What is the difference between a tensor flow in a spark discharge with a distributed tensor flow of 1.0 by default?

Question

I am trying to set shadoworflow to a spark on the server, as my boss told me, because he thought it would be easy to use. But I also studied the default distributed tensor flow on the tensorflow website. Can any expert tell me about the difference between the two distribution options? Will the automatic assignment of a parameter server or workers occur?

Thanks in advance.

Answer 1

, TensorflowOnSpark (TFOS) Tensorflow (TF) . :

:

• TFOS . , node , PS node. .
• node, .
• TFOS .

:

• - GPU ​​(K80 ). .
• .
• , node PS node.

- , , .