-/* $Id$ */
+/* amok_bandwidth - Bandwidth tests facilities */
-/* amok_bandwidth - Bandwidth tests facilities */
-
-/* Authors: Martin Quinson */
-/* Copyright (C) 2003, 2004 the OURAGAN project. */
+/* Copyright (c) 2004, 2005, 2006, 2007, 2009, 2010. The SimGrid Team.
+ * All rights reserved. */
/* This program is free software; you can redistribute it and/or modify it
- under the terms of the license (GNU LGPL) which comes with this package. */
+ * under the terms of the license (GNU LGPL) which comes with this package. */
+#include "xbt/ex.h"
#include "amok/Bandwidth/bandwidth_private.h"
#include "gras/messages.h"
-XBT_LOG_NEW_DEFAULT_SUBCATEGORY(bw,amok,"Bandwidth testing");
+XBT_LOG_NEW_DEFAULT_SUBCATEGORY(amok_bw, amok, "Bandwidth testing");
+
+
+/******************************
+ * Stuff global to the module *
+ ******************************/
static short _amok_bw_initialized = 0;
-/**** code ****/
-void amok_bw_init(void) {
- xbt_error_t errcode;
- gras_datadesc_type_t bw_request_desc, bw_res_desc, sat_request_desc;
-
- if (_amok_bw_initialized)
- return;
-
- amok_base_init();
-
- /* Build the datatype descriptions */
- bw_request_desc = gras_datadesc_struct("s_bw_request_t");
- gras_datadesc_struct_append(bw_request_desc,"host",gras_datadesc_by_name("xbt_host_t"));
- gras_datadesc_struct_append(bw_request_desc,"buf_size",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_append(bw_request_desc,"exp_size",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_append(bw_request_desc,"msg_size",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_close(bw_request_desc);
- bw_request_desc = gras_datadesc_ref("bw_request_t",bw_request_desc);
-
- bw_res_desc = gras_datadesc_struct("s_bw_res_t");
- gras_datadesc_struct_append(bw_res_desc,"err",gras_datadesc_by_name("s_amok_remoterr_t"));
- gras_datadesc_struct_append(bw_res_desc,"timestamp",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_append(bw_res_desc,"seconds",gras_datadesc_by_name("double"));
- gras_datadesc_struct_append(bw_res_desc,"bw",gras_datadesc_by_name("double"));
- gras_datadesc_struct_close(bw_res_desc);
- bw_res_desc = gras_datadesc_ref("bw_res_t",bw_res_desc);
-
- sat_request_desc = gras_datadesc_struct("s_sat_request_desc_t");
- gras_datadesc_struct_append(sat_request_desc,"host",gras_datadesc_by_name("xbt_host_t"));
- gras_datadesc_struct_append(sat_request_desc,"msg_size",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_append(sat_request_desc,"timeout",gras_datadesc_by_name("unsigned int"));
- gras_datadesc_struct_close(sat_request_desc);
- sat_request_desc = gras_datadesc_ref("sat_request_t",sat_request_desc);
-
- /* Register the bandwidth messages */
- gras_msgtype_declare("BW request", bw_request_desc);
- gras_msgtype_declare("BW result", bw_res_desc);
- gras_msgtype_declare("BW handshake", bw_request_desc);
- gras_msgtype_declare("BW handshake ACK", bw_request_desc);
-
- /* Register the saturation messages */
- gras_msgtype_declare("SAT start", sat_request_desc);
- gras_msgtype_declare("SAT started", gras_datadesc_by_name("amok_remoterr_t"));
- gras_msgtype_declare("SAT begin", sat_request_desc);
- gras_msgtype_declare("SAT begun", gras_datadesc_by_name("amok_remoterr_t"));
- gras_msgtype_declare("SAT end", NULL);
- gras_msgtype_declare("SAT ended", gras_datadesc_by_name("amok_remoterr_t"));
- gras_msgtype_declare("SAT stop", NULL);
- gras_msgtype_declare("SAT stopped", gras_datadesc_by_name("amok_remoterr_t"));
-
- /* Register the callbacks */
- gras_cb_register(gras_msgtype_by_name("BW request"),
- &amok_bw_cb_bw_request);
- gras_cb_register(gras_msgtype_by_name("BW handshake"),
- &amok_bw_cb_bw_handshake);
-
- gras_cb_register(gras_msgtype_by_name("SAT start"),
- &amok_bw_cb_sat_start);
- gras_cb_register(gras_msgtype_by_name("SAT begin"),
- &amok_bw_cb_sat_begin);
-
- _amok_bw_initialized =1;
+/** @brief module initialization; all participating nodes must run this */
+void amok_bw_init(void)
+{
+
+ if (!_amok_bw_initialized) {
+ amok_bw_bw_init();
+ amok_bw_sat_init();
+ }
+
+ amok_bw_bw_join();
+ amok_bw_sat_join();
+
+ _amok_bw_initialized++;
}
-void amok_bw_exit(void) {
- if (! _amok_bw_initialized)
+/** @brief module finalization */
+void amok_bw_exit(void)
+{
+ if (!_amok_bw_initialized)
return;
-
- gras_cb_unregister(gras_msgtype_by_name("BW request"),
- &amok_bw_cb_bw_request);
- gras_cb_unregister(gras_msgtype_by_name("BW handshake"),
- &amok_bw_cb_bw_handshake);
-
- gras_cb_unregister(gras_msgtype_by_name("SAT start"),
- &amok_bw_cb_sat_start);
- gras_cb_unregister(gras_msgtype_by_name("SAT begin"),
- &amok_bw_cb_sat_begin);
-
- _amok_bw_initialized = 0;
-}
-
+ amok_bw_bw_leave();
+ amok_bw_sat_leave();
+
+ _amok_bw_initialized--;
+}
/* ***************************************************************************
* Bandwidth tests
* ***************************************************************************/
+static int amok_bw_cb_bw_handshake(gras_msg_cb_ctx_t ctx, void *payload);
+static int amok_bw_cb_bw_request(gras_msg_cb_ctx_t ctx, void *payload);
+
+void amok_bw_bw_init()
+{
+ xbt_datadesc_type_t bw_request_desc, bw_res_desc;
+
+ /* Build the Bandwidth datatype descriptions */
+ bw_request_desc = xbt_datadesc_struct("s_bw_request_t");
+ xbt_datadesc_struct_append(bw_request_desc, "peer",
+ xbt_datadesc_by_name("s_xbt_peer_t"));
+ xbt_datadesc_struct_append(bw_request_desc, "buf_size",
+ xbt_datadesc_by_name("unsigned long int"));
+ xbt_datadesc_struct_append(bw_request_desc, "msg_size",
+ xbt_datadesc_by_name("unsigned long int"));
+ xbt_datadesc_struct_append(bw_request_desc, "msg_amount",
+ xbt_datadesc_by_name("unsigned long int"));
+ xbt_datadesc_struct_append(bw_request_desc, "min_duration",
+ xbt_datadesc_by_name("double"));
+ xbt_datadesc_struct_close(bw_request_desc);
+ bw_request_desc = xbt_datadesc_ref("bw_request_t", bw_request_desc);
+
+ bw_res_desc = xbt_datadesc_struct("s_bw_res_t");
+ xbt_datadesc_struct_append(bw_res_desc, "timestamp",
+ xbt_datadesc_by_name("unsigned int"));
+ xbt_datadesc_struct_append(bw_res_desc, "seconds",
+ xbt_datadesc_by_name("double"));
+ xbt_datadesc_struct_append(bw_res_desc, "bw",
+ xbt_datadesc_by_name("double"));
+ xbt_datadesc_struct_close(bw_res_desc);
+ bw_res_desc = xbt_datadesc_ref("bw_res_t", bw_res_desc);
+
+ gras_msgtype_declare_rpc("BW handshake", bw_request_desc,
+ bw_request_desc);
+
+ gras_msgtype_declare_rpc("BW reask", bw_request_desc, NULL);
+ gras_msgtype_declare("BW stop", NULL);
+
+ gras_msgtype_declare_rpc("BW request", bw_request_desc, bw_res_desc);
+}
+
+void amok_bw_bw_join()
+{
+ gras_cb_register("BW request", &amok_bw_cb_bw_request);
+ gras_cb_register("BW handshake", &amok_bw_cb_bw_handshake);
+}
+
+void amok_bw_bw_leave()
+{
+ gras_cb_unregister("BW request", &amok_bw_cb_bw_request);
+ gras_cb_unregister("BW handshake", &amok_bw_cb_bw_handshake);
+}
/**
- * amok_bw_test:
+ * \brief bandwidth measurement between localhost and \e peer
+ *
+ * \arg peer: A (regular) socket at which the the host with which we should conduct the experiment can be contacted
+ * \arg buf_size: Size of the socket buffer. If 0, a sain default is used (32k, but may change)
+ * \arg msg_size: Size of each message sent.
+ * \arg msg_amount: Amount of such messages to exchange
+ * \arg min_duration: The minimum wanted duration. When the test message is too little, you tend to measure the latency. This argument allows you to force the test to take at least, say one second.
+ * \arg sec: where the result (in seconds) should be stored. If the experiment was done several times because the first one was too short, this is the timing of the last run only.
+ * \arg bw: observed Bandwidth (in byte/s)
+ *
+ * Conduct a bandwidth test from the local process to the given peer.
+ * This call is blocking until the end of the experiment.
+ *
+ * If the asked experiment lasts less than \a min_duration, another one will be
+ * launched (and others, if needed). msg_size will be multiplicated by
+ * MIN(20, (\a min_duration / measured_duration) *1.1) (plus 10% to be sure to eventually
+ * reach the \a min_duration). In that case, the reported bandwidth and
+ * duration are the ones of the last run. \a msg_size cannot go over 64Mb
+ * because we need to malloc a block of this size in RL to conduct the
+ * experiment, and we still don't want to visit the swap. In such case, the
+ * number of messages is increased instead of their size.
+ *
+ * Results are reported in last args, and sizes are in byte.
+ *
+ * @warning: in SimGrid version 3.1 and previous, the experiment size were specified
+ * as the total amount of data to send and the msg_size. This
+ * was changed for the fool wanting to send more than MAXINT
+ * bytes in a fat pipe.
*
- * Conduct a test between the local host and @peer, and
- * report the result in last args
*/
-xbt_error_t amok_bw_test(gras_socket_t peer,
- unsigned int buf_size,unsigned int exp_size,unsigned int msg_size,
- /*OUT*/ double *sec, double *bw) {
- gras_socket_t rawIn,rawOut; /* raw sockets for the experiments */
- gras_socket_t sock_dummy; /* ignored arg to msg_wait */
- int port;
- xbt_error_t errcode;
- bw_request_t request,request_ack;
-
- for (port = 5000, errcode = system_error;
- errcode == system_error;
- errcode = gras_socket_server_ext(++port,buf_size,1,&rawIn));
- if (errcode != no_error) {
- ERROR1("Error %s encountered while opening a raw socket",
- xbt_error_name(errcode));
- return errcode;
- }
-
- request=xbt_new0(s_bw_request_t,1);
- request->buf_size=buf_size;
- request->exp_size=exp_size;
- request->msg_size=msg_size;
- request->host.name = NULL;
- request->host.port = gras_socket_my_port(rawIn);
- INFO1("Send an handshake to get the dude connect to port %d on me", request->host.port);
-
- if ((errcode=gras_msg_send(peer,gras_msgtype_by_name("BW handshake"),&request))) {
- ERROR1("Error %s encountered while sending the BW request.", xbt_error_name(errcode));
- return errcode;
- }
- if ((errcode=gras_msg_wait(60,gras_msgtype_by_name("BW handshake ACK"),&sock_dummy,&request_ack))) {
- ERROR1("Error %s encountered while waiting for the answer to BW request.\n",
- xbt_error_name(errcode));
- return errcode;
- }
-
- /* FIXME: What if there is a remote error? */
-
- if((errcode=gras_socket_client_ext(gras_socket_peer_name(peer),request_ack->host.port, buf_size,1,&rawOut))) {
- ERROR3("Error %s encountered while opening the raw socket to %s:%d for BW test\n",
- xbt_error_name(errcode),gras_socket_peer_name(peer),request_ack->host.port);
- return errcode;
- }
- xbt_free(request_ack);
- INFO0("Got ACK");
-
- *sec=gras_os_time();
- if ((errcode=gras_socket_raw_send(rawOut,120,exp_size,msg_size)) ||
- (errcode=gras_socket_raw_recv(rawIn,120,1,1))) {
- ERROR1("Error %s encountered while sending the BW experiment.",
- xbt_error_name(errcode));
- gras_socket_close(rawOut);
- gras_socket_close(rawIn);
- return errcode;
- }
- *sec = gras_os_time() - *sec;
- *bw = ((double)exp_size /* 8.0*/) / *sec / (1024.0 *1024.0);
-INFO0("DOOONE");
-
- gras_socket_close(rawIn);
- gras_socket_close(rawOut);
- return no_error;
+void amok_bw_test(xbt_socket_t peer,
+ unsigned long int buf_size,
+ unsigned long int msg_size,
+ unsigned long int msg_amount,
+ double min_duration, /*OUT*/ double *sec, double *bw)
+{
+
+ /* Measurement sockets for the experiments */
+ volatile xbt_socket_t measMasterIn = NULL, measIn, measOut = NULL;
+ volatile int port;
+ bw_request_t request, request_ack;
+ xbt_ex_t e;
+ int first_pass;
+
+ for (port = 5000; port < 10000 && measMasterIn == NULL; port++) {
+ TRY {
+ measMasterIn = gras_socket_server_ext(++port, buf_size, 1);
+ }
+ CATCH(e) {
+ measMasterIn = NULL;
+ if (port == 10000 - 1) {
+ RETHROWF("Error caught while opening a measurement socket: %s");
+ } else {
+ xbt_ex_free(e);
+ }
+ }
+ }
+
+ request = xbt_new0(s_bw_request_t, 1);
+ request->buf_size = buf_size;
+ request->msg_size = msg_size;
+ request->msg_amount = msg_amount;
+ request->peer.name = NULL;
+ request->peer.port = xbt_socket_my_port(measMasterIn);
+ XBT_DEBUG
+ ("Handshaking with %s:%d to connect it back on my %d (bufsize=%lu, msg_size=%lu, msg_amount=%lu)",
+ xbt_socket_peer_name(peer), xbt_socket_peer_port(peer),
+ request->peer.port, request->buf_size, request->msg_size,
+ request->msg_amount);
+
+ TRY {
+ gras_msg_rpccall(peer, 15, "BW handshake", &request, &request_ack);
+ }
+ CATCH_ANONYMOUS {
+ RETHROWF("Error encountered while sending the BW request: %s");
+ }
+ measIn = xbt_socket_meas_accept(measMasterIn);
+
+ TRY {
+ measOut = gras_socket_client_ext(xbt_socket_peer_name(peer),
+ request_ack->peer.port,
+ request->buf_size, 1);
+ }
+ CATCH_ANONYMOUS {
+ RETHROWF
+ ("Error encountered while opening the measurement socket to %s:%d for BW test: %s",
+ xbt_socket_peer_name(peer), request_ack->peer.port);
+ }
+ XBT_DEBUG
+ ("Got ACK; conduct the experiment (msg_size = %lu, msg_amount=%lu)",
+ request->msg_size, request->msg_amount);
+
+ *sec = 0;
+ first_pass = 1;
+ do {
+ if (first_pass == 0) {
+ double meas_duration = *sec;
+ double increase;
+ if (*sec != 0.0) {
+ increase = (min_duration / meas_duration) * 1.1;
+ } else {
+ increase = 4;
+ }
+ /* Do not increase the exp size too fast since our decision would be based on wrong measurements */
+ if (increase > 20)
+ increase = 20;
+
+ request->msg_size = request->msg_size * increase;
+
+ /* Do not do too large experiments messages or the sensors
+ will start to swap to store one of them.
+ And then increase the number of messages to compensate (check for overflow there, too) */
+ if (request->msg_size > 64 * 1024 * 1024) {
+ unsigned long int new_amount =
+ ((request->msg_size / ((double) 64 * 1024 * 1024))
+ * request->msg_amount) + 1;
+
+ xbt_assert(new_amount > request->msg_amount,
+ "Overflow on the number of messages! You must have a *really* fat pipe. Please fix your platform");
+ request->msg_amount = new_amount;
+
+ request->msg_size = 64 * 1024 * 1024;
+ }
+
+ XBT_VERB
+ ("The experiment was too short (%f sec<%f sec). Redo it with msg_size=%lu (nb_messages=%lu) (got %fMb/s)",
+ meas_duration, min_duration, request->msg_size,
+ request->msg_amount,
+ ((double) request->msg_size) * ((double) request->msg_amount /
+ (*sec) / 1024.0 / 1024.0));
+
+ gras_msg_rpccall(peer, 60, "BW reask", &request, NULL);
+ }
+
+ first_pass = 0;
+ *sec = gras_os_time();
+ TRY {
+ xbt_socket_meas_send(measOut, 120, request->msg_size,
+ request->msg_amount);
+ XBT_DEBUG("Data sent. Wait ACK");
+ xbt_socket_meas_recv(measIn, 120, 1, 1);
+ }
+ CATCH_ANONYMOUS {
+ gras_socket_close(measOut);
+ gras_socket_close(measMasterIn);
+ gras_socket_close(measIn);
+ RETHROWF("Unable to conduct the experiment: %s");
+ }
+ *sec = gras_os_time() - *sec;
+ if (*sec != 0.0) {
+ *bw =
+ ((double) request->msg_size) * ((double) request->msg_amount) /
+ (*sec);
+ }
+ XBT_DEBUG("Experiment done ; it took %f sec", *sec);
+ if (*sec <= 0) {
+ XBT_CRITICAL("Nonpositive value (%f) found for BW test time.", *sec);
+ }
+
+ } while (*sec < min_duration);
+
+ XBT_DEBUG
+ ("This measurement was long enough (%f sec; found %f b/s). Stop peer",
+ *sec, *bw);
+ gras_msg_send(peer, "BW stop", NULL);
+
+ free(request_ack);
+ free(request);
+ if (measIn != measMasterIn)
+ gras_socket_close(measIn);
+ gras_socket_close(measMasterIn);
+ gras_socket_close(measOut);
}
/* Callback to the "BW handshake" message:
- opens a server raw socket,
+ opens a server measurement socket,
indicate its port in an "BW handshaked" message,
- receive the corresponding data on the raw socket,
- close the raw socket
+ receive the corresponding data on the measurement socket,
+ close the measurement socket
+
+ sizes are in byte
*/
-int amok_bw_cb_bw_handshake(gras_socket_t expeditor,
- void *payload) {
- gras_socket_t rawIn,rawOut;
- bw_request_t request=*(bw_request_t*)payload;
+int amok_bw_cb_bw_handshake(gras_msg_cb_ctx_t ctx, void *payload)
+{
+ xbt_socket_t expeditor = gras_msg_cb_ctx_from(ctx);
+ volatile xbt_socket_t measMasterIn = NULL, measIn = NULL, measOut = NULL;
+ volatile bw_request_t request = *(bw_request_t *) payload;
bw_request_t answer;
- xbt_error_t errcode;
+ xbt_ex_t e;
int port;
-
- INFO2("Got an handshake to connect to %s:%d",
- request->host.name,request->host.port);
-
- answer = xbt_new0(s_bw_request_t,1);
-
- for (port = 5000, errcode = system_error;
- errcode == system_error;
- errcode = gras_socket_server_ext(++port,request->buf_size,1,&rawIn));
- if (errcode != no_error) {
- ERROR1("Error %s encountered while opening a raw socket", xbt_error_name(errcode));
+ int tooshort = 1;
+ gras_msg_cb_ctx_t ctx_reask;
+ static xbt_dynar_t msgtwaited = NULL;
+
+ XBT_DEBUG
+ ("Handshaked to connect to %s:%d (sizes: buf=%lu msg=%lu msg_amount=%lu)",
+ xbt_socket_peer_name(expeditor), request->peer.port,
+ request->buf_size, request->msg_size, request->msg_amount);
+
+ /* Build our answer */
+ answer = xbt_new0(s_bw_request_t, 1);
+
+ for (port = 6000; port <= 10000 && measMasterIn == NULL; port++) {
+ TRY {
+ measMasterIn = gras_socket_server_ext(port, request->buf_size, 1);
+ }
+ CATCH(e) {
+ measMasterIn = NULL;
+ if (port < 10000)
+ xbt_ex_free(e);
+ else
+ /* FIXME: tell error to remote */
+ RETHROWF
+ ("Error encountered while opening a measurement server socket: %s");
+ }
+ }
+
+ answer->buf_size = request->buf_size;
+ answer->msg_size = request->msg_size;
+ answer->msg_amount = request->msg_amount;
+ answer->peer.port = xbt_socket_my_port(measMasterIn);
+
+ TRY {
+ gras_msg_rpcreturn(60, ctx, &answer);
+ }
+ CATCH_ANONYMOUS {
+ gras_socket_close(measMasterIn);
/* FIXME: tell error to remote */
- return 1;
- }
-
- if ((errcode=gras_socket_client_ext(gras_socket_peer_name(expeditor),request->host.port,
- request->buf_size,1,&rawOut))) {
- ERROR3("Error '%s' encountered while opening a raw socket to %s:%d",
- xbt_error_name(errcode),gras_socket_peer_name(expeditor),request->host.port);
- /* FIXME: tell error to remote */
- return 1;
- }
-
- answer->buf_size=request->buf_size;
- answer->exp_size=request->exp_size;
- answer->msg_size=request->msg_size;
- answer->host.port=gras_socket_my_port(rawIn);
-
- if ((errcode=gras_msg_send(expeditor,gras_msgtype_by_name("BW handshake ACK"),&answer))) {
- ERROR1("Error %s encountered while sending the answer.",
- xbt_error_name(errcode));
- gras_socket_close(rawIn);
- gras_socket_close(rawOut);
- /* FIXME: tell error to remote */
- return 1;
- }
- INFO4("BW handshake answered. buf_size=%d exp_size=%d msg_size=%d port=%d",
- answer->buf_size,answer->exp_size,answer->msg_size,answer->host.port);
-
- if ((errcode=gras_socket_raw_recv(rawIn, 120,request->exp_size,request->msg_size)) ||
- (errcode=gras_socket_raw_send(rawOut,120,1,1))) {
- ERROR1("Error %s encountered while receiving the experiment.",
- xbt_error_name(errcode));
- gras_socket_close(rawIn);
- gras_socket_close(rawOut);
- /* FIXME: tell error to remote ? */
- return 1;
- }
- gras_socket_close(rawIn);
- gras_socket_close(rawOut);
- return 1;
-}
-
-int amok_bw_cb_bw_request(gras_socket_t expeditor,
- void *payload) {return 1;}
-
-int amok_bw_cb_sat_start(gras_socket_t expeditor,
- void *payload) {return 1;}
-int amok_bw_cb_sat_begin(gras_socket_t expeditor,
- void *payload) {return 1;}
-
-#if 0
-/* function to request a BW test between two external hosts */
-xbt_error_t grasbw_request(const char* from_name,unsigned int from_port,
- const char* to_name,unsigned int to_port,
- unsigned int bufSize,unsigned int expSize,unsigned int msgSize,
- /*OUT*/ double *sec, double*bw) {
-
- gras_sock_t *sock;
- gras_msg_t *answer;
- xbt_error_t errcode;
- /* The request */
- BwExp_t *request;
- msgHost_t *target;
-
- if((errcode=gras_sock_client_open(from_name,from_port,&sock))) {
- fprintf(stderr,"grasbw_request(): Error %s encountered while contacting the actuator\n",
- xbt_error_name(errcode));
- return errcode;
- }
- if (!(request=(BwExp_t *)malloc(sizeof(BwExp_t))) ||
- !(target=(msgHost_t*)malloc(sizeof(msgHost_t)))) {
- fprintf(stderr,"grasbw_test(): Malloc error\n");
- gras_sock_close(sock);
- return malloc_error;
- }
-
- request->bufSize=bufSize;
- request->expSize=expSize;
- request->msgSize=msgSize;
- strcpy(target->host,to_name);
- target->port=to_port;
-
- if ((errcode=gras_msg_new_and_send(sock,GRASMSG_BW_REQUEST, 2,
- target,1,
- request,1))) {
- fprintf(stderr,"grasbw_request(): Error %s encountered while sending the request.\n",
- xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
- }
- if ((errcode=gras_msg_wait(240,GRASMSG_BW_RESULT,&answer))) {
- fprintf(stderr,"grasbw_request(): Error %s encountered while waiting for the answer.\n",
- xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
- }
-
- if((errcode=gras_msg_ctn(answer,0,0,msgError_t).errcode)) {
- fprintf(stderr,"grasbw_request(): Peer reported error %s (%s).\n",
- xbt_error_name(errcode),gras_msg_ctn(answer,0,0,msgError_t).errmsg);
- gras_msg_free(answer);
- gras_sock_close(sock);
- return errcode;
- }
-
- /* fprintf(stderr,"sec=%p",gras_msg_ctn(answer,1,0,msgResult_t)); */
- *sec=gras_msg_ctn(answer,1,0,msgResult_t).value;
- *bw=gras_msg_ctn(answer,1,1,msgResult_t).value;
-
- gras_msg_free(answer);
- gras_sock_close(sock);
- return no_error;
-}
-
-int grasbw_cbBWRequest(gras_msg_t *msg) {
- /* specification of the test to run */
- char* to_name=gras_msg_ctn(msg,0,0,msgHost_t).host;
- unsigned int to_port=gras_msg_ctn(msg,0,0,msgHost_t).port;
-
- unsigned int bufSize=gras_msg_ctn(msg,1,0,BwExp_t).bufSize;
- unsigned int expSize=gras_msg_ctn(msg,1,0,BwExp_t).expSize;
- unsigned int msgSize=gras_msg_ctn(msg,1,0,BwExp_t).msgSize;
- /* our answer */
- msgError_t *error;
- msgResult_t *res;
-
- if (!(error=(msgError_t *)malloc(sizeof(msgError_t))) ||
- !(res=(msgResult_t *)malloc(sizeof(msgResult_t) * 2))) {
- fprintf(stderr,"%s:%d:grasbw_cbRequest: Malloc error\n",__FILE__,__LINE__);
- return malloc_error;
- }
-
- if ((error->errcode=grasbw_test(to_name,to_port,bufSize,expSize,msgSize,
- &(res[0].value),&(res[1].value) ))) {
- fprintf(stderr,
- "%s:%d:grasbw_cbRequest: Error %s encountered while doing the test\n",
- __FILE__,__LINE__,xbt_error_name(error->errcode));
- strncpy(error->errmsg,"Error within grasbw_test",ERRMSG_LEN);
- gras_msg_new_and_send(msg->sock,GRASMSG_BW_RESULT,2,
- error,1,
- res,2);
- return 1;
+ RETHROWF("Error encountered while sending the answer: %s");
}
- res[0].timestamp = (unsigned int) gras_time();
- res[1].timestamp = (unsigned int) gras_time();
- gras_msg_new_and_send(msg->sock,GRASMSG_BW_RESULT,2,
- error,1,
- res,2);
- gras_msg_free(msg);
- return 1;
-}
-/* ***************************************************************************
- * Link saturation
- * ***************************************************************************/
-xbt_error_t grasbw_saturate_start(const char* from_name,unsigned int from_port,
- const char* to_name,unsigned int to_port,
- unsigned int msgSize, unsigned int timeout) {
- gras_sock_t *sock;
- xbt_error_t errcode;
- /* The request */
- SatExp_t *request;
- msgHost_t *target;
- /* answer */
- gras_msg_t *answer;
-
- if((errcode=gras_sock_client_open(from_name,from_port,&sock))) {
- fprintf(stderr,"%s:%d:saturate_start(): Error %s encountered while contacting peer\n",
- __FILE__,__LINE__,xbt_error_name(errcode));
- return errcode;
+ /* Don't connect asap to leave time to other side to enter the accept() */
+ TRY {
+ measOut = gras_socket_client_ext(xbt_socket_peer_name(expeditor),
+ request->peer.port,
+ request->buf_size, 1);
}
- if (!(request=(SatExp_t *)malloc(sizeof(SatExp_t))) ||
- !(target=(msgHost_t*)malloc(sizeof(msgHost_t)))) {
- fprintf(stderr,"%s:%d:saturate_start(): Malloc error\n",__FILE__,__LINE__);
- gras_sock_close(sock);
- return malloc_error;
- }
-
- request->timeout=timeout;
- request->msgSize=msgSize;
-
- strcpy(target->host,to_name);
- target->port=to_port;
-
- if ((errcode=gras_msg_new_and_send(sock,GRASMSG_SAT_START, 2,
- target,1,
- request,1))) {
- fprintf(stderr,"%s:%d:saturate_start(): Error %s encountered while sending the request.\n",
- __FILE__,__LINE__,xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
- }
- if ((errcode=gras_msg_wait(120,GRASMSG_SAT_STARTED,&answer))) {
- fprintf(stderr,"%s:%d:saturate_start(): Error %s encountered while waiting for the ACK.\n",
- __FILE__,__LINE__,xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
+ CATCH_ANONYMOUS {
+ RETHROWF
+ ("Error encountered while opening a measurement socket back to %s:%d : %s",
+ xbt_socket_peer_name(expeditor), request->peer.port);
+ /* FIXME: tell error to remote */
}
- if((errcode=gras_msg_ctn(answer,0,0,msgError_t).errcode)) {
- fprintf(stderr,"%s:%d:saturate_start(): Peer reported error %s (%s).\n",
- __FILE__,__LINE__,xbt_error_name(errcode),gras_msg_ctn(answer,0,0,msgError_t).errmsg);
- gras_msg_free(answer);
- gras_sock_close(sock);
- return errcode;
+ TRY {
+ measIn = xbt_socket_meas_accept(measMasterIn);
+ XBT_DEBUG
+ ("BW handshake answered. buf_size=%lu msg_size=%lu msg_amount=%lu port=%d",
+ answer->buf_size, answer->msg_size, answer->msg_amount,
+ answer->peer.port);
}
-
- gras_msg_free(answer);
- gras_sock_close(sock);
- return no_error;
+ CATCH_ANONYMOUS {
+ gras_socket_close(measMasterIn);
+ gras_socket_close(measIn);
+ gras_socket_close(measOut);
+ /* FIXME: tell error to remote ? */
+ RETHROWF("Error encountered while opening the meas socket: %s");
+ }
+
+ if (!msgtwaited) {
+ msgtwaited = xbt_dynar_new(sizeof(gras_msgtype_t), NULL);
+ xbt_dynar_push(msgtwaited, gras_msgtype_by_name("BW stop"));
+ xbt_dynar_push(msgtwaited, gras_msgtype_by_name("BW reask"));
+ }
+
+ while (tooshort) {
+ void *payloadgot;
+ int msggot;
+ TRY {
+ xbt_socket_meas_recv(measIn, 120, request->msg_size,
+ request->msg_amount);
+ xbt_socket_meas_send(measOut, 120, 1, 1);
+ }
+ CATCH_ANONYMOUS {
+ gras_socket_close(measMasterIn);
+ gras_socket_close(measIn);
+ gras_socket_close(measOut);
+ /* FIXME: tell error to remote ? */
+ RETHROWF("Error encountered while receiving the experiment: %s");
+ }
+ gras_msg_wait_or(60, msgtwaited, &ctx_reask, &msggot, &payloadgot);
+ switch (msggot) {
+ case 0: /* BW stop */
+ tooshort = 0;
+ break;
+ case 1: /* BW reask */
+ tooshort = 1;
+ free(request);
+ request = (bw_request_t) payloadgot;
+ XBT_VERB("Return the reasking RPC");
+ gras_msg_rpcreturn(60, ctx_reask, NULL);
+ }
+ gras_msg_cb_ctx_free(ctx_reask);
+ }
+
+ if (measIn != measMasterIn)
+ gras_socket_close(measMasterIn);
+ gras_socket_close(measIn);
+ gras_socket_close(measOut);
+ free(answer);
+ free(request);
+ XBT_VERB("BW experiment done.");
+ return 0;
}
-int grasbw_cbSatStart(gras_msg_t *msg) {
- gras_rawsock_t *raw;
- gras_sock_t *sock;
- xbt_error_t errcode;
- double start; /* time to timeout */
-
- /* specification of the test to run */
- char* to_name=gras_msg_ctn(msg,0,0,msgHost_t).host;
- unsigned int to_port=gras_msg_ctn(msg,0,0,msgHost_t).port;
-
- unsigned int msgSize=gras_msg_ctn(msg,1,0,SatExp_t).msgSize;
- unsigned int timeout=gras_msg_ctn(msg,1,0,SatExp_t).timeout;
- unsigned int raw_port;
-
+/**
+ * \brief request a bandwidth measurement between two remote peers
+ *
+ * \arg from_name: Name of the first peer
+ * \arg from_port: port on which the first process is listening for messages
+ * \arg to_name: Name of the second peer
+ * \arg to_port: port on which the second process is listening (for messages, do not
+ * give a measurement socket here. The needed measurement sockets will be created
+ * automatically and negociated between the peers)
+ * \arg buf_size: Size of the socket buffer. If 0, a sain default is used (32k, but may change)
+ * \arg msg_size: Size of each message sent.
+ * \arg msg_amount: Amount of such data to exchange
+ * \arg sec: where the result (in seconds) should be stored.
+ * \arg bw: observed Bandwidth (in byte/s)
+ *
+ * Conduct a bandwidth test from the process from_peer:from_port to to_peer:to_port.
+ * This call is blocking until the end of the experiment.
+ *
+ * @warning: in SimGrid version 3.1 and previous, the experiment size were specified
+ * as the total amount of data to send and the msg_size. This
+ * was changed for the fool wanting to send more than MAXINT
+ * bytes in a fat pipe.
+ *
+ * Results are reported in last args, and sizes are in bytes.
+ */
+void amok_bw_request(const char *from_name, unsigned int from_port,
+ const char *to_name, unsigned int to_port,
+ unsigned long int buf_size,
+ unsigned long int msg_size,
+ unsigned long int msg_amount,
+ double min_duration, /*OUT*/ double *sec, double *bw)
+{
+
+ xbt_socket_t sock;
/* The request */
- SatExp_t *request;
- /* answer */
- gras_msg_t *answer;
-
- /*
- fprintf(stderr,"grasbw_cbSatStart(sd=%p)\n",msg->sock);
- fprintf(stderr,"(server=%d,raw=%d,fromPID=%d,toPID=%d,toHost=%p,toPort=%d,toChan=%d)\n",
- msg->sock->server_sock,msg->sock->raw_sock,msg->sock->from_PID,
- msg->sock->to_PID,msg->sock->to_host,msg->sock->to_port,msg->sock->to_chan);
- */
-
- /* Negociate the saturation with the peer */
- if((errcode=gras_sock_client_open(to_name,to_port,&sock))) {
- fprintf(stderr,"cbSatStart(): Error %s encountered while contacting peer\n",
- xbt_error_name(errcode));
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Cannot contact peer.\n");
- return 1;
- }
- if (!(request=(SatExp_t *)malloc(sizeof(SatExp_t)))) {
- fprintf(stderr,"cbSatStart(): Malloc error\n");
- gras_sock_close(sock);
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- malloc_error,"Cannot build request.\n");
- return 1;
- }
+ bw_request_t request;
+ bw_res_t result;
+ request = xbt_new0(s_bw_request_t, 1);
+ request->buf_size = buf_size;
+ request->msg_size = msg_size;
+ request->msg_amount = msg_amount;
+ request->min_duration = min_duration;
- request->timeout=gras_msg_ctn(msg,1,0,SatExp_t).timeout;
- request->msgSize=gras_msg_ctn(msg,1,0,SatExp_t).msgSize;
-
- if ((errcode=gras_msg_new_and_send(sock,GRASMSG_SAT_BEGIN, 1,
- request,1))) {
- fprintf(stderr,"cbSatStart(): Error %s encountered while sending the request.\n",
- xbt_error_name(errcode));
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Cannot send request.\n");
- gras_sock_close(sock);
- return 1;
- }
- if ((errcode=gras_msg_wait(120,GRASMSG_SAT_BEGUN,&answer))) {
- fprintf(stderr,"cbSatStart(): Error %s encountered while waiting for the ACK.\n",
- xbt_error_name(errcode));
- gras_sock_close(sock);
+ request->peer.name = (char *) to_name;
+ request->peer.port = to_port;
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,
- "Cannot receive the ACK.\n");
- return 1;
- }
- if((errcode=gras_msg_ctn(answer,0,0,msgError_t).errcode)) {
- fprintf(stderr,"cbSatStart(): Peer reported error %s (%s).\n",
- xbt_error_name(errcode),gras_msg_ctn(answer,0,0,msgError_t).errmsg);
-
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,
- "Peer repported '%s'.\n",gras_msg_ctn(answer,0,0,msgError_t).errmsg);
- gras_msg_free(answer);
- gras_sock_close(sock);
- return 1;
- }
+ sock = gras_socket_client(from_name, from_port);
- raw_port=gras_msg_ctn(answer,1,0,SatExp_t).port;
- if ((errcode=gras_rawsock_client_open(to_name,raw_port,msgSize,&raw))) {
- fprintf(stderr,"cbSatStart(): Error %s while opening raw socket to %s:%d.\n",
- xbt_error_name(errcode),to_name,gras_msg_ctn(answer,1,0,SatExp_t).port);
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Cannot open raw socket.\n");
- gras_sock_close(sock);
- return 1;
- }
+ XBT_DEBUG("Ask for a BW test between %s:%u and %s:%u", from_name, from_port,
+ to_name, to_port);
+ gras_msg_rpccall(sock, 20 * 60, "BW request", &request, &result);
- /* send a train of data before repporting that XP is started */
- if ((errcode=gras_rawsock_send(raw,msgSize,msgSize))) {
- fprintf(stderr,"cbSatStart: Failure %s during raw send\n",xbt_error_name(errcode));
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Cannot raw send.\n");
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- return 1;
- }
-
- grasRepportError(msg->sock,GRASMSG_SAT_STARTED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- no_error,"Saturation started");
- gras_msg_free(answer);
- gras_msg_free(msg);
-
- /* Do the saturation until we get a SAT_STOP message or until we timeout the whole XP*/
- start=gras_time();
- while (gras_msg_wait(0,GRASMSG_SAT_STOP,&msg)==timeout_error &&
- gras_time()-start < timeout) {
- if ((errcode=gras_rawsock_send(raw,msgSize,msgSize))) {
- fprintf(stderr,"cbSatStart: Failure %s during raw send\n",xbt_error_name(errcode));
- /* our error message do not interess anyone. SAT_STOP will do nothing. */
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- return 1;
- }
- }
- if (gras_time()-start > timeout) {
- fprintf(stderr,"The saturation experiment did timeout. Stop it NOW\n");
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- return 1;
- }
-
- /* Handle the SAT_STOP which broke the previous while */
-
- if ((errcode=gras_msg_new_and_send(sock, GRASMSG_SAT_END,0))) {
- fprintf(stderr,"cbSatStart(): Cannot tell peer to stop saturation\n");
-
- grasRepportError(msg->sock,GRASMSG_SAT_STOPPED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Sending SAT_END to peer failed.\n");
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- return 1;
- }
-
- if ((errcode=gras_msg_wait(60,GRASMSG_SAT_ENDED,&answer))) {
- fprintf(stderr,"cbSatStart(): Peer didn't ACK the end\n");
-
- grasRepportError(msg->sock,GRASMSG_SAT_STOPPED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- errcode,"Receiving SAT_ENDED from peer failed.\n");
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- return 1;
- }
- grasRepportError(msg->sock,GRASMSG_SAT_STOPPED,1,
- "cbSatStart: Severe error: Cannot send error status to requester!!\n",
- no_error,"");
+ if (sec)
+ *sec = result->sec;
+ if (bw)
+ *bw = result->bw;
- gras_sock_close(sock);
- gras_rawsock_close(raw);
- gras_msg_free(answer);
- gras_msg_free(msg);
+ XBT_VERB("BW test (%s:%u -> %s:%u) took %f sec (%f kb/s)",
+ from_name, from_port, to_name, to_port,
+ result->sec, ((double) result->bw) / 1024.0);
- return 1;
+ gras_socket_close(sock);
+ free(result);
+ free(request);
}
-int grasbw_cbSatBegin(gras_msg_t *msg) {
- gras_rawsock_t *raw;
- xbt_error_t errcode;
- double start; /* timer */
- /* request */
- unsigned int msgSize=gras_msg_ctn(msg,0,0,SatExp_t).msgSize;
- unsigned int timeout=gras_msg_ctn(msg,0,0,SatExp_t).timeout;
- /* answer */
- SatExp_t *request;
- msgError_t *error;
-
- if (!(request=(SatExp_t*)malloc(sizeof(SatExp_t))) ||
- !(error=(msgError_t *)malloc(sizeof(msgError_t)))) {
- fprintf(stderr,"cbSatBegin(): Malloc error\n");
- grasRepportError(msg->sock,GRASMSG_SAT_BEGUN,2,
- "cbSatBegin: Severe error: Cannot send error status to requester!!\n",
- malloc_error,"Malloc error");
- return 1;
- }
-
- if ((errcode=gras_rawsock_server_open(6666,8000,msgSize,&raw))) {
- fprintf(stderr,"cbSatBegin(): Error %s encountered while opening a raw socket\n",
- xbt_error_name(errcode));
- grasRepportError(msg->sock,GRASMSG_SAT_BEGUN,2,
- "cbSatBegin: Severe error: Cannot send error status to requester!!\n",
- errcode,"Cannot open raw socket");
- return 1;
- }
- request->port=gras_rawsock_get_peer_port(raw);
- request->msgSize=msgSize;
- error->errcode=no_error;
- error->errmsg[0]='\0';
- if ((errcode=gras_msg_new_and_send(msg->sock,GRASMSG_SAT_BEGUN,2,
- error,1,
- request,1))) {
- fprintf(stderr,"cbSatBegin(): Error %s encountered while send ACK to peer\n",
- xbt_error_name(errcode));
- return 1;
- }
- gras_msg_free(msg);
-
- start=gras_time();
- while (gras_msg_wait(0,GRASMSG_SAT_END,&msg)==timeout_error &&
- gras_time() - start < timeout) {
- errcode=gras_rawsock_recv(raw,msgSize,msgSize,1);
- if (errcode != timeout_error && errcode != no_error) {
- fprintf(stderr,"cbSatBegin: Failure %s during raw receive\n",xbt_error_name(errcode));
- /* our error message do not interess anyone. SAT_END will do nothing. */
- /* (if timeout'ed, it may be because the sender stopped emission. so survive it) */
- return 1;
- }
- }
- if (gras_time()-start > timeout) {
- fprintf(stderr,"The saturation experiment did timeout. Stop it NOW.\n");
- gras_rawsock_close(raw);
- return 1;
- }
-
- grasRepportError(msg->sock,GRASMSG_SAT_ENDED,1,
- "cbSatBegin: Cannot send SAT_ENDED.\n",
- no_error,"");
- gras_rawsock_close(raw);
- gras_msg_free(msg);
- return 1;
+int amok_bw_cb_bw_request(gras_msg_cb_ctx_t ctx, void *payload)
+{
+
+ /* specification of the test to run, and our answer */
+ bw_request_t request = *(bw_request_t *) payload;
+ bw_res_t result = xbt_new0(s_bw_res_t, 1);
+ xbt_socket_t peer, asker;
+
+ asker = gras_msg_cb_ctx_from(ctx);
+ XBT_VERB("Asked by %s:%d to conduct a bw XP with %s:%d (request: %lu %lu)",
+ xbt_socket_peer_name(asker), xbt_socket_peer_port(asker),
+ request->peer.name, request->peer.port,
+ request->msg_size, request->msg_amount);
+ peer = gras_socket_client(request->peer.name, request->peer.port);
+ amok_bw_test(peer,
+ request->buf_size, request->msg_size, request->msg_amount,
+ request->min_duration, &(result->sec), &(result->bw));
+
+ gras_msg_rpcreturn(240, ctx, &result);
+
+ gras_os_sleep(1);
+ gras_socket_close(peer); /* FIXME: it should be blocking in RL until everything is sent */
+ free(request->peer.name);
+ free(request);
+ free(result);
+
+ return 0;
}
-xbt_error_t grasbw_saturate_stop(const char* from_name,unsigned int from_port,
- const char* to_name,unsigned int to_port) {
- xbt_error_t errcode;
- gras_sock_t *sock;
- gras_msg_t *answer;
-
- if((errcode=gras_sock_client_open(from_name,from_port,&sock))) {
- fprintf(stderr,"saturate_stop(): Error %s encountered while contacting peer\n",
- xbt_error_name(errcode));
- return errcode;
- }
-
- if ((errcode=gras_msg_new_and_send(sock,GRASMSG_SAT_STOP,0))) {
- fprintf(stderr,"saturate_stop(): Error %s encountered while sending request\n",
- xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
- }
-
- if ((errcode=gras_msg_wait(120,GRASMSG_SAT_STOPPED,&answer))) {
- fprintf(stderr,"saturate_stop(): Error %s encountered while receiving ACK\n",
- xbt_error_name(errcode));
- gras_sock_close(sock);
- return errcode;
- }
-
- if((errcode=gras_msg_ctn(answer,0,0,msgError_t).errcode)) {
- fprintf(stderr,"saturate_stop(): Peer reported error %s (%s).\n",
- xbt_error_name(errcode),gras_msg_ctn(answer,0,0,msgError_t).errmsg);
- gras_msg_free(answer);
- gras_sock_close(sock);
- return errcode;
- }
-
- gras_msg_free(answer);
- gras_sock_close(sock);
-
- return no_error;
+/** \brief builds a matrix of results of bandwidth measurement
+ *
+ * @warning: in SimGrid version 3.1 and previous, the experiment size were specified
+ * as the total amount of data to send and the msg_size. This
+ * was changed for the fool wanting to send more than MAXINT
+ * bytes in a fat pipe.
+ */
+double *amok_bw_matrix(xbt_dynar_t peers,
+ int buf_size_bw, int msg_size_bw, int msg_amount_bw,
+ double min_duration)
+{
+ double sec;
+ /* construction of matrices for bandwith and latency */
+
+
+ unsigned int i, j;
+ int len = xbt_dynar_length(peers);
+
+ double *matrix_res = xbt_new0(double, len * len);
+ xbt_peer_t p1, p2;
+
+ xbt_dynar_foreach(peers, i, p1) {
+ xbt_dynar_foreach(peers, j, p2) {
+ if (i != j) {
+ /* Mesurements of Bandwidth */
+ amok_bw_request(p1->name, p1->port, p2->name, p2->port,
+ buf_size_bw, msg_size_bw, msg_amount_bw,
+ min_duration, &sec, &matrix_res[i * len + j]);
+ }
+ }
+ }
+ return matrix_res;
}
-#endif
