使用Docker方式来运行GPU任务计算。英伟达(nvidia)官方提供了一个插件NVIDIA-Docker，封装docker的相关参数来绑定GPU相关信息，以给容器提供gpu环境。下面主要介绍使用nvidia-docker插件运行容器以共享宿主机的GPU资源，后面会捎带的讲解如何使用docker原生的方式来运行GPU容器环境。
1. 快速部署安装
# 安装 nvidia-docker and nvidia-docker-plugin
$ wget -P /tmp https://github.com/NVIDIA/nvidia-docker/releases/download/v1.0.1/nvidia-docker-1.0.1-1.x86_64.rpm
$ rpm -i /tmp/nvidia-docker*.rpm && rm /tmp/nvidia-docker*.rpm
$ systemctl start nvidia-docker
# 测试docker容器内部可以识别宿主机的GPU环境
# nvidia-docker run --rm idockerhub.jd.com/nvidia-docker/cuda8.0-runtime:centos6-17-10-19 nvidia-smi
Thu Oct 19 08:07:09 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 375.39 Driver Version: 375.39 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla M40 24GB On | 0000:04:00.0 Off | 0 |
| N/A 28C P8 18W / 250W | 0MiB / 22939MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 Tesla M40 On | 0000:05:00.0 Off | 0 |
| N/A 31C P8 17W / 250W | 0MiB / 11443MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 Tesla M40 24GB On | 0000:06:00.0 Off | 0 |
| N/A 26C P8 18W / 250W | 0MiB / 22939MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 Tesla M40 On | 0000:07:00.0 Off | 0 |
| N/A 27C P8 16W / 250W | 0MiB / 11443MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
注意：上述输出的表格中可用看到，使用nvidia-docker工具创建的容器内部其实是可以识别到宿主机的GPU设备的
2. 容器内部GPU环境验证
使用官方的gpu版本的tensorflow框架来测试：
# nvidia-docker run -it --rm -v /usr/lib64/libcuda.so.1:/usr/local/nvidia/lib64/libcuda.so.1 idockerhub.xxb.com/jdjr/tensorflow-gpu:17-10-17 bash
root@6b4ad215279e:/notebooks# python
Python 2.7.12 (default, Nov 19 2016, 06:48:10)
[GCC 5.4.0 20160609] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
>>> b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
>>> c = tf.matmul(a, b)
>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
2017-10-19 08:01:39.862500: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.1 instructions, but these are available on your machine and could speed up CPU computations.
2017-10-19 08:01:39.862600: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use SSE4.2 instructions, but these are available on your machine and could speed up CPU computations.
2017-10-19 08:01:39.862646: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX instructions, but these are available on your machine and could speed up CPU computations.
2017-10-19 08:01:39.862676: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations.
2017-10-19 08:01:39.862711: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
2017-10-19 08:01:40.388656: I tensorflow/core/common_runtime/gpu/gpu_device.cc:955] Found device 0 with properties:
name: Tesla M40 24GB
major: 5 minor: 2 memoryClockRate (GHz) 1.112
pciBusID 0000:04:00.0
Total memory: 22.40GiB
Free memory: 22.29GiB
2017-10-19 08:01:40.682810: W tensorflow/stream_executor/cuda/cuda_driver.cc:523] A non-primary context 0x1e2dac0 exists before initializing the StreamExecutor. We haven't verified StreamExecutor works with that.
2017-10-19 08:01:40.684222: I tensorflow/core/common_runtime/gpu/gpu_device.cc:955] Found device 1 with properties:
name: Tesla M40
major: 5 minor: 2 memoryClockRate (GHz) 1.112
pciBusID 0000:05:00.0
Total memory: 11.17GiB
Free memory: 11.07GiB
2017-10-19 08:01:40.995170: W tensorflow/stream_executor/cuda/cuda_driver.cc:523] A non-primary context 0x329a280 exists before initializing the StreamExecutor. We haven't verified StreamExecutor works with that.
2017-10-19 08:01:40.998560: I tensorflow/core/common_runtime/gpu/gpu_device.cc:955] Found device 2 with properties:
name: Tesla M40 24GB
major: 5 minor: 2 memoryClockRate (GHz) 1.112
pciBusID 0000:06:00.0
Total memory: 22.40GiB
Free memory: 22.29GiB
2017-10-19 08:01:41.289133: W tensorflow/stream_executor/cuda/cuda_driver.cc:523] A non-primary context 0x329dc00 exists before initializing the StreamExecutor. We haven't verified StreamExecutor works with that.
2017-10-19 08:01:41.290444: I tensorflow/core/common_runtime/gpu/gpu_device.cc:955] Found device 3 with properties:
name: Tesla M40
major: 5 minor: 2 memoryClockRate (GHz) 1.112
pciBusID 0000:07:00.0
Total memory: 11.17GiB
Free memory: 11.07GiB
2017-10-19 08:01:41.294062: I tensorflow/core/common_runtime/gpu/gpu_device.cc:976] DMA: 0 1 2 3
2017-10-19 08:01:41.294083: I tensorflow/core/common_runtime/gpu/gpu_device.cc:986] 0: Y Y Y Y
2017-10-19 08:01:41.294093: I tensorflow/core/common_runtime/gpu/gpu_device.cc:986] 1: Y Y Y Y
2017-10-19 08:01:41.294156: I tensorflow/core/common_runtime/gpu/gpu_device.cc:986] 2: Y Y Y Y
2017-10-19 08:01:41.294178: I tensorflow/core/common_runtime/gpu/gpu_device.cc:986] 3: Y Y Y Y
2017-10-19 08:01:41.294215: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1045] Creating TensorFlow device (/gpu:0) -> (device: 0, name: Tesla M40 24GB, pci bus id: 0000:04:00.0)
2017-10-19 08:01:41.294229: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1045] Creating TensorFlow device (/gpu:1) -> (device: 1, name: Tesla M40, pci bus id: 0000:05:00.0)
2017-10-19 08:01:41.294239: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1045] Creating TensorFlow device (/gpu:2) -> (device: 2, name: Tesla M40 24GB, pci bus id: 0000:06:00.0)
2017-10-19 08:01:41.294248: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1045] Creating TensorFlow device (/gpu:3) -> (device: 3, name: Tesla M40, pci bus id: 0000:07:00.0)
Device mapping:
/job:localhost/replica:0/task:0/gpu:0 -> device: 0, name: Tesla M40 24GB, pci bus id: 0000:04:00.0
/job:localhost/replica:0/task:0/gpu:1 -> device: 1, name: Tesla M40, pci bus id: 0000:05:00.0
/job:localhost/replica:0/task:0/gpu:2 -> device: 2, name: Tesla M40 24GB, pci bus id: 0000:06:00.0
/job:localhost/replica:0/task:0/gpu:3 -> device: 3, name: Tesla M40, pci bus id: 0000:07:00.0
2017-10-19 08:01:41.875931: I tensorflow/core/common_runtime/direct_session.cc:300] Device mapping:
/job:localhost/replica:0/task:0/gpu:0 -> device: 0, name: Tesla M40 24GB, pci bus id: 0000:04:00.0
/job:localhost/replica:0/task:0/gpu:1 -> device: 1, name: Tesla M40, pci bus id: 0000:05:00.0
/job:localhost/replica:0/task:0/gpu:2 -> device: 2, name: Tesla M40 24GB, pci bus id: 0000:06:00.0
/job:localhost/replica:0/task:0/gpu:3 -> device: 3, name: Tesla M40, pci bus id: 0000:07:00.0
>>> print(sess.run(c))
MatMul: (MatMul): /job:localhost/replica:0/task:0/gpu:0
2017-10-19 08:01:51.333248: I tensorflow/core/common_runtime/simple_placer.cc:872] MatMul: (MatMul)/job:localhost/replica:0/task:0/gpu:0
b: (Const): /job:localhost/replica:0/task:0/gpu:0
2017-10-19 08:01:51.333346: I tensorflow/core/common_runtime/simple_placer.cc:872] b: (Const)/job:localhost/replica:0/task:0/gpu:0
a: (Const): /job:localhost/replica:0/task:0/gpu:0
2017-10-19 08:01:51.333408: I tensorflow/core/common_runtime/simple_placer.cc:872] a: (Const)/job:localhost/replica:0/task:0/gpu:0
[[ 22. 28.]
[ 49. 64.]]
>>>
注意：由以上完整输出可以看到当前的容器环境内部可以正常识别到GPU设备，并可以进行GPU计算
使用docker原生方式运行GPU容器环境
# docker run $DEVICES -it --rm -v /usr/lib64/libcuda.so.1:/usr/local/nvidia/lib64/libcuda.so.1 -v /usr/lib64/libnvidia-fatbinaryloader.so.375.39:/usr/local/nvidia/lib64/libnvidia-fatbinaryloader.so.375.39 -v /root/gpu-example/:/tmp idockerhub.xxb.com/jdjr/tensorflow-gpu:17-10-17 bash
# docker run --privileged -it --rm -v /usr/lib64/libcuda.so.1:/usr/local/nvidia/lib64/libcuda.so.1 -v /usr/lib64/libnvidia-fatbinaryloader.so.375.39:/usr/local/nvidia/lib64/libnvidia-fatbinaryloader.so.375.39 -v /root/gpu-example/:/tmp idockerhub.xxb.com/jdjr/tensorflow-gpu:17-10-17 bash
网友评论