#include "xbt/module.h"
#include "mc/mc.h"
#include "surf/surf_resource.h"
-//#include "surf/surf_resource_lmm.h"
-
+#include "xbt/xbt_os_thread.h"
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(surf_kernel, surf,
"Logging specific to SURF (kernel)");
-/* Additional declarations for Windows potability. */
+/* Additional declarations for Windows portability. */
#ifndef MAX_DRIVE
#define MAX_DRIVE 26
/* Don't forget to update the option description in smx_config when you change this */
s_surf_model_description_t surf_network_model_description[] = {
+ {"LV08",
+ "Realistic network analytic model (slow-start modeled by multiplying latency by 10.4, bandwidth by .92; bottleneck sharing uses a payload of S=8775 for evaluating RTT). ",
+ surf_network_model_init_LegrandVelho},
{"Constant",
- "Simplistic network model where all communication take a constant time (one second)",
+ "Simplistic network model where all communication take a constant time (one second). This model provides the lowest realism, but is (marginally) faster.",
surf_network_model_init_Constant},
- {"CM02",
- "Realistic network model with lmm_solve and no correction factors",
- surf_network_model_init_CM02},
- {"LV08",
- "Realistic network model with lmm_solve, adequate correction factors (latency*=10.4, bandwidth*=.92, S=8775) and partial invalidation optimization",
- im_surf_network_model_init_LegrandVelho},
- {"LV08_fullupdate",
- "Realistic network model wit lmm_solve, adequate correction factors (latency*=10.4, bandwidth*=.92, S=8775) but no further optimization. Should produce the same results as LV08, only slower.",
- surf_network_model_init_LegrandVelho},
{"SMPI",
- "Realistic network model with lmm_solve and correction factors on three intervals (< 1KiB, < 64 KiB, >= 64 KiB)",
+ "Realistic network model specifically tailored for HPC settings (accurate modeling of slow start with correction factors on three intervals: < 1KiB, < 64 KiB, >= 64 KiB)",
surf_network_model_init_SMPI},
+ {"CM02",
+ "Legacy network analytic model (Very similar to LV08, but without corrective factors. The timings of small messages are thus poorly modeled).",
+ surf_network_model_init_CM02},
#ifdef HAVE_GTNETS
{"GTNets",
- "Network Pseudo-model using the GTNets simulator instead of an analytic model",
+ "Network pseudo-model using the GTNets simulator instead of an analytic model",
surf_network_model_init_GTNETS},
#endif
#ifdef HAVE_NS3
{"NS3",
- "Use NS3 tcp model",
+ "Network pseudo-model using the NS3 tcp model instead of an analytic model",
surf_network_model_init_NS3},
#endif
{"Reno",
- "Model using lagrange_solve instead of lmm_solve (experts only)",
+ "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
surf_network_model_init_Reno},
{"Reno2",
- "Model using lagrange_solve instead of lmm_solve (experts only)",
+ "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
surf_network_model_init_Reno2},
{"Vegas",
- "Model using lagrange_solve instead of lmm_solve (experts only)",
+ "Model from Steven H. Low using lagrange_solve instead of lmm_solve (experts only; check the code for more info).",
surf_network_model_init_Vegas},
{NULL, NULL, NULL} /* this array must be NULL terminated */
};
s_surf_model_description_t surf_cpu_model_description[] = {
- {"Cas01_fullupdate", "CPU classical model time=size/power",
- surf_cpu_model_init_Cas01},
{"Cas01",
- "Variation of Cas01_fullupdate with partial invalidation optimization of lmm system. Should produce the same values, only faster",
- surf_cpu_model_init_Cas01_im},
- {"CpuTI",
- "Variation of Cas01 with also trace integration. Should produce the same values, only faster if you use availability traces",
- surf_cpu_model_init_ti},
+ "Simplistic CPU model (time=size/power).",
+ surf_cpu_model_init_Cas01},
{NULL, NULL, NULL} /* this array must be NULL terminated */
};
s_surf_model_description_t surf_workstation_model_description[] = {
- {"CLM03",
- "Default workstation model, using LV08 and CM02 as network and CPU",
- surf_workstation_model_init_CLM03},
+ {"default",
+ "Default workstation model. Currently, CPU:Cas01 and network:LV08 (with cross traffic enabled)",
+ surf_workstation_model_init_current_default},
{"compound",
- "Workstation model allowing you to use other network and CPU models",
+ "Workstation model that is automatically chosen if you change the network and CPU models",
surf_workstation_model_init_compound},
- {"ptask_L07", "Workstation model with better parallel task modeling",
+ {"ptask_L07", "Workstation model somehow similar to Cas01+CM02 but allowing parallel tasks",
surf_workstation_model_init_ptask_L07},
{NULL, NULL, NULL} /* this array must be NULL terminated */
};
+s_surf_model_description_t surf_optimization_mode_description[] = {
+ {"Lazy",
+ "Lazy action management (partial invalidation in lmm + heap in action remaining).",
+ NULL},
+ {"TI",
+ "Trace integration. Highly optimized mode when using availability traces (only available for the Cas01 CPU model for now).",
+ NULL},
+ {"Full",
+ "Full update of remaining and variables. Slow but may be useful when debugging.",
+ NULL},
+ {NULL, NULL, NULL} /* this array must be NULL terminated */
+};
+
+#ifdef CONTEXT_THREADS
+static xbt_parmap_t surf_parmap = NULL; /* parallel map on models */
+#endif
+
+static int surf_nthreads = 1; /* number of threads of the parmap (1 means no parallelism) */
+static double *surf_mins = NULL; /* return value of share_resources for each model */
+static int surf_min_index; /* current index in surf_mins */
+static double min; /* duration determined by surf_solve */
+
+static void surf_share_resources(surf_model_t model);
+static void surf_update_actions_state(surf_model_t model);
+
/** Displays the long description of all registered models, and quit */
void model_help(const char *category, s_surf_model_description_t * table)
{
for (i = 1; table[i].name; i++) {
name_list =
xbt_realloc(name_list,
- strlen(name_list) + strlen(table[i].name) + 2);
+ strlen(name_list) + strlen(table[i].name) + 3);
strcat(name_list, ", ");
strcat(name_list, table[i].name);
}
xbt_dynar_foreach(model_list, iter, model)
model->model_private->finalize();
xbt_dynar_free(&model_list);
+ routing_exit();
if (maxmin_system) {
lmm_system_free(maxmin_system);
}
surf_action_exit();
+#ifdef CONTEXT_THREADS
+ xbt_parmap_destroy(surf_parmap);
+ xbt_free(surf_mins);
+ surf_mins = NULL;
+#endif
+
xbt_dynar_free(&surf_path);
xbt_lib_free(&host_lib);
double surf_solve(double max_date)
{
- double min = -1.0; /* duration */
+ min = -1.0; /* duration */
double next_event_date = -1.0;
double model_next_action_end = -1.0;
double value = -1.0;
}
XBT_DEBUG("Looking for next action end for all models except NS3");
- xbt_dynar_foreach(model_list, iter, model) {
- if(strcmp(model->name,"network NS3") ){
- XBT_DEBUG("Running for Resource [%s]", model->name);
- model_next_action_end = model->model_private->share_resources(NOW);
- XBT_DEBUG("Resource [%s] : next action end = %f",
- model->name, model_next_action_end);
- if (((min < 0.0) || (model_next_action_end < min))
- && (model_next_action_end >= 0.0))
- min = model_next_action_end;
+
+ if (surf_mins == NULL) {
+ surf_mins = xbt_new(double, xbt_dynar_length(model_list));
+ }
+ surf_min_index = 0;
+
+
+ if (surf_get_nthreads() > 1) {
+ /* parallel version */
+#ifdef CONTEXT_THREADS
+ xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_share_resources, model_list);
+#endif
+ }
+ else {
+ /* sequential version */
+ xbt_dynar_foreach(model_list, iter, model) {
+ surf_share_resources(model);
+ }
+ }
+
+ unsigned i;
+ for (i = 0; i < xbt_dynar_length(model_list); i++) {
+ if ((min < 0.0 || surf_mins[i] < min)
+ && surf_mins[i] >= 0.0) {
+ min = surf_mins[i];
}
}
- XBT_DEBUG("Min for resources (except NS3) : %f", min);
+ XBT_DEBUG("Min for resources (remember that NS3 dont update that value) : %f", min);
XBT_DEBUG("Looking for next trace event");
min = model_next_action_end;
}
- if (next_event_date == -1.0) break;
+ if (next_event_date == -1.0) {
+ XBT_DEBUG("no next TRACE event. Stop searching for it");
+ break;
+ }
if ((min != -1.0) && (next_event_date > NOW + min)) break;
}
} while (1);
- /* FIXME: Moved this test to here to avoid stoping simulation if there are actions running on cpus and all cpus are with availability = 0.
+ /* FIXME: Moved this test to here to avoid stopping simulation if there are actions running on cpus and all cpus are with availability = 0.
* This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a trace with periodicity > 0.
* The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed */
- if (min < 0.0)
+ if (min == -1.0) {
+ XBT_DEBUG("No next event at all. Bail out now.");
return -1.0;
+ }
XBT_DEBUG("Duration set to %f", min);
NOW = NOW + min;
- xbt_dynar_foreach(model_list, iter, model)
- model->model_private->update_actions_state(NOW, min);
+ if (surf_get_nthreads() > 1) {
+ /* parallel version */
+#ifdef CONTEXT_THREADS
+ xbt_parmap_apply(surf_parmap, (void_f_pvoid_t) surf_update_actions_state, model_list);
+#endif
+ }
+ else {
+ /* sequential version */
+ xbt_dynar_foreach(model_list, iter, model) {
+ surf_update_actions_state(model);
+ }
+ }
#ifdef HAVE_TRACING
TRACE_paje_dump_buffer (0);
{
return NOW;
}
+
+static void surf_share_resources(surf_model_t model)
+{
+ if (strcmp(model->name,"network NS3")) {
+ XBT_DEBUG("Running for Resource [%s]", model->name);
+ double next_action_end = model->model_private->share_resources(NOW);
+ XBT_DEBUG("Resource [%s] : next action end = %f",
+ model->name, next_action_end);
+ int i = __sync_fetch_and_add(&surf_min_index, 1);
+ surf_mins[i] = next_action_end;
+ }
+}
+
+static void surf_update_actions_state(surf_model_t model)
+{
+ model->model_private->update_actions_state(NOW, min);
+}
+
+/**
+ * \brief Returns the number of parallel threads used to update the models.
+ * \return the number of threads (1 means no parallelism)
+ */
+int surf_get_nthreads(void) {
+ return surf_nthreads;
+}
+
+/**
+ * \brief Sets the number of parallel threads used to update the models.
+ *
+ * A value of 1 means no parallelism.
+ *
+ * \param nb_threads the number of threads to use
+ */
+void surf_set_nthreads(int nthreads) {
+
+ if (nthreads<=0) {
+ nthreads = xbt_os_get_numcores();
+ XBT_INFO("Auto-setting surf/nthreads to %d",nthreads);
+ }
+
+#ifdef CONTEXT_THREADS
+ xbt_parmap_destroy(surf_parmap);
+ surf_parmap = NULL;
+#endif
+
+ if (nthreads > 1) {
+#ifdef CONTEXT_THREADS
+ surf_parmap = xbt_parmap_new(nthreads, XBT_PARMAP_DEFAULT);
+#else
+ THROWF(arg_error, 0, "Cannot activate parallel threads in Surf: your architecture does not support threads");
+#endif
+ }
+
+ surf_nthreads = nthreads;
+}
