Merge pull request #4859 from BOINC/dpa_cpu_usage2

fix non-BOINC CPU usage limit on Win and Linux

Author: davidpanderson

client/app.cpp

@@ -359,8 +359,8 @@ void ACTIVE_TASK_SET::get_memory_usage() {
     unsigned int i;
     int retval;
     static bool first = true;
-    static double last_cpu_time;
     double diff=0;
+    bool using_vbox = false;
 
     if (!first) {
         diff = gstate.now - last_mem_time;
@@ -412,6 +412,7 @@ void ACTIVE_TASK_SET::get_memory_usage() {
         }
         procinfo_app(pi, v, pm, atp->app_version->graphics_exec_file);
         if (atp->app_version->is_vm_app) {
+            using_vbox = true;
             // the memory of virtual machine apps is not reported correctly,
             // at least on Windows.  Use the VM size instead.
             //
@@ -485,13 +486,32 @@ void ACTIVE_TASK_SET::get_memory_usage() {
         }
     }
 
-    // get info on non-BOINC processes.
+#if defined(__linux__) || defined(_WIN32)
+    // compute non_boinc_cpu_usage
+    // Improved version for systems where we can get total CPU (Win, Linux)
+    //
+    static double last_nbrc=0;
+    double nbrc = total_cpu_time() - boinc_related_cpu_time(pm, using_vbox);
+    double delta_nbrc = nbrc - last_nbrc;
+    if (delta_nbrc < 0) delta_nbrc = 0;
+    last_nbrc = nbrc;
+    if (!first) {
+        non_boinc_cpu_usage = delta_nbrc/(diff*gstate.host_info.p_ncpus);
+        //printf("non_boinc_cpu_usage %f\n", non_boinc_cpu_usage);
+    }
+#else
+    // compute non_boinc_cpu_usage
+    //
+    // NOTE: this is flawed because it doesn't count short-lived processes
+    // correctly.  Linux and Win use a better approach (see above).
+    //
     // mem usage info is not useful because most OSs don't
     // move idle processes out of RAM, so physical memory is always full.
     // Also (at least on Win) page faults are used for various things,
     // not all of them generate disk I/O,
     // so they're not useful for detecting paging/thrashing.
     //
+    static double last_cpu_time;
     PROCINFO pi;
     procinfo_non_boinc(pi, pm);
     if (log_flags.mem_usage_debug) {
@@ -508,13 +528,17 @@ void ACTIVE_TASK_SET::get_memory_usage() {
         // processes might have exited in the last 10 sec,
         // causing this to be negative.
         if (non_boinc_cpu_usage < 0) non_boinc_cpu_usage = 0;
+    }
+    last_cpu_time = new_cpu_time;
+#endif
+
+    if (!first) {
         if (log_flags.mem_usage_debug) {
             msg_printf(NULL, MSG_INFO,
                 "[mem_usage] non-BOINC CPU usage: %.2f%%", non_boinc_cpu_usage*100
             );
         }
     }
-    last_cpu_time = new_cpu_time;
     first = false;
 }
 

client/cs_prefs.cpp

@@ -203,6 +203,8 @@ void CLIENT_STATE::get_disk_shares() {
 // and if it's zero set gpu_suspend_reason
 //
 int CLIENT_STATE::check_suspend_processing() {
+    static double last_cpu_usage_suspend=0;
+
     if (benchmarks_running) {
         return SUSPEND_REASON_BENCHMARKS;
     }
@@ -247,8 +249,18 @@ int CLIENT_STATE::check_suspend_processing() {
         if (now - exclusive_app_running < MEMORY_USAGE_PERIOD + EXCLUSIVE_APP_WAIT) {
             return SUSPEND_REASON_EXCLUSIVE_APP_RUNNING;
         }
-        if (global_prefs.suspend_cpu_usage && non_boinc_cpu_usage*100 > global_prefs.suspend_cpu_usage) {
-            return SUSPEND_REASON_CPU_USAGE;
+
+        // if we suspended because of CPU usage,
+        // don't unsuspend for at least 2*MEMORY_USAGE_PERIOD
+        //
+        if (global_prefs.suspend_cpu_usage) {
+            if (now < last_cpu_usage_suspend+2*MEMORY_USAGE_PERIOD) {
+                return SUSPEND_REASON_CPU_USAGE;
+            }
+            if (non_boinc_cpu_usage*100 > global_prefs.suspend_cpu_usage) {
+                last_cpu_usage_suspend = now;
+                return SUSPEND_REASON_CPU_USAGE;
+            }
         }
     }
 

lib/procinfo.cpp

@@ -135,10 +135,34 @@ void procinfo_non_boinc(PROCINFO& procinfo, PROC_MAP& pm) {
         procinfo.working_set_size += p.working_set_size;
     }
 #if 0
-    fprintf(stderr, "total non-boinc: %f %f\n", procinfo.user_time, procinfo.kernel_time);
+    fprintf(stderr,
+        "total non-boinc: %f %f\n", procinfo.user_time, procinfo.kernel_time
+    );
 #endif
 }
 
+// get CPU time of BOINC-related processes, low-priority processes,
+// and (if we're using Vbox) the Vbox daemon.
+//
+double boinc_related_cpu_time(PROC_MAP& pm, bool using_vbox) {
+    double sum = 0;
+    PROC_MAP::iterator i;
+    for (i=pm.begin(); i!=pm.end(); ++i) {
+        PROCINFO& p = i->second;
+#ifdef _WIN32
+        if (p.id == 0) continue;    // idle process
+#endif
+        if (
+            p.is_boinc_app
+            || p.is_low_priority
+            || (using_vbox && strstr(p.command, "VBoxSVC"))
+        ) {
+            sum += p.user_time;
+        }
+    }
+    return sum;
+}
+
 double process_tree_cpu_time(int pid) {
     PROC_MAP pm;
     PROCINFO procinfo;

lib/procinfo.h

@@ -82,4 +82,10 @@ extern void procinfo_non_boinc(PROCINFO&, PROC_MAP&);
 extern double process_tree_cpu_time(int pid);
     // get the CPU time of the given process and its descendants
 
+extern double total_cpu_time();
+    // total user-mode CPU time, as reported by OS
+
+extern double boinc_related_cpu_time(PROC_MAP&, bool using_vbox);
+    // total CPU of current BOINC processes, low-priority processes,
+    // and (if using vbox) the Vbox daemon
 #endif

lib/procinfo_unix.cpp

@@ -263,3 +263,35 @@ int procinfo_setup(PROC_MAP& pm) {
     find_children(pm);
     return 0;
 }
+
+// get total user-mode CPU time
+// see https://www.baeldung.com/linux/get-cpu-usage
+//
+double total_cpu_time() {
+    char buf[1024];
+    static FILE *f=NULL;
+    static double scale;
+    if (!f) {
+        f = fopen("/proc/stat", "r");
+        if (!f) {
+            fprintf(stderr, "can't open /proc/stat\n");
+            return 0;
+        }
+        long hz = sysconf(_SC_CLK_TCK);
+        scale = 1./hz;
+    } else {
+        fflush(f);
+        rewind(f);
+    }
+    if (!fgets(buf, 256, f)) {
+        fprintf(stderr, "can't read /proc/stat\n");
+        return 0;
+    }
+    double user, nice;
+    int n = sscanf(buf, "cpu %lf %lf", &user, &nice);
+    if (n != 2) {
+        fprintf(stderr, "can't parse /proc/stat: %s\n", buf);
+        return 0;
+    }
+    return (user+nice)*scale;
+}

lib/procinfo_win.cpp

@@ -149,3 +149,15 @@ int procinfo_setup(PROC_MAP& pm) {
     }
     return 0;
 }
+
+// get total CPU time
+// see https://docs.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-getsystemtimes
+//
+double total_cpu_time() {
+    FILETIME i, s, u;
+    GetSystemTimes(&i, &s, &u);
+    ULARGE_INTEGER x;
+    x.LowPart = u.dwLowDateTime;
+    x.HighPart = u.dwHighDateTime;
+    return (double)x.QuadPart/1e7;
+}