fluid support asynchronous training

[jacquesqiao]

Project

https://github.com/PaddlePaddle/Paddle/projects/61

Design

• Add async update design doc. Add async update design doc #9932
• Add distributed training overview doc. Add distributed training overview doc #9937

Operators

• VariableResponse support deserialize var into local scope. VariableResponse support deserialize var into local scope #10060
• Refine listen and serve op, Separate RunSyncLoop to a method, prepare for RunAsyncLoop. Refine listen and serve op #10080
• split optimization ops on pserver to independenty blocks split optimization ops on pserver to independenty blocks #10123
• Create sub socpe when it is necessary Create sub socpe when it is necessary #10124
• Add an RunAsyncUpdate(no barrier and no lock) to listen_and_serv_op listen_and_serv_op support async update #9997
  • Prepare optimization block and PrepareContext for each parameter.
  • Add BlockQueue for each parameter block. The queue is used to store the gradient VariableMessage of this parameter from trainers.
  • Add a thread for each parameter to run optimization block.
  • The thread will read gradient from its BlockQueue, create a subscope to deserialize it and then use this subscope to run optimization block.
  • Add one thread to get parameter from the global scope for trainers.(Maybe we need a thread pool to speed up the get process. but it seems that GRPC interface can only work in one thread. Can have a test)
• send_vars and read_vars from pserver without send_barrier and get_barrier.
• Use multi thread todo update Use multi thread to do update #10228

Transpiler #9997

• dist transpile async trainer program. Do not need to add .trainer_n suffix to gradient block in async mode.
• dist transpile async pserver program. Do not need to aggregate gradient block.

Consider

• need to consider how to add learning rate decay in asynchronous training. Do we need lr_decay?

Benchmark

• benchmark of fluid async training benchmark of fluid async training #10180

[typhoonzero]

Add an async_listen_and_serv_op(no barrier and no lock)

I think if we can decouple barrier operations and the grpc_server/client, it's easy to pass an Attr to the op to enable async or not.

[jacquesqiao]

@typhoonzero Ok, I will write some demo code to see if we need an independent async_listen_and_serv_op.

[shanyi15]

您好，此issue在近一个月内暂无更新，我们将于今天内关闭。若在关闭后您仍需跟进提问，可重新开启此问题，我们将在24小时内回复您。因关闭带来的不便我们深表歉意，请您谅解~感谢您对PaddlePaddle的支持!
Hello, this issue has not been updated in the past month. We will close it today for the sake of other user‘s experience. If you still need to follow up on this question after closing, please feel free to reopen it. In that case, we will get back to you within 24 hours. We apologize for the inconvenience caused by the closure and thank you so much for your support of PaddlePaddle Group!