return predecessor_id;
834 * \brief Returns the closest preceding finger of an id
835 * with respect to the finger table of the current node.
836 * \param node the current node
837 * \param id the id to find
838 * \return the closest preceding finger of that id
840 int closest_preceding_node(node_t node, int id)
843 for (i = nb_bits - 1; i >= 0; i--) {
844 if (is_in_interval(node->fingers[i].id, node->id + 1, id - 1)) {
845 return node->fingers[i].id;
852 * \brief This function is called periodically. It checks the immediate
853 * successor of the current node.
854 * \param node the current node
856 static void stabilize(node_t node)
858 XBT_DEBUG("Stabilizing node");
860 // get the predecessor of my immediate successor
862 int successor_id = node->fingers[0].id;
863 if (successor_id != node->id) {
864 candidate_id = remote_get_predecessor(node, successor_id);
867 candidate_id = node->pred_id;
870 // this node is a candidate to become my new successor
871 if (candidate_id != -1
872 && is_in_interval(candidate_id, node->id + 1, successor_id - 1)) {
873 set_finger(node, 0, candidate_id);
875 if (successor_id != node->id) {
876 remote_notify(node, successor_id, node->id);
881 * \brief Notifies the current node that its predecessor may have changed.
882 * \param node the current node
883 * \param candidate_id the possible new predecessor
885 static void notify(node_t node, int predecessor_candidate_id) {
887 if (node->pred_id == -1
888 || is_in_interval(predecessor_candidate_id, node->pred_id + 1, node->id - 1)) {
890 set_predecessor(node, predecessor_candidate_id);
891 print_finger_table(node);
894 XBT_DEBUG("I don't have to change my predecessor to %d", predecessor_candidate_id);
899 * \brief Notifies a remote node that its predecessor may have changed.
900 * \param node the current node
901 * \param notify_id id of the node to notify
902 * \param candidate_id the possible new predecessor
904 static void remote_notify(node_t node, int notify_id, int predecessor_candidate_id) {
906 task_data_t req_data = xbt_new0(s_task_data_t, 1);
907 req_data->type = TASK_NOTIFY;
908 req_data->request_id = predecessor_candidate_id;
909 req_data->issuer_host_name = MSG_host_get_name(MSG_host_self());
911 // send a "Notify" request to notify_id
912 msg_task_t task = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data);
913 XBT_DEBUG("Sending a 'Notify' request (task %p) to %d", task, notify_id);
914 char mailbox[MAILBOX_NAME_SIZE];
915 get_mailbox(notify_id, mailbox);
916 MSG_task_dsend(task, mailbox, task_free);
920 * \brief This function is called periodically.
921 * It refreshes the finger table of the current node.
922 * \param node the current node
924 static void fix_fingers(node_t node) {
926 XBT_DEBUG("Fixing fingers");
927 int i = node->next_finger_to_fix;
928 int id = find_successor(node, node->id + powers2[i]);
931 if (id != node->fingers[i].id) {
932 set_finger(node, i, id);
933 print_finger_table(node);
935 node->next_finger_to_fix = (i + 1) % nb_bits;
940 * \brief This function is called periodically.
941 * It checks whether the predecessor has failed
942 * \param node the current node
944 static void check_predecessor(node_t node)
946 XBT_DEBUG("Checking whether my predecessor is alive");
948 if(node->pred_id == -1)
953 char mailbox[MAILBOX_NAME_SIZE];
954 get_mailbox(node->pred_id, mailbox);
955 task_data_t req_data = xbt_new0(s_task_data_t,1);
956 req_data->type = TASK_PREDECESSOR_ALIVE;
957 req_data->request_id = node->pred_id;
958 get_mailbox(node->id, req_data->answer_to);
959 req_data->issuer_host_name = MSG_host_get_name(MSG_host_self());
961 msg_task_t task_sent = MSG_task_create(NULL, COMP_SIZE, COMM_SIZE, req_data);
962 XBT_DEBUG("Sending a 'Predecessor Alive' request to my predecessor %d", node->pred_id);
964 msg_error_t res = MSG_task_send_with_timeout(task_sent, mailbox, timeout);
967 XBT_DEBUG("Failed to send the 'Predecessor Alive' request (task %p) to %d", task_sent, node->pred_id);
968 task_free(task_sent);
971 // receive the answer
972 XBT_DEBUG("Sent 'Predecessor Alive' request (task %p) to %d, waiting for the answer on my mailbox '%s'",
973 task_sent, node->pred_id, req_data->answer_to);
976 if (node->comm_receive == NULL) { // FIXME simplify this
977 msg_task_t task_received = NULL;
978 node->comm_receive = MSG_task_irecv(&task_received, node->mailbox);
981 res = MSG_comm_wait(node->comm_receive, timeout);
984 XBT_DEBUG("Failed to receive the answer to my 'Predecessor Alive' request (task %p): %d",
985 task_sent, (int)res);
987 MSG_comm_destroy(node->comm_receive);
988 node->comm_receive = NULL;
991 msg_task_t task_received = MSG_comm_get_task(node->comm_receive);
992 if (task_received != task_sent) {
993 MSG_comm_destroy(node->comm_receive);
994 node->comm_receive = NULL;
995 handle_task(node, task_received);
997 XBT_DEBUG("Received the answer to my 'Predecessor Alive' request (task %p) : my predecessor %d is alive", task_received, node->pred_id);
999 MSG_comm_destroy(node->comm_receive);
1000 node->comm_receive = NULL;
1001 task_free(task_received);
1009 * \brief Performs a find successor request to a random id.
1010 * \param node the current node
1012 static void random_lookup(node_t node)
1014 int random_index = RngStream_RandInt (node->stream, 0, nb_bits - 1);
1015 int random_id = node->fingers[random_index].id;
1016 XBT_DEBUG("Making a lookup request for id %d", random_id);
1017 int res = find_successor(node, random_id);
1018 XBT_DEBUG("The successor of node %d is %d", random_id, res);
1023 * \brief Main function.
1025 int main(int argc, char *argv[])
1027 MSG_init(&argc, argv);
1029 printf("Usage: %s [-nb_bits=n] [-timeout=t] platform_file deployment_file\n", argv[0]);
1030 printf("example: %s ../msg_platform.xml chord.xml\n", argv[0]);
1034 char **options = &argv[1];
1035 while (!strncmp(options[0], "-", 1)) {
1037 int length = strlen("-nb_bits=");
1038 if (!strncmp(options[0], "-nb_bits=", length) && strlen(options[0]) > length) {
1039 nb_bits = atoi(options[0] + length);
1040 XBT_DEBUG("Set nb_bits to %d", nb_bits);
1044 length = strlen("-timeout=");
1045 if (!strncmp(options[0], "-timeout=", length) && strlen(options[0]) > length) {
1046 timeout = atoi(options[0] + length);
1047 XBT_DEBUG("Set timeout to %d", timeout);
1050 xbt_die("Invalid chord option '%s'", options[0]);
1056 const char* platform_file = options[0];
1057 const char* application_file = options[1];
1059 MSG_create_environment(platform_file);
1063 MSG_function_register("node", node);
1064 MSG_launch_application(application_file);
1066 msg_error_t res = MSG_main();
1067 XBT_CRITICAL("Messages created: %ld", smx_total_comms);
1068 XBT_INFO("Simulated time: %g", MSG_get_clock());