Shortcuts

Profiling PyTorch RPC-Based Workloads

In this recipe, you will learn:

  • An overview of the Distributed RPC Framework.
  • An overview of the PyTorch Profiler.
  • How to use the profiler to profile RPC-based workloads.
  • A short example showcasing how to use the profiler to tune RPC parameters.

Requirements

  • PyTorch 1.6+

The instructions for installing PyTorch are available at pytorch.org.

What is the Distributed RPC Framework?

The Distributed RPC Framework provides mechanisms for multi-machine model training through a set of primitives to allow for remote communication, and a higher-level API to automatically differentiate models split across several machines. For this recipe, it would be helpful to be familiar with the Distributed RPC Framework as well as the RPC Tutorials.

What is the PyTorch Profiler?

The profiler is a context manager based API that allows for on-demand profiling of operators in a model’s workload. The profiler can be used to analyze various aspects of a model including execution time, operators invoked, and memory consumption. For a detailed tutorial on using the profiler to profile a single-node model, please see the Profiler Recipe.

How to use the Profiler for RPC-based workloads

The profiler supports profiling of calls made of RPC and allows the user to have a detailed view into the operations that take place on different nodes. To demonstrate an example of this, let’s first set up the RPC framework. The below code snippet will initialize two RPC workers on the same host, named worker0 and worker1 respectively. The workers will be spawned as subprocesses, and we set some environment variables required for proper initialization.

import torch
import torch.distributed.rpc as rpc
import torch.autograd.profiler as profiler
import torch.multiprocessing as mp
import os
import logging
import sys

logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
logger = logging.getLogger()

def random_tensor():
    return torch.rand((3, 3), requires_grad=True)


def worker(rank, world_size):
    os.environ["MASTER_ADDR"] = "localhost"
    os.environ["MASTER_PORT"] = "29500"
    worker_name = f"worker{rank}"

    # Initialize RPC framework.
    rpc.init_rpc(
        name=worker_name,
        rank=rank,
        world_size=world_size
    )
    logger.debug(f"{worker_name} successfully initialized RPC.")

    pass # to be continued below

    logger.debug(f"Rank {rank} waiting for workers and shutting down RPC")
    rpc.shutdown()
    logger.debug(f"Rank {rank} shutdown RPC")


if __name__ == '__main__':
    # Run 2 RPC workers.
    world_size = 2
    mp.spawn(worker, args=(world_size,), nprocs=world_size)

Running the above program should present you with the following output:

DEBUG:root:worker1 successfully initialized RPC.
DEBUG:root:worker0 successfully initialized RPC.
DEBUG:root:Rank 0 waiting for workers and shutting down RPC
DEBUG:root:Rank 1 waiting for workers and shutting down RPC
DEBUG:root:Rank 1 shutdown RPC
DEBUG:root:Rank 0 shutdown RPC

Now that we have a skeleton setup of our RPC framework, we can move on to sending RPCs back and forth and using the profiler to obtain a view of what’s happening under the hood. Let’s add to the above worker function:

def worker(rank, world_size):
    # Above code omitted...
    if rank == 0:
        dst_worker_rank = (rank + 1) % world_size
        dst_worker_name = f"worker{dst_worker_rank}"
        t1, t2 = random_tensor(), random_tensor()
        # Send and wait RPC completion under profiling scope.
        with profiler.profile() as prof:
            fut1 = rpc.rpc_async(dst_worker_name, torch.add, args=(t1, t2))
            fut2 = rpc.rpc_async(dst_worker_name, torch.mul, args=(t1, t2))
            # RPCs must be awaited within profiling scope.
            fut1.wait()
            fut2.wait()

        print(prof.key_averages().table())

The aforementioned code creates 2 RPCs, specifying torch.add and torch.mul, respectively, to be run with two random input tensors on worker 1. Since we use the rpc_async API, we are returned a torch.futures.Future object, which must be awaited for the result of the computation. Note that this wait must take place within the scope created by the profiling context manager in order for the RPC to be accurately profiled. Running the code with this new worker function should result in the following output:

# Some columns are omitted for brevity, exact output subject to randomness
----------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------
Name                                                              Self CPU total %  Self CPU total   CPU total %      CPU total        CPU time avg     Number of Calls  Node ID
----------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------
rpc_async#aten::add(worker0 -> worker1)                           0.00%            0.000us          0                20.462ms         20.462ms         1                0
rpc_async#aten::mul(worker0 -> worker1)                           0.00%            0.000us          0                5.712ms          5.712ms          1                0
rpc_async#aten::mul(worker0 -> worker1)#remote_op: mul            1.84%            206.864us        2.69%            302.162us        151.081us        2                1
rpc_async#aten::add(worker0 -> worker1)#remote_op: add            1.41%            158.501us        1.57%            176.924us        176.924us        1                1
rpc_async#aten::mul(worker0 -> worker1)#remote_op: output_nr      0.04%            4.980us          0.04%            4.980us          2.490us          2                1
rpc_async#aten::mul(worker0 -> worker1)#remote_op: is_leaf        0.07%            7.806us          0.07%            7.806us          1.952us          4                1
rpc_async#aten::add(worker0 -> worker1)#remote_op: empty          0.16%            18.423us         0.16%            18.423us         18.423us         1                1
rpc_async#aten::mul(worker0 -> worker1)#remote_op: empty          0.14%            15.712us         0.14%            15.712us         15.712us         1                1
----------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------
Self CPU time total: 11.237ms

Here we can see that the profiler has profiled our rpc_async calls made to worker1 from worker0. In particular, the first 2 entries in the table show details (such as the operator name, originating worker, and destination worker) about each RPC call made and the CPU total column indicates the end-to-end latency of the RPC call.

We also have visibility into the actual operators invoked remotely on worker 1 due to RPC. We can see operations that took place on worker1 by checking the Node ID column. For example, we can interpret the row with name rpc_async#aten::mul(worker0 -> worker1)#remote_op: mul as a mul operation taking place on the remote node, as a result of the RPC sent to worker1 from worker0, specifying worker1 to run the builtin mul operator on the input tensors. Note that names of remote operations are prefixed with the name of the RPC event that resulted in them. For example, remote operations corresponding to the rpc.rpc_async(dst_worker_name, torch.add, args=(t1, t2)) call are prefixed with rpc_async#aten::mul(worker0 -> worker1).

We can also use the profiler to gain insight into user-defined functions that are executed over RPC. For example, let’s add the following to the above worker function:

# Define somewhere outside of worker() func.
def udf_with_ops():
    import time
    time.sleep(1)
    t1, t2 = random_tensor(), random_tensor()
    torch.add(t1, t2)
    torch.mul(t1, t2)

def worker(rank, world_size):
    # Above code omitted
    with profiler.profile() as p:
        fut = rpc.rpc_async(dst_worker_name, udf_with_ops)
        fut.wait()
    print(p.key_averages().table())

The above code creates a user-defined function that sleeps for 1 second, and then executes various operators. Similar to what we’ve done above, we send an RPC to the remote worker, specifying it to run our user-defined function. Running this code should result in the following output:

# Exact output subject to randomness
--------------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------
Name                                                                  Self CPU total %  Self CPU total   CPU total %      CPU total        CPU time avg     Number of Calls  Node ID
--------------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------
rpc_async#udf_with_ops(worker0 -> worker1)                            0.00%            0.000us          0                1.008s           1.008s           1                0
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: rand            12.58%           80.037us         47.09%           299.589us        149.795us        2                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: empty           15.40%           98.013us         15.40%           98.013us         24.503us         4                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: uniform_        22.85%           145.358us        23.87%           151.870us        75.935us         2                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: is_complex      1.02%            6.512us          1.02%            6.512us          3.256us          2                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: add             25.80%           164.179us        28.43%           180.867us        180.867us        1                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: mul             20.48%           130.293us        31.43%           199.949us        99.975us         2                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: output_nr       0.71%            4.506us          0.71%            4.506us          2.253us          2                1
rpc_async#udf_with_ops(worker0 -> worker1)#remote_op: is_leaf         1.16%            7.367us          1.16%            7.367us          1.842us          4                1
--------------------------------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------  ---------------

Here we can see that the user-defined function has successfully been profiled with its name (rpc_async#udf_with_ops(worker0 -> worker1)), and has the CPU total time we would roughly expect (slightly greater than 1s given the sleep). Similar to the above profiling output, we can see the remote operators that have been executed on worker 1 as part of executing this RPC request.

In addition, we can visualize remote execution using the tracing functionality provided by the profiler. Let’s add the following code to the above worker function:

def worker(rank, world_size):
    # Above code omitted
    # Will generate trace for above profiling output
    trace_file = "/tmp/trace.json"
    prof.export_chrome_trace(trace_file)
    logger.debug(f"Wrote trace to {trace_file}")

Now, we can load the trace file in Chrome (chrome://tracing). We should see output similar to the following:

../_images/rpc_trace_img1.png

As we can see, we have traced our RPC requests and can also visualize traces of the remote operations, in this case, given in the trace row for node_id: 1.

Example: Using profiler to tune RPC initialization parameters

The following exercise is intended to be a simple example into how one can use statistics and traces from the profiler to guide tuning RPC initialization parameters. In particular, we will focus on tuning the num_worker_threads parameter used during RPC initialization. First, we modify our rpc.init_rpc call to the following:

# Initialize RPC framework.
num_worker_threads = 1
rpc.init_rpc(
  name=worker_name,
  rank=rank,
  world_size=world_size,
  rpc_backend_options = rpc.TensorPipeRpcBackendOptions(num_worker_threads=num_worker_threads)
)

This will initialize the [TensorPipe RPC backend](https://pytorch.org/docs/stable/rpc.html#tensorpipe-backend) with only one thread for processing RPC requests. Next, add the following function somewhere outside of the worker main function:

def num_workers_udf_with_ops():
  t = torch.randn((100, 100))
  for i in range(10):
    t.mul(t)
    t.add(t)
    t = t.relu()
    t = t.sigmoid()
return t

This function is mainly intended to be a dummy CPU-intensive function for demonstration purposes. Next, we add the following RPC and profiling code to our main worker function:

with profiler.profile() as p:
  futs = []
  for i in range(4):
    fut = rpc.rpc_async(dst_worker_name, num_workers_udf_with_ops)
    futs.append(fut)
  for f in futs:
    f.wait()

print(p.key_averages().table())

trace_file = "/tmp/trace.json"
# Export the trace.
p.export_chrome_trace(trace_file)
logger.debug(f"Wrote trace to {trace_file}")

Running the code should return the following profiling statistics (exact output subject to randomness):

-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------
                                               Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg    # of Calls       Node ID
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------
                                            aten::zeros         0.33%     143.557us         0.47%     203.125us      50.781us             4             0
                                            aten::empty         0.24%     101.487us         0.24%     101.487us      12.686us             8             0
                                            aten::zero_         0.04%      17.758us         0.04%      17.758us       4.439us             4             0
rpc_async#num_workers_udf_with_ops(worker0 -> worker...         0.00%       0.000us             0     189.757ms      47.439ms             4             0
# additional columns omitted for brevity
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------

We can see that there were 4 RPC calls as expected taking a total of 190ms. Let’s now tune the num_worker_threads parameter we set earlier, by changing it to num_worker_threads = 8. Running the code with that change should return the following profiling statistics (exact output subject to randomness):

-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------
                                               Name    Self CPU %      Self CPU   CPU total %     CPU total  CPU time avg    # of Calls       Node ID
-------------------------------------------------------  ------------  ------------  ------------  ------------  ------------  ------------  ------------
                                            aten::zeros         0.31%     127.320us         0.53%     217.203us      54.301us             4             0
                                            aten::empty         0.27%     113.529us         0.27%     113.529us      14.191us             8             0
                                            aten::zero_         0.04%      18.032us         0.04%      18.032us       4.508us             4             0
rpc_async#num_workers_udf_with_ops(worker0 -> worker...         0.00%       0.000us             0      94.776ms      23.694ms             4             0

We see a clear ~2x speedup, and hypothesize that this speedup is due to exploiting parallelism on the server due to the additional cores available. However, how can we ensure that this speedup is due to the increase in cores? Taking a look at the trace visualization helps with this. Below is the trace when we set num_worker_threads=1:

../_images/oneworker.png

Focusing on the trace for node 1, we can see that the RPCs are ran serially on the server.

Next, the following is the trace where we set num_worker_threads=8:

../_images/8_workers.png

Based on the latter trace, we can see node 1 was able to execute the RPCs in parallel on the server, due to having additional worker threads. To summarize, we were able to leverage both the profiler’s output report and trace to pick an appropriate num_worker_threads parameter for RPC initialization in this simple exercise.

Putting it all together, we have the following code for this recipe:

import torch
import torch.distributed.rpc as rpc
import torch.autograd.profiler as profiler
import torch.multiprocessing as mp
import os
import logging
import sys

logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
logger = logging.getLogger()

def random_tensor():
  return torch.rand((3, 3), requires_grad=True)

def udf_with_ops():
  import time
  time.sleep(1)
  t1, t2 = random_tensor(), random_tensor()
  torch.add(t1, t2)
  torch.mul(t1, t2)

def num_workers_udf_with_ops():
  t = torch.randn((100, 100))
  for i in range(10):
      t.mul(t)
      t.add(t)
      t = t.relu()
      t = t.sigmoid()
  return t

def worker(rank, world_size):
  os.environ["MASTER_ADDR"] = "localhost"
  os.environ["MASTER_PORT"] = "29500"
  worker_name = f"worker{rank}"

  # Initialize RPC framework.
  num_worker_threads =8
  rpc.init_rpc(
    name=worker_name,
    rank=rank,
    world_size=world_size,
    rpc_backend_options = rpc.TensorPipeRpcBackendOptions(num_worker_threads=num_worker_threads),
  )
  logger.debug(f"{worker_name} successfully initialized RPC.")

  if rank == 0:
    dst_worker_rank = (rank + 1) % world_size
    dst_worker_name = f"worker{dst_worker_rank}"
    t1, t2 = random_tensor(), random_tensor()
    # Send and wait RPC completion under profiling scope.
    with profiler.profile() as prof:
      fut1 = rpc.rpc_async(dst_worker_name, torch.add, args=(t1, t2))
      fut2 = rpc.rpc_async(dst_worker_name, torch.mul, args=(t1, t2))
      # RPCs must be awaited within profiling scope.
      fut1.wait()
      fut2.wait()
    print(prof.key_averages().table())

    with profiler.profile() as p:
      futs = []
      for i in range(4):
        fut = rpc.rpc_async(dst_worker_name, num_workers_udf_with_ops)
        futs.append(fut)
        for f in futs:
          f.wait()

    print(p.key_averages().table())

    trace_file = "/tmp/trace.json"
    # Export the trace.
    p.export_chrome_trace(trace_file)
    logger.debug(f"Wrote trace to {trace_file}")


  logger.debug(f"Rank {rank} waiting for workers and shutting down RPC")
  rpc.shutdown()
  logger.debug(f"Rank {rank} shutdown RPC")



if __name__ == '__main__':
  # Run 2 RPC workers.
  world_size = 2
  mp.spawn(worker, args=(world_size,), nprocs=world_size)

Learn More

공식 문서 (영어)

PyTorch 공식 문서입니다.

공식 문서로 이동

한국어 튜토리얼

한국어로 번역 중인 PyTorch 튜토리얼입니다.

튜토리얼로 이동

커뮤니티

다른 사용자들과 의견을 나눠보세요!

커뮤니티로 이동