int main(int argc, char *argv[])
{
SetErrorMode(SEM_NOGPFAULTERRORBOX);
avmshell::WinPlatform platformInstance;
gPlatformHandle = &platformInstance;
int code = executeShell(argc, argv);
return code;
}
//monitor the submitted tasks
// by polling each server for its load information. If a server is idle then it will return load = -4 (queue length - num of idle cores)
// so if every server returns load = -4 it implies that all submitted tasks are complete
void *monitor_function(void* args) {
ZHTClient *clientRet = (ZHTClient*)args;
Package loadPackage, shutdownPackage;
string loadmessage("Monitoring Information!");
loadPackage.set_virtualpath(loadmessage);
loadPackage.set_operation(15);
string loadstr = loadPackage.SerializeAsString();
string endmessage("Shutdown!");
loadPackage.set_virtualpath(endmessage);
loadPackage.set_operation(98);
string endstr = loadPackage.SerializeAsString();
//int num_worker = clientRet.memberList.size();
int num_worker = clientRet->memberList.size();
int num_cores = ncores;
int index = 0;
long termination_value = num_worker * num_cores * -1;
int total_avail_cores = num_cores * num_worker;
int32_t total_queued = 0;
int32_t total_idle = 0;
int32_t queued_busy = 0;
int32_t queued_idle = 0;
int32_t queued = 0;
int32_t num_idle = 0;
int32_t num_busy = 0;
int32_t load = 0;
int32_t total_busy = 0;
//int32_t status = 0;
int32_t finished = 0;
int32_t total_msg_count = 0;
int32_t ret = 0;
//sleep(60);
int min_lines = num_worker;
int num = num_worker - 1;
stringstream num_ss;
num_ss << num;
long num_monitor = 0;
//min_lines++;
// not sure why we need this
string filename(shared);
filename = filename + "startinfo" + num_ss.str();
string cmd("cat ");
cmd = cmd + filename + " | wc -l";
string result = executeShell(cmd);
cout << cmd << " " << result << endl;
//cout << "client: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
/*string filename(shared);
filename = filename + "start_info";
string cmd("wc -l ");
cmd = cmd + filename + " | awk {\'print $1\'}";
string result = executeShell(cmd);*/
while(atoi(result.c_str()) < 1) {
usleep(minterval);
result = executeShell(cmd); cout << " temp result = " << result << endl;
}
cout << "client: minlines = 1 " << " cmd = " << cmd << " result = " << result << endl;
//cout << "starting to monitor" << endl;
cout << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << endl;
timespec local_start, local_diff;
clock_gettime(CLOCK_REALTIME, &local_start);
local_diff = timediff(start_tasks, local_start);
if (client_logfile.is_open() && cl_LOGGING) {
client_logfile << "Submission time: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << endl;
client_logfile << "Monitoring time: " << local_start.tv_sec << " SECONDS " << local_start.tv_nsec << " NANOSECONDS" << endl;
client_logfile << "TIME TAKEN: " << local_diff.tv_sec << " SECONDS " << local_diff.tv_nsec << " NANOSECONDS" << endl;
}
int total_fin = 0;
while(1) {
// total_queued = 0;
// total_idle = 0;
// queued_busy = 0;
total_fin = 0;
stringstream worker_load;
for(index = 0; index < num_worker; index++) {
//int32_t queued_idle = clientRet.svrtosvr(loadstr, loadstr.length(), index);
queued_idle = clientRet->svrtosvr(loadstr, loadstr.length(), index);
// queued = queued_idle/10; // summation of the lengths of the three queues
// num_idle = queued_idle%10; // number of idle cores
// total_queued = total_queued + queued;
// total_idle = total_idle + num_idle;
// num_busy = num_cores - num_idle;
// load = queued + num_busy;
total_fin += queued_idle;
// worker_load << load << " ";
worker_load << queued_idle << " ";
}
// loadfile << worker_load.str() << endl;
// total_busy = total_avail_cores - total_idle;
// queued_busy = total_queued + total_busy;
// finished = total_num_tasks - queued_busy;
num_monitor++;
clock_gettime(CLOCK_REALTIME, &end_tasks);
//cout << "Total busy cores " << total_busy << " Total Load on all workers = " << queued_busy << " No. of tasks finished = " << finished << " Total tasks submitted = " << total_num_tasks << endl;//" time = " << end_tasks.tv_sec << " " << end_tasks.tv_nsec << endl;
if (client_logfile.is_open() && cl_LOGGING) {
// client_logfile << "Total busy cores " << total_busy << " Total Load on all workers = " << queued_busy << " No. of tasks finished = " << finished << " Total tasks submitted = " << total_num_tasks << endl;
//client_logfile << "No. of tasks finished is:" << total_fin <<", No. of tasks submitted is:" << total_num_tasks << endl;
}
// if(finished == total_num_tasks)
if (total_fin == total_num_tasks)
{
clock_gettime(CLOCK_REALTIME, &end_tasks);
cout << "\n\n\n\n==============================All tasks finished===========================\n\n\n\n";
break;
}
usleep(minterval);
}
total_msg_count = 0;
for(index = 0; index < num_worker; index++) {
//clientRet.svrtosvr(endstr, endstr.length(), index);
ret = clientRet->svrtosvr(endstr, endstr.length(), index);
total_msg_count += ret;
}
cout << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << "\n";
cout << "TIME END: " << end_tasks.tv_sec << " SECONDS " << end_tasks.tv_nsec << " NANOSECONDS" << "\n";
timespec diff = timediff(start_tasks, end_tasks);
cout << "TIME TAKEN: " << diff.tv_sec << " SECONDS " << diff.tv_nsec << " NANOSECONDS" << "\n";
cout << "Total messages between all servers = " << total_msg_count << endl;
if (client_logfile.is_open() && cl_LOGGING) {
client_logfile << "\n\n\n\n==============================All tasks finished===========================\n\n\n\n";
client_logfile << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << "\n";
client_logfile << "TIME END: " << end_tasks.tv_sec << " SECONDS " << end_tasks.tv_nsec << " NANOSECONDS" << "\n";
client_logfile << "TIME TAKEN: " << diff.tv_sec << " SECONDS " << diff.tv_nsec << " NANOSECONDS" << endl;
client_logfile << "Total messages between all servers = " << total_msg_count << endl;
client_logfile << "Total monitoring times is = " << num_monitor << endl;
client_logfile.flush();
client_logfile.close();
if (loadfile.is_open())
{
loadfile.flush();
loadfile.close();
}
//return 1;
}
pthread_exit(NULL);
}
Worker::Worker(char *parameters[], NoVoHT *novoht) {
/* set thread detachstate attribute to DETACHED */
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM);
/* filename definitions */
set_dir(parameters[9], parameters[10]);
file_worker_start.append(shared);
file_worker_start.append("startinfo");
file_task_fp.append(prefix);
file_task_fp.append("pkgs");
file_migrate_fp.append(prefix);
file_migrate_fp.append("log_migrate");
file_fin_fp.append(prefix);
file_fin_fp.append("finish");
file_log_fp.append(prefix);
file_log_fp.append("log_worker");
pmap = novoht;
Env_var::set_env_var(parameters);
svrclient.initialize(Env_var::cfgFile, Env_var::membershipFile,
Env_var::TCP);
//svrzht.initialize(Env_var::cfgFile, Env_var::membershipFile, Env_var::TCP);
//svrmig.initialize(Env_var::cfgFile, Env_var::membershipFile, Env_var::TCP);
if (set_ip(ip)) {
printf("Could not get the IP address of this machine!\n");
exit(1);
}
for (int i = 0; i < 10; i++) {
msg_count[i] = 0;
}
poll_interval = start_poll;
poll_threshold = start_thresh;
num_nodes = svrclient.memberList.size();
num_cores = atoi(parameters[11]);
ws_sleep = atoi(parameters[12]);
num_idle_cores = num_cores;
neigh_mode = 'd';
//worker.num_neigh = (int)(sqrt(worker.num_nodes));
num_neigh = (int) (sqrt(num_nodes));
neigh_index = new int[num_neigh];
selfIndex = getSelfIndex(ip, atoi(parameters[1]), svrclient.memberList);// replace "localhost" with proper hostname, host is the IP in C++ string
ostringstream oss;
oss << selfIndex;
printf("<ip:selfIndex>: <%s:%d>\n", ip.c_str(), selfIndex);
//string f1 = file_fin_fp;
//f1 = f1 + oss.str();
//fin_fp.open(f1.c_str(), ios_base::app);
if (LOGGING) {
string f2 = file_task_fp;
f2 = f2 + oss.str();
task_fp.open(f2.c_str(), ios_base::app);
string f3 = file_log_fp;
f3 = f3 + oss.str();
log_fp.open(f3.c_str(), ios_base::app);
string f4 = file_migrate_fp;
f4 = f4 + oss.str();
migrate_fp.open(f4.c_str(), ios_base::app);
}
migratev = bitvec(num_nodes);
Package loadPackage, tasksPackage;
string loadmessage("Load Information!");
loadPackage.set_virtualpath(loadmessage);
loadPackage.set_operation(13);
loadstr = loadPackage.SerializeAsString();
stringstream selfIndexstream;
selfIndexstream << selfIndex;
string taskmessage(selfIndexstream.str());
tasksPackage.set_virtualpath(taskmessage);
tasksPackage.set_operation(14);
taskstr = tasksPackage.SerializeAsString();
srand((selfIndex + 1) * (selfIndex + 5));
int rand_wait = rand() % 1000000;
cout << "Worker ip = " << ip << " selfIndex = " << selfIndex << endl;
//cout << "Worker ip = " << ip << " selfIndex = " << selfIndex << " going to wait for " << rand_wait << " seconds" << endl;
usleep(rand_wait);
file_worker_start.append(oss.str());
string cmd("touch ");
cmd.append(file_worker_start);
//executeShell(cmd);
system(cmd.c_str());
FILE *fp = fopen(file_worker_start.c_str(), "w+");
if (fp != NULL) {
//fputs("fopen example", fp);
char fbuf[100];
memset(fbuf, 0, sizeof(fbuf));
sprintf(fbuf, "%s:%d ", ip.c_str(), selfIndex);
fwrite(fbuf, sizeof(char), strlen(fbuf), fp);
fflush(fp);
fclose(fp);
}
/*worker_start.open(file_worker_start.c_str(),
std::ofstream::out | ios_base::app);
if (worker_start.is_open()) {
worker_start << ip << ":" << selfIndex << " ";
worker_start.flush();
worker_start.close();
worker_start.open(file_worker_start.c_str(), ios_base::app);
}
worker_start.open(file_worker_start.c_str(),
std::ofstream::out | ios_base::app);
*/
clock_gettime(CLOCK_REALTIME, &poll_start);
int err;
/*pthread_t *ready_queue_thread = new pthread_t();//(pthread_t*)malloc(sizeof(pthread_t));
pthread_create(ready_queue_thread, &attr, check_ready_queue, NULL);*/
try {
pthread_t *ready_queue_thread = new pthread_t[num_cores];
for (int i = 0; i < num_cores; i++) {
err = pthread_create(&ready_queue_thread[i], &attr,
check_ready_queue, (void*) this);
if (err) {
printf(
"work_steal_init: pthread_create: ready_queue_thread: %s\n",
strerror(errno));
exit(1);
}
}
pthread_t *wait_queue_thread = new pthread_t();
err = pthread_create(wait_queue_thread, &attr, check_wait_queue,
(void*) this);
if (err) {
printf("work_steal_init: pthread_create: wait_queue_thread: %s\n",
strerror(errno));
exit(1);
}
pthread_t *complete_queue_thread = new pthread_t();
err = pthread_create(complete_queue_thread, &attr, check_complete_queue,
(void*) this);
if (err) {
printf(
"work_steal_init: pthread_create: complete_queue_thread: %s\n",
strerror(errno));
exit(1);
}
package_thread_args rq_args, wq_args;
rq_args.source = &insertq_new;
wq_args.source = &waitq;
rq_args.dest = &rqueue;
wq_args.dest = &wqueue;
rq_args.slock = &iq_new_lock;
wq_args.slock = &waitq_lock;
rq_args.dlock = &lock;
wq_args.dlock = &w_lock;
rq_args.worker = this;
wq_args.worker = this;
pthread_t *waitq_thread = new pthread_t();
err = pthread_create(waitq_thread, &attr, HB_insertQ_new,
(void*) &wq_args);
if (err) {
printf("work_steal_init: pthread_create: waitq_thread: %s\n",
strerror(errno));
exit(1);
}
pthread_t *readyq_thread = new pthread_t();
err = pthread_create(readyq_thread, &attr, HB_insertQ_new,
(void*) &rq_args);
if (err) {
printf("work_steal_init: pthread_create: ready_thread: %s\n",
strerror(errno));
exit(1);
}
pthread_t *migrateq_thread = new pthread_t();
err = pthread_create(migrateq_thread, &attr, migrateTasks,
(void*) this);
if (err) {
printf("work_steal_init: pthread_create: migrateq_thread: %s\n",
strerror(errno));
exit(1);
}
pthread_t *notq_thread = new pthread_t();
err = pthread_create(notq_thread, &attr, notQueue, (void*) this);
if (err) {
printf("work_steal_init: pthread_create: notq_thread: %s\n",
strerror(errno));
exit(1);
}
int min_lines = svrclient.memberList.size();
//min_lines++;
string filename(file_worker_start);
//string cmd("wc -l ");
//cmd = cmd + filename + " | awk {\'print $1\'}";
string cmd2("ls -l ");
cmd2.append(shared);
cmd2.append("startinfo*");
cmd2.append(" | wc -l");
string result = executeShell(cmd2);
//cout << "server: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
while (atoi(result.c_str()) < min_lines) {
sleep(2);
//cout << "server: " << worker.selfIndex << " minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
result = executeShell(cmd2);
}
//cout << "server: " << selfIndex << " minlines = " << min_lines << " cmd = " << cmd2 << " result = " << result << endl;
/*int num = min_lines - 1;
stringstream num_ss;
num_ss << num;
//min_lines++;
string cmd1("cat "); cmd1.append(shared); cmd1.append("startinfo"); cmd1.append(num_ss.str()); cmd1.append(" | wc -l");
string result1 = executeShell(cmd1);
//cout << "server: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
while(atoi(result1.c_str()) < 1) {
sleep(2);
result1 = executeShell(cmd1);
}
cout << "worksteal started: server: " << selfIndex << " minlines = " << 1 << " cmd = " << cmd1 << " result = " << result1 << endl;*/
pthread_t *work_steal_thread = new pthread_t(); //(pthread_t*)malloc(sizeof(pthread_t));
err = pthread_create(work_steal_thread, &attr, worksteal, (void*) this);
if (err) {
printf("work_steal_init: pthread_create: work_steal_thread: %s\n",
strerror(errno));
exit(1);
}
delete ready_queue_thread;
delete wait_queue_thread;
delete complete_queue_thread;
delete work_steal_thread;
delete readyq_thread;
delete waitq_thread;
delete migrateq_thread;
delete notq_thread;
} catch (std::bad_alloc& exc) {
cout
<< "work_steal_init: failed to allocate memory while creating threads"
<< endl;
exit(1);
}
}
//monitor the submitted tasks
// by polling each server for its load information. If a server is idle then it will return load = -4 (queue length - num of idle cores)
// so if every server returns load = -4 it implies that all submitted tasks are complete
void *monitor_function(void* args) {
ZHTClient *clientRet = (ZHTClient*)args;
Package loadPackage, loadhpcPackage, shutdownPackage;
string loadmessage("Monitoring Information!");
loadPackage.set_virtualpath(loadmessage);
loadPackage.set_operation(15);
string loadstr = loadPackage.SerializeAsString();
string loadhpcmessage("Monitoring HPC Information!");
loadhpcPackage.set_virtualpath(loadhpcmessage);
loadhpcPackage.set_operation(24);
string loadhpcstr = loadhpcPackage.SerializeAsString();
string endmessage("Shutdown!");
loadPackage.set_virtualpath(endmessage);
loadPackage.set_operation(98);
string endstr = loadPackage.SerializeAsString();
//int num_worker = clientRet.memberList.size();
int num_worker = clientRet->memberList.size();
int num_cores = 2;
int index = 0;
long termination_value = num_worker * num_cores * -1;
int total_avail_cores = num_cores * num_worker;
int32_t total_queued = 0;
int32_t total_idle = 0;
int32_t queued_busy = 0;
int32_t queued_idle = 0;
int32_t queued = 0;
int32_t num_idle = 0;
int32_t num_busy = 0;
int32_t load = 0;
int32_t total_busy = 0;
//int32_t status = 0;
int32_t finished = 0;
int32_t total_msg_count = 0;
int32_t ret = 0;
//sleep(60);
int min_lines = num_worker;
int num = num_worker - 1;
cout << "The Number is " << num << endl;
stringstream num_ss;
num_ss << num;
//min_lines++;
string filename(shared);
filename = filename + "startinfo" + num_ss.str();
cout << "The filename is " << filename << endl;
string cmd("cat ");
cmd = cmd + filename + " | wc -l";
cout << "The command is " << cmd << endl;
string result = executeShell(cmd);
//cout << cmd << " " << result << endl;
//cout << "client: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
/*string filename(shared);
filename = filename + "start_info";
string cmd("wc -l ");
cmd = cmd + filename + " | awk {\'print $1\'}";
string result = executeShell(cmd);*/
while(atoi(result.c_str()) < 1) {
sleep(5);
result = executeShell(cmd); cout << " temp result = " << result << endl;
}
cout << "client: minlines = 1 " << " cmd = " << cmd << " result = " << result << endl;
//cout << "starting to monitor" << endl;
cout << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << endl;
while(1) {
//If mtc task or only complete queue values are taken
if(NUM_OF_CORES == 1)
{
total_queued = 0;
total_idle = 0;
queued_busy = 0;
stringstream worker_load;
for(index = 0; index < num_worker; index++) {
//int32_t queued_idle = clientRet.svrtosvr(loadstr, loadstr.length(), index);
queued_idle = clientRet->svrtosvr(loadstr, loadstr.length(), index);
queued = queued_idle/10; // summation of the lengths of the three queues
num_idle = queued_idle%10; // number of idle cores
total_queued = total_queued + queued;
total_idle = total_idle + num_idle;
num_busy = num_cores - num_idle;
load = queued + num_busy;
worker_load << load << " ";
}
loadfile << worker_load.str() << endl;
total_busy = total_avail_cores - total_idle;
queued_busy = total_queued + total_busy;
finished = total_num_tasks - queued_busy;
clock_gettime(CLOCK_REALTIME, &end_tasks);
cout << "Total busy cores " << total_busy << " Total Load on all workers = " << queued_busy << " No. of tasks finished = " << finished << " Total tasks submitted = " << total_num_tasks << endl;//" time = " << end_tasks.tv_sec << " " << end_tasks.tv_nsec << endl;
if (client_logfile.is_open() && cl_LOGGING) {
client_logfile << "Total busy cores " << total_busy << " Total Load on all workers = " << queued_busy << " No. of tasks finished = " << finished << " Total tasks submitted = " << total_num_tasks << endl;
}
if(finished == total_num_tasks) {
clock_gettime(CLOCK_REALTIME, &end_tasks);
cout << "\n\n\n\n==============================All tasks finished===========================\n\n\n\n";
break;
}
usleep(200000);
}
else
{
finished = 0;
for(index = 0; index < num_worker; index++)
{
finished += clientRet->svrtosvr(loadhpcstr, loadhpcstr.length(), index);
}
cout << "The number of finished tasks = " << finished << endl;
if(finished >= total_num_tasks-50) {
clock_gettime(CLOCK_REALTIME, &end_tasks);
cout << "\n\n\n\n==============================All tasks finished===========================\n\n\n\n";
break;
}
usleep(200000);
}
}
total_msg_count = 0;
for(index = 0; index < num_worker; index++) {
//clientRet.svrtosvr(endstr, endstr.length(), index);
ret = clientRet->svrtosvr(endstr, endstr.length(), index);
total_msg_count += ret;
}
cout << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << "\n";
cout << "TIME END: " << end_tasks.tv_sec << " SECONDS " << end_tasks.tv_nsec << " NANOSECONDS" << "\n";
timespec diff = timediff(start_tasks, end_tasks);
cout << "TIME TAKEN: " << diff.tv_sec << " SECONDS " << diff.tv_nsec << " NANOSECONDS" << "\n";
cout << "Total messages between all servers = " << total_msg_count << endl;
if (client_logfile.is_open() && cl_LOGGING) {
client_logfile << "\n\n\n\n==============================All tasks finished===========================\n\n\n\n";
client_logfile << "TIME START: " << start_tasks.tv_sec << " SECONDS " << start_tasks.tv_nsec << " NANOSECONDS" << "\n";
client_logfile << "TIME END: " << end_tasks.tv_sec << " SECONDS " << end_tasks.tv_nsec << " NANOSECONDS" << "\n";
client_logfile << "TIME TAKEN: " << diff.tv_sec << " SECONDS " << diff.tv_nsec << " NANOSECONDS" << endl;
client_logfile << "Total messages between all servers = " << total_msg_count << endl;
client_logfile.close();
//return 1;
}
pthread_exit(NULL);
}
// thread to steal tasks it ready queue length becomes zero
void* worksteal(void* args) {
//cout << "entered worksteal thread" << endl;
Worker *worker = (Worker*) args;
/*int min_lines = worker->svrclient.memberList.size();
//min_lines++;
string filename(file_worker_start);
//string cmd("wc -l ");
//cmd = cmd + filename + " | awk {\'print $1\'}";
string cmd("ls -l "); cmd.append(shared); cmd.append("startinfo*"); cmd.append(" | wc -l");
string result = executeShell(cmd);
//cout << "server: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
while(atoi(result.c_str()) < min_lines) {
sleep(2);
//cout << "server: " << worker.selfIndex << " minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
result = executeShell(cmd);
}
cout << "server: " << worker->selfIndex << " minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;*/
int num = worker->svrclient.memberList.size() - 1;
stringstream num_ss;
num_ss << num;
//min_lines++;
string cmd1("cat ");
cmd1.append(shared);
cmd1.append("startinfo");
cmd1.append(num_ss.str());
cmd1.append(" | wc -l");
string result1 = executeShell(cmd1);
//cout << "server: minlines = " << min_lines << " cmd = " << cmd << " result = " << result << endl;
while (atoi(result1.c_str()) < 1) {
usleep(ws_sleep);
result1 = executeShell(cmd1);
}
//cout << "worksteal started: server: " << worker->selfIndex << " minlines = " << 1 << " cmd = " << cmd1 << " result = " << result1 << endl;
while (work_steal_signal) {
//while(ready_queue->get_length() > 0) { }
while (rqueue.size() > 0) {
}
usleep(1000);
// If there are no waiting ready tasks, do work stealing
//if (worker.num_nodes > 1 && ready_queue->get_length() < 1)
if (worker->num_nodes > 1 && rqueue.size() < 1) {
int32_t success = worker->steal_task();
// Do work stealing until succeed
while (success == 0) {
failed_attempts++;
if (failed_attempts >= fail_threshold) {
work_steal_signal = 0;
cout << worker->selfIndex << " stopping worksteal" << endl;
break;
}
usleep(worker->poll_interval);
success = worker->steal_task();
}
failed_attempts = 0;
//cout << "Received " << success << " tasks" << endl;
}
}
}
