TensorFlow集群部署

TensorFlow

CenterOS 7 安装 Python3

Python3.6.5 安装

tar -zxvf Python-3.6.5.tgz
cd Python-3.6.5/
./configure --prefix=/usr/local/python3
make
make install
# 删除原来的python引用，替换为python3
rm /usr/bin/python
ln -s /usr/local/python3/bin/python3 /usr/bin/python
ln -s /usr/local/python3/bin/pip3 /usr/bin/pip
#测试
python

修复YUM命令

yum命令需要 python2 环境执行
修改 /usr/bin/yum 文件中的第一行为 #!/usr/bin/python2

TensorFlow集群

Tensorflow分布式原理理解

安装TensorFlow

因为CPU型号较老，只能安装 1.5 版本
在节点上分别执行 pip install tensorflow==1.5

server 配置

在第 1 台机器上执行 vi server.py
在另外 2 台机器上分别修改task_index为 1，2

import tensorflow as tf
worker1 = "hbbw214:12000"
worker2 = "hbbw112:10000"
worker3 = "hbbw113:10000"
worker_hosts = [worker1, worker2, worker3]
cluster_spec = tf.train.ClusterSpec({ "worker": worker_hosts})
server = tf.train.Server(cluster_spec, job_name="worker", task_index=0)
server.join()

在第2台机器上执行vi server.py

import tensorflow as tf
worker1 = "hbbw214:12000"
worker2 = "hbbw112:12000"
worker3 = "hbbw113:12000"
worker_hosts = [worker1, worker2, worker3]
cluster_spec = tf.train.ClusterSpec({ "worker": worker_hosts})
server = tf.train.Server(cluster_spec, job_name="worker", task_index=1)
server.join()

在第3台机器上执行vi server.py

import tensorflow as tf
worker1 = "hbbw214:12000"
worker2 = "hbbw112:12000"
worker3 = "hbbw113:12000"
worker_hosts = [worker1, worker2, worker3]
cluster_spec = tf.train.ClusterSpec({ "worker": worker_hosts})
server = tf.train.Server(cluster_spec, job_name="worker", task_index=2)
server.join()

在任意一台机器上执行 vi client.py

import tensorflow as tf
import numpy as np
train_X = np.linspace(-1,1,1000000)
train_Y = 2*train_X + np.random.randn(*train_X.shape)*0.33+10
X = tf.placeholder("float")
Y = tf.placeholder("float")
w = tf.Variable(0.0, name="weight")
b = tf.Variable(0.0, name="reminder")
init_op = tf.global_variables_initializer()
cost_op = tf.square(Y - tf.multiply(X,w) - b)
train_op = tf.train.GradientDescentOptimizer(0.01).minimize(cost_op)
with tf.Session("grpc://hbbw214:10000") as sess:
    with tf.device("/job:worker/task:0"):
        sess.run(init_op)
    for i in range(10):
        for (x, y) in zip(train_X, train_Y):
            sess.run(train_op, feed_dict={X:x, Y:y})
    print(sess.run(w))
    print(sess.run(b))

启动

先分别执行 python server.py
最后执行 python client.py