Rooba
/
boinc
mirror of https://github.com/BOINC/boinc.git
1
0
Fork 0
Code Issues 1 Packages Projects Releases Wiki Activity

- client (Unix) Add /usr/local/cuda/lib/ to LD_LIBRARY_PATH 

before looking for CUDA library
- scheduler: some additional work on matchmaker scheduling
    Changed check_app_filter() so that it doesn't depend on
    the current multi-phase approach;
    move that logic to scan_array()


svn path=/trunk/boinc/; revision=15109
This commit is contained in:
David Anderson 2008-04-30 20:31:33 +00:00
parent 7da4959dff
commit 938d77ce4c
6 changed files with 157 additions and 104 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
checkin_notes
View File

@ -3453,3 +3453,17 @@ Rom April 30 2008
graphics2_unix.C
graphics2_win.C
David April 30 2008
- client (Unix) Add /usr/local/cuda/lib/ to LD_LIBRARY_PATH
before looking for CUDA library
- scheduler: some additional work on matchmaker scheduling
Changed check_app_filter() so that it doesn't depend on
the current multi-phase approach;
move that logic to scan_array()
lib/
coproc.C
sched/
sched_array.C
sched_locality.C
sched_send.C,h

56
lib/coproc.C
View File

@ -96,24 +96,24 @@ COPROC* COPROCS::lookup(char* name) {
void COPROC_CUDA::get(COPROCS& coprocs) {
int count, retval;
#ifdef _WIN32
#ifdef _WIN32
int (__stdcall* __cudaGetDeviceCount)( int * );
int (__stdcall* __cudaGetDeviceProperties) ( cudaDeviceProp*, int );
int bufsize=256;
char buf[256], path[256];
HKEY key;
retval = RegOpenKeyEx(
HKEY_LOCAL_MACHINE,
_T("SOFTWARE\\NVIDIA Corporation\\Installed Products\\NVIDIA CUDA"),
NULL,
KEY_READ,
&key
);
if (retval != ERROR_SUCCESS) return;
retval = RegQueryValueEx(key, "InstallDir", NULL, NULL, (LPBYTE)buf, (LPDWORD)&bufsize);
RegCloseKey(key);
if (retval != ERROR_SUCCESS) return;
sprintf(path, "%s\\bin\\cudart.dll", buf);
int bufsize=256;
char buf[256], path[256];
HKEY key;
retval = RegOpenKeyEx(
HKEY_LOCAL_MACHINE,
_T("SOFTWARE\\NVIDIA Corporation\\Installed Products\\NVIDIA CUDA"),
NULL,
KEY_READ,
&key
);
if (retval != ERROR_SUCCESS) return;
retval = RegQueryValueEx(key, "InstallDir", NULL, NULL, (LPBYTE)buf, (LPDWORD)&bufsize);
RegCloseKey(key);
if (retval != ERROR_SUCCESS) return;
sprintf(path, "%s\\bin\\cudart.dll", buf);
HMODULE cudalib = LoadLibrary(path);
__cudaGetDeviceCount = (int(__stdcall*)(int*)) GetProcAddress(cudalib, "cudaGetDeviceCount");
@ -121,11 +121,22 @@ void COPROC_CUDA::get(COPROCS& coprocs) {
__cudaGetDeviceProperties = (int(__stdcall*)(cudaDeviceProp*, int)) GetProcAddress( cudalib, "cudaGetDeviceProperties" );
if (!__cudaGetDeviceProperties) return;
#else
int (*__cudaGetDeviceCount)( int * );
int (*__cudaGetDeviceProperties) ( cudaDeviceProp*, int );
void (*__cudaGetDeviceCount)( int * );
void (*__cudaGetDeviceProperties) ( cudaDeviceProp*, int );
// Add CUDA dir to library path while looking for CUDA lib.
// Leave it there, shouldn't hurt.
//
char* p = getenv("LD_LIBRARY_PATH");
if (!strstr(p, "/usr/local/cuda/lib")) {
char libpath[8192];
sprintf(libpath, "%s:/usr/local/cuda/lib", p);
setenv("LD_LIBRARY_PATH", libpath, 1);
}
#ifdef __APPLE__
void *cudalib = dlopen ("libcudart.dylib", RTLD_NOW );
#else
#else
void *cudalib = dlopen ("libcudart.so", RTLD_NOW );
#endif
if(!cudalib) return;
@ -134,13 +145,12 @@ void COPROC_CUDA::get(COPROCS& coprocs) {
__cudaGetDeviceProperties = (void(*)(cudaDeviceProp*, int)) dlsym( cudalib, "cudaGetDeviceProperties" );
if (!__cudaGetDeviceProperties) return;
#endif
retval = (*__cudaGetDeviceCount)(&count);
if (retval || count < 1) return;
(*__cudaGetDeviceCount)(&count);
if (count < 1) return;
for (int i=0; i<count; i++) {
COPROC_CUDA* cc = new COPROC_CUDA;
retval = (*__cudaGetDeviceProperties)(&cc->prop, i);
if (retval) continue;
(*__cudaGetDeviceProperties)(&cc->prop, i);
cc->count = 1;
strcpy(cc->name, "CUDA");
coprocs.coprocs.push_back(cc);

30
sched/sched_array.C
View File

@ -110,15 +110,33 @@ void scan_work_array(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
continue;
}
wu = wu_result.workunit;
// check app filter if needed
//
if (reply.wreq.user_apps_only &&
(!reply.wreq.beta_only || config.distinct_beta_apps)
) {
if (app_not_selected(wu, sreq, reply)) {
reply.wreq.no_allowed_apps_available = true;
if (config.debug_send) {
log_messages.printf(MSG_DEBUG,
"[USER#%d] [WU#%d] user doesn't want work for this application\n",
reply.user.id, wu.id
);
}
continue;
}
}
// don't send if host can't handle it
//
wu = wu_result.workunit;
retval = wu_is_infeasible(wu, sreq, reply, *app);
retval = wu_is_infeasible_fast(wu, sreq, reply, *app);
if (retval) {
log_messages.printf(MSG_DEBUG,
"[HOST#%d] [WU#%d %s] WU is infeasible: %s\n",
reply.host.id, wu.id, wu.name, infeasible_string(retval)
);
log_messages.printf(MSG_DEBUG,
"[HOST#%d] [WU#%d %s] WU is infeasible: %s\n",
reply.host.id, wu.id, wu.name, infeasible_string(retval)
);
continue;
}

6
sched/sched_locality.C
View File

@ -296,12 +296,8 @@ static int possibly_send_result(
if (!bavp) return ERR_NO_APP_VERSION;
// wu_is_infeasible() returns the reason why the WU is not feasible;
// INFEASIBLE_MEM, INFEASIBLE_DISK, INFEASIBLE_CPU.
// see sched_send.h.
//
APP* app = ssp->lookup_app(wu.appid);
if (wu_is_infeasible(wu, sreq, reply, *app)) {
if (wu_is_infeasible_fast(wu, sreq, reply, *app)) {
return ERR_INSUFFICIENT_RESOURCE;
}

152
sched/sched_send.C
View File

@ -362,7 +362,7 @@ static double estimate_wallclock_duration(
//
static int get_host_info(SCHEDULER_REPLY& reply) {
char buf[8096];
std::string str;
string str;
unsigned int pos = 0;
int temp_int;
bool flag;
@ -389,12 +389,7 @@ static int get_host_info(SCHEDULER_REPLY& reply) {
reply.wreq.host_info.allow_beta_work = flag;
}
// Decide whether or not this computer is a 'reliable' computer
//
double expavg_credit = reply.host.expavg_credit;
double expavg_time = reply.host.expavg_time;
double avg_turnaround = reply.host.avg_turnaround;
update_average(0, 0, CREDIT_HALF_LIFE, expavg_credit, expavg_time);
// Decide whether or not this computer is 'reliable'
// A computer is reliable if the following conditions are true
// (for those that are set in the config file)
// 1) The host average turnaround is less than the config
@ -402,6 +397,10 @@ static int get_host_info(SCHEDULER_REPLY& reply) {
// 2) The host error rate is less then the config max error rate
// 3) The host results per day is equal to the config file value
//
double expavg_credit = reply.host.expavg_credit;
double expavg_time = reply.host.expavg_time;
double avg_turnaround = reply.host.avg_turnaround;
update_average(0, 0, CREDIT_HALF_LIFE, expavg_credit, expavg_time);
if (config.debug_send) {
log_messages.printf(MSG_DEBUG,
"[HOST#%d] Checking if reliable (OS = %s) error_rate = %.3f avg_turnaround = %.0f hrs\n",
@ -414,7 +413,11 @@ static int get_host_info(SCHEDULER_REPLY& reply) {
// larger set of computers to be marked reliable
//
double multiplier = 1.0;
if (strstr(reply.host.os_name,"Windows") || strstr(reply.host.os_name,"Linux") || (strstr(reply.host.os_name,"Darwin") && !(strstr(reply.host.p_vendor,"Power Macintosh")))) {
if (strstr(reply.host.os_name,"Windows")
|| strstr(reply.host.os_name,"Linux")
|| (strstr(reply.host.os_name,"Darwin")
&& !(strstr(reply.host.p_vendor,"Power Macintosh"))
)) {
multiplier = 1.0;
} else {
multiplier = 1.8;
@ -436,36 +439,22 @@ static int get_host_info(SCHEDULER_REPLY& reply) {
return 0;
}
// Check to see if the user has set application preferences.
// If they have, then only send work for the allowed applications
// Return true if the user has set application preferences,
// and this job is not for a selected app
//
static inline int check_app_filter(
bool app_not_selected(
WORKUNIT& wu, SCHEDULER_REQUEST& , SCHEDULER_REPLY& reply
) {
unsigned int i;
if (reply.wreq.host_info.preferred_apps.size() == 0) return 0;
bool app_allowed = false;
if (reply.wreq.host_info.preferred_apps.size() == 0) return false;
for (i=0; i<reply.wreq.host_info.preferred_apps.size(); i++) {
if (wu.appid==reply.wreq.host_info.preferred_apps[i].appid) {
app_allowed = true;
if (wu.appid == reply.wreq.host_info.preferred_apps[i].appid) {
reply.wreq.host_info.preferred_apps[i].work_available = true;
break;
return false;
}
}
if (!app_allowed && reply.wreq.user_apps_only &&
(!reply.wreq.beta_only || config.distinct_beta_apps)
) {
reply.wreq.no_allowed_apps_available = true;
if (config.debug_send) {
log_messages.printf(MSG_DEBUG,
"[USER#%d] [WU#%d] user doesn't want work for this application\n",
reply.user.id, wu.id
);
}
return INFEASIBLE_APP_SETTING;
}
return 0;
return true;
}
// see how much RAM we can use on this machine
@ -586,7 +575,7 @@ static inline int check_deadline(
return 0;
}
// Quick checks (no DB access) to see if the WU can be sent on the host.
// Fast checks (no DB access) to see if the job can be sent to the host.
// Reasons why not include:
// 1) the host doesn't have enough memory;
// 2) the host doesn't have enough disk space;
@ -594,10 +583,7 @@ static inline int check_deadline(
// the host probably won't get the result done within the delay bound
// 4) app isn't in user's "approved apps" list
//
// TODO: this should be used in locality scheduling case too.
// Should move a few other checks from sched_array.C
//
int wu_is_infeasible(
int wu_is_infeasible_fast(
WORKUNIT& wu, SCHEDULER_REQUEST& request, SCHEDULER_REPLY& reply, APP& app
) {
int retval;
@ -631,8 +617,6 @@ int wu_is_infeasible(
}
}
retval = check_app_filter(wu, request, reply);
if (retval) return retval;
retval = check_memory(wu, request, reply);
if (retval) return retval;
retval = check_disk(wu, request, reply);
@ -1061,6 +1045,7 @@ int add_result_to_reply(
//
static void explain_to_user(SCHEDULER_REPLY& reply) {
char helpful[512];
unsigned int i;
// If work was sent from apps // the user did not select, explain
//
@ -1069,7 +1054,7 @@ static void explain_to_user(SCHEDULER_REPLY& reply) {
reply.insert_message(um);
// Inform the user about applications with no work
for (int i=0; i<reply.wreq.host_info.preferred_apps.size(); i++) {
for (i=0; i<reply.wreq.host_info.preferred_apps.size(); i++) {
if (!reply.wreq.host_info.preferred_apps[i].work_available) {
APP* app = ssp->lookup_app(reply.wreq.host_info.preferred_apps[i].appid);
// don't write message if the app is deprecated
@ -1087,7 +1072,7 @@ static void explain_to_user(SCHEDULER_REPLY& reply) {
// Tell the user about applications they didn't qualify for
//
for(int i=0;i<preferred_app_message_index;i++){
for(i=0;i<preferred_app_message_index;i++){
reply.insert_message(reply.wreq.no_work_messages.at(i));
}
USER_MESSAGE um1("You have selected to receive work from other applications if no work is available for the applications you selected", "high");
@ -1103,8 +1088,8 @@ static void explain_to_user(SCHEDULER_REPLY& reply) {
USER_MESSAGE um2("No work sent", "high");
reply.insert_message(um2);
// Inform the user about applications with no work
for(int i=0; i<reply.wreq.host_info.preferred_apps.size(); i++) {
if ( !reply.wreq.host_info.preferred_apps[i].work_available ) {
for(i=0; i<reply.wreq.host_info.preferred_apps.size(); i++) {
if (!reply.wreq.host_info.preferred_apps[i].work_available) {
APP* app = ssp->lookup_app(reply.wreq.host_info.preferred_apps[i].appid);
// don't write message if the app is deprecated
if ( app != NULL ) {
@ -1124,7 +1109,7 @@ static void explain_to_user(SCHEDULER_REPLY& reply) {
}
if (reply.wreq.no_allowed_apps_available) {
USER_MESSAGE um(
"No work available for the applications you have selected. Please check your settings on the website.",
"No work available for the applications you have selected. Please check your settings on the web site.",
"high"
);
reply.insert_message(um);
@ -1306,7 +1291,8 @@ void send_work(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
reply.wreq.infeasible_only = false;
scan_work_array(sreq, reply);
// If the user has said they prefer to only receive work from certain apps
// If user has selected apps but will accept any,
// and we haven't found any jobs for selected apps, try others
//
if (!reply.wreq.nresults && reply.wreq.host_info.allow_non_preferred_apps ) {
reply.wreq.user_apps_only = false;
@ -1334,13 +1320,13 @@ void send_work(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
struct JOB{
int index;
double value;
double score;
double est_time;
double disk_usage;
APP* app;
BEST_APP_VERSION* bavp;
void get_value(SCHEDULER_REQUEST&, SCHEDULER_REPLY&);
void get_score(SCHEDULER_REQUEST&, SCHEDULER_REPLY&);
};
struct JOB_SET {
@ -1351,8 +1337,8 @@ struct JOB_SET {
std::list<JOB> jobs; // sorted high to low
void add_job(JOB&);
double higher_value_disk_usage(double);
double lowest_value();
double higher_score_disk_usage(double);
double lowest_score();
inline bool request_satisfied() {
return est_time >= work_req;
}
@ -1382,10 +1368,10 @@ int read_sendable_result(DB_RESULT& result) {
return 0;
}
// compute a "value" for sending this WU to this host.
// compute a "score" for sending this job to this host.
// return 0 if the WU is infeasible
//
void JOB::get_value(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
void JOB::get_score(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
bool found;
WORKUNIT wu;
int retval;
@ -1393,7 +1379,7 @@ void JOB::get_value(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
WU_RESULT& wu_result = ssp->wu_results[index];
wu = wu_result.workunit;
value = 0;
score = 0;
// Find the app and app_version for the client's platform.
//
@ -1406,9 +1392,8 @@ void JOB::get_value(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
bavp = get_app_version(sreq, reply, wu);
if (!bavp) return;
}
if (app == NULL) return; // this should never happen
retval = wu_is_infeasible(wu, sreq, reply, *app);
retval = wu_is_infeasible_fast(wu, sreq, reply, *app);
if (retval) {
if (config.debug_send) {
log_messages.printf(MSG_DEBUG,
@ -1419,24 +1404,50 @@ void JOB::get_value(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
return;
}
value = 1;
double val = 1;
score = 1;
// check if user has selected apps
//
if (!reply.wreq.host_info.allow_beta_work || config.distinct_beta_apps) {
if (app_not_selected(wu, sreq, reply)) {
if (!reply.wreq.host_info.allow_non_preferred_apps) {
score = 0;
return;
}
} else {
if (reply.wreq.host_info.allow_non_preferred_apps) {
score += 1;
}
}
}
// if it's a beta user, try to send beta jobs
//
if (app->beta) {
if (reply.wreq.host_info.allow_beta_work) {
value += 1;
score += 1;
} else {
value = 0;
score = 0;
return;
}
} else {
if (reply.wreq.host_info.reliable && (wu_result.need_reliable)) {
value += 1;
}
}
// if job needs to get done fast, send to fast/reliable host
//
if (reply.wreq.host_info.reliable && (wu_result.need_reliable)) {
score += 1;
}
// if job already committed to an HR class,
// try to send to host in that class
//
if (wu_result.infeasible_count) {
value += 1;
score += 1;
}
// If user has selected some apps but will accept jobs from others,
// try to send them jobs from the selected apps
//
}
bool wu_is_infeasible_slow(
@ -1523,12 +1534,12 @@ bool wu_is_infeasible_slow(
return false;
}
double JOB_SET::lowest_value() {
double JOB_SET::lowest_score() {
if (jobs.empty()) return 0;
return jobs.back().value;
return jobs.back().score;
}
// add the given job, and remove lowest-value jobs
// add the given job, and remove lowest-score jobs
// that are in excess of work request
// or that cause the disk limit to be exceeded
//
@ -1551,7 +1562,7 @@ void JOB_SET::add_job(JOB& job) {
}
list<JOB>::iterator i = jobs.begin();
while (i != jobs.end()) {
if (i->value < job.value) {
if (i->score < job.score) {
jobs.insert(i, job);
break;
}
@ -1564,13 +1575,13 @@ void JOB_SET::add_job(JOB& job) {
disk_usage += job.disk_usage;
}
// return the disk usage of jobs above the given value
// return the disk usage of jobs above the given score
//
double JOB_SET::higher_value_disk_usage(double v) {
double JOB_SET::higher_score_disk_usage(double v) {
double sum = 0;
list<JOB>::iterator i = jobs.begin();
while (i != jobs.end()) {
if (i->value < v) break;
if (i->score < v) break;
sum += i->disk_usage;
i++;
}
@ -1623,8 +1634,11 @@ void send_work_matchmaker(SCHEDULER_REQUEST& sreq, SCHEDULER_REPLY& reply) {
JOB job;
job.index = i;
job.get_value(sreq, reply);
if (job.value > jobs.lowest_value()) {
job.get_score(sreq, reply);
log_messages.printf(MSG_NORMAL,
"score for %s: %f\n", wu_result.workunit.name, job.score
);
if (job.score > jobs.lowest_score()) {
ssp->wu_results[i].state = pid;
unlock_sema();
if (wu_is_infeasible_slow(wu_result, sreq, reply)) {

3
sched/sched_send.h
View File

@ -44,7 +44,7 @@ extern bool app_core_compatible(WORK_REQ& wreq, APP_VERSION& av);
#define INFEASIBLE_HR 8
#define INFEASIBLE_BANDWIDTH 9
extern int wu_is_infeasible(
extern int wu_is_infeasible_fast(
WORKUNIT&, SCHEDULER_REQUEST&, SCHEDULER_REPLY&, APP&
);
@ -59,3 +59,4 @@ extern int update_wu_transition_time(WORKUNIT wu, time_t x);
extern void lock_sema();
extern void unlock_sema();
extern const char* infeasible_string(int);
extern bool app_not_selected(WORKUNIT&, SCHEDULER_REQUEST&, SCHEDULER_REPLY&);