Addition of Nvidia GPU monitoring module

CMakeLists.txt

@@ -10,6 +10,7 @@ project (konro)
 list (APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake/")
 
 option(TIMING "Enable low level timing" OFF)
+option(NVIDIA "Enable NVIDIA GPU management" OFF)
 
 if (NOT DEFINED CGROUPBASEDIR)
 set(CGROUPBASEDIR "/sys/fs/cgroup" CACHE STRING "The cgroup base directory")
@@ -22,6 +23,10 @@ if (TIMING)
     add_definitions(-DTIMING)
 endif()
 
+if (NVIDIA)
+    add_definitions(-DNVIDIA)
+endif()
+
 add_subdirectory (common)
 add_subdirectory (lib)
 add_subdirectory (rm)
@@ -30,6 +35,7 @@ add_subdirectory (konrolib)
 add_subdirectory (testplatformcontrol)
 add_subdirectory (testthreadsafequeue)
 add_subdirectory (testsensors)
+add_subdirectory (testnvml)
 add_subdirectory (testhwloc)
 add_subdirectory (testkonrolib)
 add_subdirectory (testtimer)

README.md

@@ -17,10 +17,19 @@ A modern resource manager
 - libhwloc-dev
 - libsensors-dev
 - libopencv-dev
+- Cuda Toolkit 
 
 OpenCV is required only for the peopledetect demo.
 If it is not installed, peopledetect is not compiled.
+# Optional: Nvidia GPU support
+nvidia-cuda-toolkit is required in order to compile with NVIDIA GPU monitoring. It can be installed from your package manager or by manually by following this link https://developer.nvidia.com/cuda-downloads.
+In order to compile, you need to add the NVIDIA flag to cmake:
+* `mkdir build && cd build`
+* `cmake -DNVIDIA=ON ..`  
+* `make`
 
+If no NVIDIA GPUs are present, the manager will still work, but the monitor will just skip them.  
+If nothing is specified, NVIDIA GPU support will be disabled.
 # Development
 
 QtCreator has been used as IDE for development on Ubuntu 22.04.2 LTS.

rm/CMakeLists.txt

@@ -9,6 +9,10 @@ add_subdirectory (platformcontrol)
 add_subdirectory (workloadmanager)
 add_subdirectory (policymanager)
 add_subdirectory (platformmonitor)
+if(NVIDIA)
+    # Enable nvml code
+    add_subdirectory (nvidiamonitor)
+endif(NVIDIA)
 
 file(GLOB konro_SOURCES "src/*.cpp")
 file(GLOB konro_HEADERS "src/*.h")
@@ -17,6 +21,18 @@ include_directories(${PROJECT_SOURCE_DIR}/rm/include)
 
 add_executable (konro ${konro_SOURCES} ${konro_HEADERS})
 target_compile_definitions(konro PUBLIC KONRO_RM)
+
+if(NVIDIA)
+target_link_libraries(konro PUBLIC
+    konro-common
+    rmcommon
+    platformcontrol
+    workloadmanager
+    policymanager
+    platformmonitor
+        nvidiamonitor
+    ${LOG4CPP_LIBRARIES})
+else(NVIDIA)
 target_link_libraries(konro PUBLIC
     konro-common
     rmcommon
@@ -25,5 +41,6 @@ target_link_libraries(konro PUBLIC
     policymanager
     platformmonitor
     ${LOG4CPP_LIBRARIES})
+endif(NVIDIA)
 
 configure_file(konro.ini.in konro.ini)

rm/librmcommon/events/monitorgpuevent.h

@@ -0,0 +1,52 @@
+#ifndef MONITORGPUEVENT_H
+#define MONITORGPUEVENT_H
+
+#include "baseevent.h"
+#include "../gpuload.h"
+#include "../gpupower.h"
+#include "../gputemperature.h"
+#include <vector>
+#include <iostream>
+
+namespace rmcommon {
+
+/*!
+ * \class event generated by the periodical sampling of machine status information.
+ */
+class MonitorGpuEvent : public BaseEvent {
+    GpuTemperature gpuTemp_;
+    GpuPower gpuActualPower_;
+    GpuLoad gpuLoad_;
+public:
+    MonitorGpuEvent(GpuTemperature temp, GpuPower power, GpuLoad load) :
+        BaseEvent("MonitorEvent"),
+        gpuTemp_(temp),
+        gpuActualPower_(power),
+        gpuLoad_(load)
+    {}
+    virtual ~MonitorGpuEvent() = default;
+
+    GpuTemperature getGpuTemperature() const {
+        return gpuTemp_;
+    }
+
+    GpuPower getGpuPower() const {
+        return gpuActualPower_;
+    }
+
+    GpuLoad getGpuLoad() const {
+        return gpuLoad_;
+    }
+
+    virtual void printOnOstream(std::ostream &os) const {
+        os << "{";
+        os << "\"gpuTemp\":" << gpuTemp_;
+        os << ",\"gpuPower\":" << gpuActualPower_;
+        os << ",\"gpuLoad\":" << gpuLoad_;
+        os << "}";
+    }
+};
+
+}   // namespace rmcommon
+
+#endif // MONITORGPUEVENT_H

rm/librmcommon/gpuload.h

@@ -0,0 +1,46 @@
+#ifndef GPULOAD_H
+#define GPULOAD_H
+
+#include <vector>
+#include <iostream>
+
+namespace rmcommon {
+/*!
+ * \class encapsulates GPU utilization information
+ */
+class GpuLoad {
+    /*! % Usage of each GPU */
+    std::vector<int> gpu_;
+
+public:
+    void addGpuLoad(int load) {
+        gpu_.push_back(load);
+    }
+
+    int getGpuLoad(int idx) {
+        return gpu_[idx];
+    }
+
+
+    friend std::ostream &operator << (std::ostream &os, const GpuLoad &pt) {
+        bool first;
+        os << '{'
+           << "\"gpus\":";
+        os << '[';
+        first = true;
+        for (const auto &gpu: pt.gpu_) {
+            if (first)
+                first = false;
+            else
+                os << ',';
+            os << gpu;
+        }
+        os << ']';
+        os << '}';
+        return os;
+    }
+};
+
+}   // namespace rmcommon
+
+#endif // GPULOAD_H

rm/librmcommon/gpupower.h

@@ -0,0 +1,44 @@
+#ifndef GPUPOWER_H
+#define GPUPOWER_H
+
+#include <iostream>
+
+namespace rmcommon {
+/*!
+ * \class encapsulates power information about the GPU
+ */
+class GpuPower {
+    /* Current power in mW */
+    std::vector<int> gpu_;
+public:
+
+    void addGpuPower(int val) {
+        gpu_.push_back(val);
+    }
+
+    int getGpuPower(int index) {
+        return gpu_[index];
+    }
+
+    friend std::ostream &operator << (std::ostream &os, const GpuPower &pp) {
+        bool first;
+        os << '{'
+           << "\"gpus\":";
+        os << '[';
+        first = true;
+        for (const auto &gpu: pp.gpu_) {
+            if (first)
+                first = false;
+            else
+                os << ',';
+            os << gpu;
+        }
+        os << ']';
+        os << '}';
+        return os;
+    }
+};
+
+}   // namespace rmcommon
+
+#endif // GPUPOWER_H

rm/librmcommon/gputemperature.h

@@ -0,0 +1,44 @@
+#ifndef GPUTEMPERATURE_H
+#define GPUTEMPERATURE_H
+
+#include <iostream>
+
+namespace rmcommon {
+/*!
+ * \class encapsulates temperature information about the GPU
+ */
+    class GpuTemperature {
+        /* Current temperature in CELSIUS */
+        std::vector<int> gpu_;
+    public:
+
+        void addGpuTemperature(int val) {
+            gpu_.push_back(val);
+        }
+
+        int getGpuTemperature(int index) {
+            return gpu_[index];
+        }
+
+        friend std::ostream &operator << (std::ostream &os, const GpuTemperature &pp) {
+            bool first;
+            os << '{'
+               << "\"gpus\":";
+            os << '[';
+            first = true;
+            for (const auto &gpu: pp.gpu_) {
+                if (first)
+                    first = false;
+                else
+                    os << ',';
+                os << gpu;
+            }
+            os << ']';
+            os << '}';
+            return os;
+        }
+    };
+
+}   // namespace rmcommon
+
+#endif // GPUTEMPERATURE_H

rm/nvidiamonitor/CMakeLists.txt

@@ -0,0 +1,13 @@
+set(CMAKE_CXX_STANDARD 14)
+
+find_package(CUDAToolkit REQUIRED)
+
+file(GLOB nvidiamonitor_SOURCES "*.cpp")
+file(GLOB nvidiamonitor_HEADERS "*.h")
+add_library(nvidiamonitor
+            ${nvidiamonitor_HEADERS}
+            ${nvidiamonitor_SOURCES}
+)
+
+target_include_directories(nvidiamonitor INTERFACE .)
+target_link_libraries(nvidiamonitor PUBLIC policymanager CUDA::nvml)