Skip to content

Latest commit

 

History

History

video-analytics

README.md

Video Analytics

diagram

This benchmark models a video analysis system, where image recognition is performed on the individual frames from a video feed. Three functions are involved:

  • The Streaming function retrieves a video fragment and sends it to the decoder.
  • The Decoder splits video fragments into individual frames which are sent to the object recognition function
  • The Object Recognition function (a.k.a. recog) performs object recognition on an image.

The video streaming function is the interface function implementing the standard helloworld grpc service. Upon invokation, it will send the video fragment to the driver which then sends singular frames to the recogniser. The result of the object recognition is relayed back to the callee, and is displayed in order from most likely to least likely tag.

Running this Benchmark

  1. If s3 is used, make sure to set the AWS_ACCESS_KEY and AWS_SECRET_KEY environment variables! The kn_deploy script will then substitute these values into the knative manifests. Example:

    export AWS_ACCESS_KEY=ABCDEFGHIJKLMNOPQRST
export AWS_SECRET_KEY=ABCDEFGHIJKLMNOPQRSTUVWXYZABCDEFGHIJKLMN

  2. Deploy the necessary functions using the kn_deploy script.

    ../../tools/kn_deploy.sh ./knative/knative/s3/*

    Any sub-folder in the knative_yamls directory can be used, and all of the manifests therein must be deployed.

    s3 contains manifests configured to use s3 for file transfer, with tracing disabled. The functions deployed using this set of manifests will have TRANSFER_TYPE set to S3.

    inline contains manifests that use the inline transfer type with tracing disabled. Functions deployed using this set of manifests will have TRANSFER_TYPE set to INLINE.

  3. Invoke the benchmark. The interface function of this benchmark is named streaming. It can be invoked using the invoker or our test client, as described in the running benchmarks document.

Instances

Number of instances per function in a stable flow:

Function Instances Is Configurable
Streaming 1 No
Decoder 1 No
Recog 6 Yes - Set in recog knative manifest and should be equal decoder DecoderFrames.

Parameters

Flags

  • addr - The address of the Decoder
  • p - The port used by the Decoder
  • sp - The port to which the streaming function will listen (which is used for invokation)
  • d - Debug toggle, enables extra logs
  • video - The location of the video file, reference/video.mp4 by default.
  • zipkin - Address of the zipkin span collector

Environment Variables

  • TRANSFER_TYPE - The transfer type to use. Can be INLINE (default), S3, or XDT. Not all benchmarks support all transfer types.
  • AWS_ACCESS_KEY, AWS_SECRET_KEY, AWS_REGION - Standard s3 keys, only needed if the s3 transfer type is used
  • ENABLE_TRACING - Toggles tracing.
  • DecoderFrames - Used to set the number of frames being sent and hence the number of object recognition function instances that are used. 6 by default.
  • CONCURRENT_RECOG - Used in the decoder to toggle if recog functions are called concurrently.