/HAMi-core

HAMi-core compiles libvgpu.so, which ensures hard limit on GPU in container

Primary LanguageC

HAMi-core —— Hook library for CUDA Environments

Introduction

HAMi-core is the in-container gpu resource controller, it has beed adopted by HAMi, volcano

Features

HAMi-core has the following features:

  1. Virtualize device meory

image

  1. Limit device utilization by self-implemented time shard

  2. Real-time device utilization monitor

Design

HAMi-core operates by Hijacking the API-call between CUDA-Runtime(libcudart.so) and CUDA-Driver(libcuda.so), as the figure below:

Build

sh build.sh

Build in Docker

docker build . -f dockerfiles/Dockerfile

Usage

CUDA_DEVICE_MEMORY_LIMIT indicates the upper limit of device memory (eg 1g,1024m,1048576k,1073741824)

CUDA_DEVICE_SM_LIMIT indicates the sm utility percentage of each device

# Add 1GB bytes limit And set max sm utility to 50% for all devices
export LD_PRELOAD=./libvgpu.so
export CUDA_DEVICE_MEMORY_LIMIT=1g
export CUDA_DEVICE_SM_LIMIT=50

Docker Images

# Make docker image
docker build . -f=dockerfiles/Dockerfile-tf1.8-cu90

# Launch the docker image
export DEVICE_MOUNTS="--device /dev/nvidia0:/dev/nvidia0 --device /dev/nvidia-uvm:/dev/nvidia-uvm --device /dev/nvidiactl:/dev/nvidiactl"
export LIBRARY_MOUNTS="-v /usr/cuda_files:/usr/cuda_files -v $(which nvidia-smi):/bin/nvidia-smi"

docker run ${LIBRARY_MOUNTS} ${DEVICE_MOUNTS} -it \
    -e CUDA_DEVICE_MEMORY_LIMIT=2g \
    cuda_vmem:tf1.8-cu90 \
    python -c "import tensorflow; tensorflow.Session()"

Log

Use environment variable LIBCUDA_LOG_LEVEL to set the visibility of logs

LIBCUDA_LOG_LEVEL description
0 errors only
1(default),2 errors,warnings,messages
3 infos,errors,warnings,messages
4 debugs,errors,warnings,messages

Test Raw APIs

./test/test_alloc