Xiulong Yuan
Xiulong Yuan
OK～
可以的，这估计得单独放一个通信的章节。点对点通信库，集合通讯库，通信优化算法，通信backend这些～
这块要写的话我也可以参与写一下。对这块最近研究比较多 @luomai
> @eedalong 你准备写tensorpipe？ tensorpipe可能不太会写，感觉tensorpipe从技术上没有很多的新意可以写主要是，更多是软件架构设计层面比较全
Great
> I requested for you to be added to the Slack, hopefully it will be accepted soon. > > As for connection/pipe/listener/..., those are the names of the individual classes...
本书的不少作者来自于华为，自然对MindSpore熟悉一些，但是机器学习框架的实现原理大同小异，所以以MindSpore为例子进行介绍并没有什么问题。然而我们肯定希望给读者一个更全面的视角，于是再横向对比各个框架在同一个技术点或者需求点的设计与实现。我理解这也没有什么问题。
> If I keep calling `auto pipe = context->connect();pipe->write()` until write successfully in client side, will I hit any potential resource leak or run out issue? Just assume the server...
Tensor applications always need to transfer dense tensors which consumes pretty large memory. Traditional socket stack always involves memory buffer copy overhead which we want to avoid。  RDMA can...