提交 c2a16b5c 编写于 作者: H Helin Wang

update OP based parameter server design

上级 74b22c37
...@@ -4,13 +4,13 @@ ...@@ -4,13 +4,13 @@
We propose an approach to implement the parameter server. In this We propose an approach to implement the parameter server. In this
approach, there is no fundamental difference between the trainer and approach, there is no fundamental difference between the trainer and
the parameter server: they both run sub-graphs, but sub-graphs of the parameter server: they both run subgraphs, but subgraphs of
different purposes. different purposes.
## Background ## Background
The previous implementations of the parameter server does not run a The previous implementations of the parameter server does not run a
sub-graph. parameter initialization, optimizer computation, network subgraph. parameter initialization, optimizer computation, network
communication and checkpointing are implemented twice on both the communication and checkpointing are implemented twice on both the
trainer and the parameter server. trainer and the parameter server.
...@@ -26,35 +26,40 @@ server becomes a natural extension. ...@@ -26,35 +26,40 @@ server becomes a natural extension.
### Graph Converter ### Graph Converter
The *graph converter* converts the user-defined operation (OP) graph The *graph converter* converts the user-defined operation (OP) graph
into sub-graphs to be scheduled on different nodes. into subgraphs to be scheduled on different nodes with the following
steps:
1. The user-defined OP graph will be cut into sub-graphs of 1. OP placement: the OPs will be placed on different nodes according
different purposes (e.g., trainer, parameter server) to run on to heuristic that minimizes estimated total computation
different workers. time. Currently we will use a simple heuristic that puts parameter
varable on parameter server workers and everything else on trainer
workers.
1. OPs will be added to the subgraphs, so the subgraphs can 1. Add communication OPs to enable the communication between nodes.
communicate with each other. We will need these OPs: *send*, *recv*,
*gradient accumulator*, *string accumulator*, *loop forever*. We will need these OPs: *Send*, *Recv*, *Enqueue*, *Dequeue*.
Below is an example of converting the user defined graph to the Below is an example of converting the user defined graph to the
sub-graphs for the trainer and the parameter server: subgraphs for the trainer and the parameter server:
<img src="src/local-graph.png" width="300"/> <img src="src/local-graph.png" width="300"/>
After converting: After converting:
<img src="src/dist-graph.png" width="500"/> <img src="src/dist-graph.png" width="700"/>
1. The parameter variable W and it's optimizer subgraph are placed on the parameter server. 1. The parameter variable W and it's optimizer subgraph are placed on the parameter server.
1. Operators are added to the sub-graphs. 1. Operators are added to the subgraphs.
- *send* operator sends data and sender's address to the destination. - *Send* sends data to the connected *Recv* operator. The
- *recv* operator receives data and sender's address from the scheduler on the receive node will only schedule *Recv* operator
destination. It will block until data has been received. to run when the *Send* operator has ran (the *Send* OP will mark
- *gradient accumulator* operator accumulates *N* pieces of the *Recv* OP runnable automatically).
gradients. N=1 in Async-SGD, N>1 in Sync-SGD. - *Enueue* enqueues the input variable, it can block until space
- *string accumulator* accumulates *N* pieces of strings into a become available in the queue.
list of strings. N=1 in Async-SGD, N>1 in Sync-SGD. - *Dequeue* outputs configurable numbers of tensors from the
- *loop forever* runs itself as a target forever. queue. It will block until the queue have the required number of
tensors.
### Benefits ### Benefits
...@@ -71,8 +76,8 @@ After converting: ...@@ -71,8 +76,8 @@ After converting:
### Challenges ### Challenges
- It might be hard for the graph converter to cut a general graph - It might be hard for the graph converter to cut a general graph
(without any hint for which sub-graph is the optimizer). We may need (without any hint for which subgraph is the optimizer). We may need
to label which sub-graph inside the OP graph is the optimizer. to label which subgraph inside the OP graph is the optimizer.
- It's important to balance the parameter shards of on multiple - It's important to balance the parameter shards of on multiple
parameter server. If a single parameter is very big (some parameter server. If a single parameter is very big (some
...@@ -80,3 +85,19 @@ After converting: ...@@ -80,3 +85,19 @@ After converting:
automatically partition the single parameter onto different automatically partition the single parameter onto different
parameter servers when possible (only element-wise optimizer depends parameter servers when possible (only element-wise optimizer depends
on the parameter variable). on the parameter variable).
### Discussion
- In the "Aync SGD" figure, the "W" variable on the parameter server
could be read and wrote concurrently, what is our locking strategy?
- Does our current tensor design supports enqueue (put the input tensor
into the queue tensor)?
- *Dequeue* OP will have variable numbers of output (depends on the
`min_count` attribute), does our current design support it? (similar
question for the *Add* OP)
References:
[1] (TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems)[https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/45166.pdf]
doc/design/ops/src/dist-graph.png

222.7 KB | W: | H:

doc/design/ops/src/dist-graph.png

222.2 KB | W: | H:

doc/design/ops/src/dist-graph.png
doc/design/ops/src/dist-graph.png
doc/design/ops/src/dist-graph.png
doc/design/ops/src/dist-graph.png
  • 2-up
  • Swipe
  • Onion skin
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册