// show a notice if we can't get work from this project,
// and there's something the user could do about it.
//
void PROJECT::show_no_work_notice() {
bool some_banned = false;
for (int i=0; i<coprocs.n_rsc; i++) {
if (no_rsc_apps[i]) continue;
bool banned_by_user = no_rsc_pref[i] || no_rsc_config[i];
if (!gstate.acct_mgr_info.dynamic) {
// dynamic account managers manage rsc usage themselves, not user
//
banned_by_user = banned_by_user || no_rsc_ams[i];
// note to self: ||= doesn't exist
}
if (!banned_by_user) {
continue;
}
string x;
x = NO_WORK_MSG;
x += " ";
x += rsc_name_long(i);
x += ". ";
x += _("To fix this, you can ");
bool first = true;
if (no_rsc_pref[i]) {
x += _("change Project Preferences on the project's web site");
first = false;
}
if (no_rsc_config[i]) {
if (!first) x += ", or ";
x += _("remove GPU exclusions in your cc_config.xml file");
first = false;
}
if (no_rsc_ams[i] && !gstate.acct_mgr_info.dynamic) {
if (!first) x += ", or ";
x += _("change your settings at your account manager web site");
}
x += ".";
msg_printf(this, MSG_USER_ALERT, "%s", x.c_str());
some_banned = true;
}
if (!some_banned) {
notices.remove_notices(this, REMOVE_NO_WORK_MSG);
return;
}
}
int RESULT::write_gui(MIOFILE& out) {
out.printf(
"<result>\n"
" <name>%s</name>\n"
" <wu_name>%s</wu_name>\n"
" <version_num>%d</version_num>\n"
" <plan_class>%s</plan_class>\n"
" <project_url>%s</project_url>\n"
" <final_cpu_time>%f</final_cpu_time>\n"
" <final_elapsed_time>%f</final_elapsed_time>\n"
" <exit_status>%d</exit_status>\n"
" <state>%d</state>\n"
" <report_deadline>%f</report_deadline>\n"
" <received_time>%f</received_time>\n"
" <estimated_cpu_time_remaining>%f</estimated_cpu_time_remaining>\n",
name,
wu_name,
version_num,
plan_class,
project->master_url,
final_cpu_time,
final_elapsed_time,
exit_status,
state(),
report_deadline,
received_time,
estimated_runtime_remaining()
);
if (got_server_ack) out.printf(" <got_server_ack/>\n");
if (ready_to_report) out.printf(" <ready_to_report/>\n");
if (completed_time) out.printf(" <completed_time>%f</completed_time>\n", completed_time);
if (suspended_via_gui) out.printf(" <suspended_via_gui/>\n");
if (project->suspended_via_gui) out.printf(" <project_suspended_via_gui/>\n");
if (report_immediately) out.printf(" <report_immediately/>\n");
if (edf_scheduled) out.printf(" <edf_scheduled/>\n");
if (coproc_missing) out.printf(" <coproc_missing/>\n");
if (schedule_backoff > gstate.now) {
out.printf(" <scheduler_wait/>\n");
if (strlen(schedule_backoff_reason)) {
out.printf(
" <scheduler_wait_reason>%s</scheduler_wait_reason>\n",
schedule_backoff_reason
);
}
}
if (avp->needs_network && gstate.network_suspended) out.printf(" <network_wait/>\n");
ACTIVE_TASK* atp = gstate.active_tasks.lookup_result(this);
if (atp) {
atp->write_gui(out);
}
if (!strlen(resources)) {
// only need to compute this string once
//
if (avp->gpu_usage.rsc_type) {
if (avp->gpu_usage.usage == 1) {
sprintf(resources,
"%.3g CPUs + 1 %s",
avp->avg_ncpus,
rsc_name_long(avp->gpu_usage.rsc_type)
);
} else {
sprintf(resources,
"%.3g CPUs + %.3g %ss",
avp->avg_ncpus,
avp->gpu_usage.usage,
rsc_name_long(avp->gpu_usage.rsc_type)
);
}
} else if (avp->missing_coproc) {
sprintf(resources, "%.3g CPUs + %s GPU (missing)",
avp->avg_ncpus, avp->missing_coproc_name
);
} else if (!project->non_cpu_intensive && (avp->avg_ncpus != 1)) {
sprintf(resources, "%.3g CPUs", avp->avg_ncpus);
} else {
strcpy(resources, " ");
}
}
if (strlen(resources)>1) {
char buf[256];
strcpy(buf, "");
if (atp && atp->scheduler_state == CPU_SCHED_SCHEDULED) {
if (avp->gpu_usage.rsc_type) {
COPROC& cp = coprocs.coprocs[avp->gpu_usage.rsc_type];
if (cp.count > 1) {
// if there are multiple GPUs of this type,
// show the user which one(s) are being used
//
int n = (int)ceil(avp->gpu_usage.usage);
strcpy(buf, " (device ");
for (int i=0; i<n; i++) {
char buf2[256];
sprintf(buf2, "%d", cp.device_nums[coproc_indices[i]]);
if (i > 0) {
strcat(buf, "/");
}
strcat(buf, buf2);
}
strcat(buf, ")");
}
}
}
out.printf(
" <resources>%s%s</resources>\n", resources, buf
);
}
out.printf("</result>\n");
return 0;
}
void simulate() {
bool action;
double start = START_TIME;
gstate.now = start;
html_start();
fprintf(summary_file,
"Hardware summary\n %d CPUs, %.1f GFLOPS\n",
gstate.host_info.p_ncpus, gstate.host_info.p_fpops/1e9
);
for (int i=1; i<coprocs.n_rsc; i++) {
fprintf(summary_file,
" %d %s GPUs, %.1f GFLOPS\n",
coprocs.coprocs[i].count,
coprocs.coprocs[i].type,
coprocs.coprocs[i].peak_flops/1e9
);
}
fprintf(summary_file,
"Preferences summary\n"
" work buf min %f max %f\n"
" Scheduling period %f\n"
"Scheduling policies\n"
" Round-robin only: %s\n"
" Scheduler EDF simulation: %s\n"
" REC half-life: %f\n",
gstate.work_buf_min(), gstate.work_buf_total(),
gstate.global_prefs.cpu_scheduling_period(),
cpu_sched_rr_only?"yes":"no",
server_uses_workload?"yes":"no",
cc_config.rec_half_life
);
fprintf(summary_file, "Jobs\n");
for (unsigned int i=0; i<gstate.results.size(); i++) {
RESULT* rp = gstate.results[i];
fprintf(summary_file,
" %s %s (%s)\n time left %s deadline %s\n",
rp->project->project_name,
rp->name,
rsc_name_long(rp->avp->gpu_usage.rsc_type),
timediff_format(rp->sim_flops_left/rp->avp->flops).c_str(),
timediff_format(rp->report_deadline - START_TIME).c_str()
);
}
fprintf(summary_file,
"Simulation parameters\n"
" time step %f, duration %f\n"
"-------------------\n",
delta, duration
);
write_inputs();
while (1) {
on = on_proc.sample(delta);
if (on) {
active = active_proc.sample(delta);
if (active) {
gpu_active = gpu_active_proc.sample(delta);
} else {
gpu_active = false;
}
connected = connected_proc.sample(delta);
} else {
active = gpu_active = connected = false;
}
// do accounting for the period that just ended,
// even if we're now in an "off" state.
//
// need both of the following, else crash
//
action |= gstate.active_tasks.poll();
action |= gstate.handle_finished_apps();
if (on) {
while (1) {
action = false;
action |= gstate.schedule_cpus();
if (connected) {
action |= gstate.scheduler_rpc_poll();
// this deletes completed results
}
action |= gstate.active_tasks.poll();
action |= gstate.handle_finished_apps();
gpu_suspend_reason = gpu_active?0:1;
//msg_printf(0, MSG_INFO, action?"did action":"did no action");
if (!action) break;
}
}
//msg_printf(0, MSG_INFO, "took time step");
for (unsigned int i=0; i<gstate.active_tasks.active_tasks.size(); i++) {
ACTIVE_TASK* atp = gstate.active_tasks.active_tasks[i];
if (atp->task_state() == PROCESS_EXECUTING) {
atp->elapsed_time += delta;
}
}
html_rec();
write_recs();
gstate.now += delta;
if (gstate.now > start + duration) break;
}
html_end();
}
