Function Compute (2.0):Best practices for audio and video processing

Scenarios and benefits

With the emergence of highly interactive application scenarios, such as social live streaming, online classrooms, and telemedicine, real-time and quasi-real-time Internet traffic is becoming the trend. In most cases, video platforms must transcode a source video content to output multiple distribution video formats in a 1:N manner based on factors such as bitrate, resolution, channel patch, and playback platform to provide viewers on different playback platforms with different network qualities. Video transcoding is a key step in video production and distribution. An ideal video transcoding solution must be cost-effective in terms of cost (RMB/stream) and power efficiency (watts/stream).

This section describes the benefits of GPU-accelerated instances compared with the instances with no GPU acceleration in Function Compute:

• Real-time and quasi-real-time application scenarios

  GPU-accelerated instances can transcode videos several times faster and push production content to users more efficiently.

• Cost-prioritized GPU application scenarios

  GPU-accelerated instances can be elastically provisioned based on your business requirements, and are more cost-effective than self-purchased virtual machines (VMs).

• Efficiency-prioritized GPU application scenarios

  Focus on code development and the business objectives without the need to perform O&M on GPU clusters, such as driver and CUDA version management, machine operation management, and faulty GPU management.

For more information about GPU-accelerated instances, see Instance types and instance modes.

Performance comparison

The GPU-accelerated instances of Function Compute are based on the Turing architecture and support the following encoding and decoding formats:

• Encoding formats

  H.264 (AVCHD) YUV 4:2:0

  H.264 (AVCHD) YUV 4:4:4

  H.264 (AVCHD) Lossless

  H.265 (HEVC) 4K YUV 4:2:0

  H.265 (HEVC) 4K YUV 4:4:4

  H.265 (HEVC) 4K Lossless

  H.265 (HEVC) 8k

  HEVC10-bitsupport

  HEVCB Framesupport

• Decoding formats

  MPEG-1

  MPEG-2

  VC-1

  VP8

  VP9

  H.264 (AVCHD)

  H.265 (HEVC) 4:2:0

  *H.265 (HEVC) 4:4:4

  8 bit

  10 bit

  12 bit

  8 bit

  10 bit

  12 bit

  8 bit

  10 bit

  12 bit

The following table describes the information about the source video.

The following table describes information about the test machine that uses GPU acceleration and the machine without GPU acceleration.

Video transcoding (1:1)

Performance test: 1 input steram and 1 output stream

Video transcoding (1: N)

Performance test: 1 input stream and 3 output streams

Transcoding commands

• Commands for transcoding without GPU acceleration

  • Single-stream transcoding (1:1)

    docker run --rm -it --volume $PWD:/workspace --runtime=nvidia willprice/nvidia-ffmpeg -y -i input.mp4 -c:v h264 -vf scale=1920:1080 -b:v 5M output.mp4

  • Multiple-stream transcoding (1:N)

    docker run --rm -it --volume $PWD:/workspace --runtime=nvidia willprice/nvidia-ffmpeg \
    -y -i input.mp4 \
    -c:a copy -c:v h264 -vf scale=1920:1080 -b:v 5M output_1080.mp4 \
    -c:a copy -c:v h264 -vf scale=1280:720 -b:v 5M output_720.mp4 \
    -c:a copy -c:v h264 -vf scale=640:480 -b:v 5M output_480.mp4

  Table 1. Parameter description

  Parameter	Description
  -c:a copy	The audio stream can be copied without recoding.
  -c:v h264	Select the software H.264 encoder for the output stream.
  -b:v 5M	Set the output bitrate to 5 Mb/s.

• Commands for transcoding with GPU acceleration

  • Single-stream transcoding (1:1)

    docker run --rm -it --volume $PWD:/workspace --runtime=nvidia willprice/nvidia-ffmpeg -y -hwaccel cuda -hwaccel_output_format cuda -i input.mp4 -c:v h264_nvenc -vf scale_cuda=1920:1080:1:4 -b:v 5M output.mp4

  • Multiple-stream transcoding (1:N)

    docker run --rm -it --volume $PWD:/workspace --runtime=nvidia willprice/nvidia-ffmpeg \
    -y -hwaccel cuda -hwaccel_output_format cuda -i input.mp4 \
    -c:a copy -c:v h264_nvenc -vf scale_npp=1920:1080 -b:v 5M output_1080.mp4 \
    -c:a copy -c:v h264_nvenc -vf scale_npp=1280:720 -b:v 5M output_720.mp4 \
    -c:a copy -c:v h264_nvenc -vf scale_npp=640:480 -b:v 5M output_480.mp4

  Table 2. Parameter description

  Parameter	Description
  -hwaccel cuda	Select a proper hardware accelerator.
  -hwaccel_output_format cuda	Save the decoded frames in the GPU memory.
  -c:v h264_nvenc	Use the NVIDIA hardware to accelerate the H.264 encoder.

Preparations

• For optimal user experience, join the DingTalk group (ID 11721331) and provide the following information:

  • The organization name, such as the name of your company.

  • The ID of your Alibaba Cloud account.

  • The region where you want to use GPU-accelerated instances, such as China (Shenzhen).

  • The contact information, such as your mobile number, email address, or DingTalk account.

• Perform the following operations in the region where the GPU-accelerated instances reside:

  • Create a Container Registry Enterprise Edition instance or Personal Edition instance. We recommend that you create an Enterprise Edition instance. For more information, see Step 1: Create a Container Registry Enterprise Edition instance.

  • Create a namespace and an image repository. For more information, see Step 2: Create a namespace and Step 3: Create an image repository.

• Compile FFmpeg.

  FFmpeg must be compiled before you can use GPU acceleration. The following items describe how to compile FFmpeg:

  • (Recommended) Use compiled FFmpeg through Docker. Download address: nvidia-ffmpeg or ffmpeg.

  • Manually compile FFmpeg. For more information, see the compilation guide.

• Upload the audio and video resources that you want to process to an Object Storage Service (OSS) bucket in the region where the GPU-accelerated instances are located. Make sure that you have the read and write permissions on the objects in the bucket. For more information about how to upload audio and video resources, see Upload objects. For more information about permissions, see Modify the ACL of a bucket.

Deploy a GPU application by using Serverless Devs

Before you start

• Install Serverless Devs and Docker

• Configure Serverless Devs

Procedure

1. Create a project.

   s init devsapp/start-fc-custom-container-event-python3.9 -d fc-gpu-prj

   The following sample code shows the directory of the created project:

   fc-gpu-prj
   ├── code
   │   ├── app.py        # Function code.
   │   └── Dockerfile    # Dockerfile: The image Dockerfile that contains the code.
   ├── README.md
   └── s.yaml            # Project configurations, which specify how the image is deployed in Function Compute

2. Go to the project directory.

   cd fc-gpu-prj

3. Modify the parameter configurations in the following files based on your business requirements.

   • Edit the s.yaml file.

     For more information about the parameters in the YAML file, see YAML specifications.

     edition: 1.0.0
     name: container-demo
     access: default
     vars:
       region: cn-shenzhen
     services:
       customContainer-demo:
         component: devsapp/fc
         props:
           region: ${vars.region}
           service:
             name: tgpu_ffmpeg_service
             internetAccess: true
           function:
             name: tgpu_ffmpeg_func
             description: test gpu for ffmpeg
             handler: not-used
             timeout: 600
             caPort: 9000
             instanceType: fc.gpu.tesla.1
             gpuMemorySize: 8192
             cpu: 4
             memorySize: 16384
             diskSize: 512
             runtime: custom-container
             customContainerConfig:
               #1. Make sure that the namespace:demo namespace and the repo:gpu-transcoding_s repository are created in advance in Alibaba Cloud Container Registry. 
               #2. Modify the tag from v0.1 to v0.2 when you update the function later and run s build && s deploy again. 
               image: registry.cn-shanghai.aliyuncs.com/demo/gpu-transcoding_s:v0.1
             codeUri: ./code

   • Edit the app.py file.

     Example:

     # -*- coding: utf-8 -*-
     # python2 and python3
     
     from __future__ import print_function
     from http.server import HTTPServer, BaseHTTPRequestHandler
     import json
     import sys
     import logging
     import os
     import time
     import urllib.request
     import subprocess
     
     class Resquest(BaseHTTPRequestHandler):
         def download(self, url, path):
             print("enter download:", url)
             f = urllib.request.urlopen(url)
             with open(path, "wb") as local_file:
                 local_file.write(f.read())
     
         def upload(self, url, path):
             print("enter upload:", url)
             headers = {
                 'Content-Type': 'application/octet-stream',
                 'Content-Length': os.stat(path).st_size,
             }
             req = urllib.request.Request(url, open(path, 'rb'), headers=headers, method='PUT')
             urllib.request.urlopen(req)
     
         def trans(self, input_path, output_path, enable_gpu):
             print("enter trans input:", input_path, " output:", output_path, " enable_gpu:", enable_gpu)
     
             cmd = ['ffmpeg', '-y', '-i', input_path, "-c:a", "copy", "-c:v", "h264", "-b:v", "5M", output_path]
             if enable_gpu:
                 cmd = ["ffmpeg", "-y", "-hwaccel", "cuda", "-hwaccel_output_format", "cuda", "-i", input_path, "-c:v", "h264_nvenc", "-b:v", "5M", output_path]
     
             try:
                 subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True)
             except subprocess.CalledProcessError as exc:
                 print('\nreturncode:{}'.format(exc.returncode))
                 print('\ncmd:{}'.format(exc.cmd))
                 print('\noutput:{}'.format(exc.output))
                 print('\nstderr:{}'.format(exc.stderr))
                 print('\nstdout:{}'.format(exc.stdout))
     
         def trans_wrapper(self, enable_gpu):
             src_url = "https://your.domain/input.mp4"  # Use the path of the OSS object under your Alibaba Cloud account. You must have the read and write permissions on the object. 
             dst_url = "https://your.domain/output.flv" # Use the path of the OSS object under your Alibaba Cloud account. You must have the read and write permissions on the object. 
             src_path = "/tmp/input_c.flv"
             dst_path = "/tmp/output_c.mp4"
     
             if enable_gpu:
                 src_url = "https://your.domain/input.mp4"  # Use the path of the OSS object under your Alibaba Cloud account. You must have the read and write permissions on the object. 
                 dst_url = "https://your.domain/output.flv" # Use the path of the OSS object under your Alibaba Cloud account. You must have the read and write permissions on the object. 
                 src_path = "/tmp/input_g.flv"
                 dst_path = "/tmp/output_g.mp4"
     
             local_time = time.time()
             self.download(src_url, src_path)
             download_time = time.time() - local_time
     
             local_time = time.time()
             self.trans(src_path, dst_path, enable_gpu)
             trans_time = time.time() - local_time
     
             local_time = time.time()
             self.upload(dst_url, dst_path)
             upload_time = time.time() - local_time
     
             data = {'result':'ok', 'download_time':download_time, 'trans_time':trans_time, 'upload_time':upload_time}
             self.send_response(200)
             self.send_header('Content-type', 'application/json')
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def pong(self):
             data = {"function":"trans_gpu"}
             self.send_response(200)
             self.send_header('Content-type', 'application/json')
             self.end_headers()
             self.wfile.write(json.dumps(data).encode())
     
         def dispatch(self):
             mode = self.headers.get('TRANS-MODE')
     
             if mode == "ping":
                 self.pong()
             elif mode == "gpu":
                 self.trans_wrapper(True)
             elif mode == "cpu":
                 self.trans_wrapper(False)
             else:
                 self.pong()
     
         def do_GET(self):
             self.dispatch()
     
         def do_POST(self):
             self.dispatch()
     
     if __name__ == '__main__':
         host = ('0.0.0.0', 9000)
         server = HTTPServer(host, Resquest)
         print("Starting server, listen at: %s:%s" % host)
         server.serve_forever()

   • Edit the Dockerfile file.

     Example:

     FROM registry.cn-shanghai.aliyuncs.com/serverless_devs/nvidia-ffmpeg:latest
     WORKDIR /usr/src/app
     RUN apt-get update --fix-missing
     RUN apt-get install -y python3
     RUN apt-get install -y python3-pip
     COPY . .
     ENTRYPOINT [ "python3", "-u", "/usr/src/app/app.py" ]
     EXPOSE 9000

4. Build an image.

   s build --dockerfile ./code/Dockerfile

5. Deploy the code to Function Compute.

   s deploy

   Note

   If you run the preceding command repeatedly and service name and function name remain unchanged, run the use local command to use local configurations.

6. Configure provisioned instances.

   s provision put --target 1 --qualifier LATEST

7. Check whether the provisioned instances are ready.

   s provision get --qualifier LATEST

   If the value of current is 1, the provisioned mode of GPU-accelerated instances is ready. Example:

   [2021-12-14 08:45:24] [INFO] [S-CLI] - Start ...
   [2021-12-14 08:45:24] [INFO] [FC] - Getting provision: tgpu_ffmpeg_service.LATEST/tgpu_ffmpeg_func
   customContainer-demo:
    serviceName:      tgpu_ffmpeg_service
    functionName:      tgpu_ffmpeg_func
    qualifier:       LATEST
    resource:        188077086902****#tgpu_ffmpeg_service#LATEST#tgpu_ffmpeg_func
    target:         1
    current:        1
    scheduledActions:    (empty array)
    targetTrackingPolicies: (empty array)
    currentError:     

8. Invoke the function.

   • View the version of the online function

     s invoke

   • Use CPUs for transcoding

     s invoke -e '{"method":"GET","headers":{"TRANS-MODE":"cpu"}}'

   • Use GPU for transcoding

     s invoke -e '{"method":"GET","headers":{"TRANS-MODE":"gpu"}}'

9. Release the GPU-accelerated instances.

   s provision put --target 0 --qualifier LATEST

Use the Function Compute console to deploy a GPU application

1. Deploy an image.

   1. Create a Container Registry Enterprise Edition instance or Container Registry Personal Edition instance.

      We recommend that you create an Enterprise Edition instance. For more information, see Create a Container Registry Enterprise Edition instance.

   2. Create a namespace and an image repository.

      For more information, see the Step 2: Create a namespace and Step 3: Create an image repository sections of the "Use Container Registry Enterprise Edition instances to build images" topic.

   3. Perform operations on Docker as prompted in the Container Registry console. Then, push the preceding sample app.py and Dockerfile to the instance image repository. For more information about the files, see app.py and Dockerfile in the /code directory when you deploy a GPU application by using Serverless Devs.

      

2. Create a service. For more information, see the "Create a service“ section of the Manage services section.

3. Create a function. For more information, see Create a custom container function.

   Note

   Select GPU Instance for Instance Type and Process HTTP Requests for Request Handler Type.

4. Change the execution timeout period of the function.

   1. Find the function that you want to manage and click Configure in the Actions column.

   2. In the Environment Information section, change the value of Execution Timeout Period and click Save.

      

   Note

   Transcoding duration by using CPU exceeds the default value of 60 seconds. Therefore, we recommend that you set the value of Execution Timeout Period to a larger value.

5. Configure provisioned GPU-accelerated instances.

   1. On the Function Details page, click the Auto Scaling tab and click Create Rule.

   2. On the page that appears, configure the following parameters to provision GPU-accelerated instances and click Create.

      For more information about how to configure provisioned instances, see Configure auto scaling rules.

      After the configuration is complete, you can check whether the provisioned GPU-accelerated instances are ready in the rule list. Specifically, check whether the value of Current Reserved Instances is the specified number of provisioned instances.

6. Use cURL to test the function

   1. On the function details page, click the Triggers tab to view the configurations of the trigger and obtain the endpoint of the trigger.

   2. Run the following command in the command line interface (CLI) to invoke the GPU function:

      • View the version of the online function

        curl -v "https://tgpu-ff-console-tgpu-ff-console-ajezot****.cn-shenzhen.fcapp.run"
        {"function": "trans_gpu"}

      • Use CPUs for transcoding

        curl "https://tgpu-ff-console-tgpu-ff-console-ajezot****.cn-shenzhen.fcapp.run" -H 'TRANS-MODE: cpu'
        {"result": "ok", "upload_time": 8.75510573387146, "download_time": 4.910430669784546, "trans_time": 105.37688875198364}

      • Use GPU for transcoding

        curl "https://tgpu-ff-console-tgpu-ff-console-ajezotchpx.cn-shenzhen.fcapp.run" -H 'TRANS-MODE: gpu'
        {"result": "ok", "upload_time": 8.313958644866943, "download_time": 5.096682548522949, "trans_time": 8.72346019744873}