Experiment: Running Apache OpenWhisk for Heavy Workloads on Public CloudsΒΆ

Apache OpenWhisk

Lightweight dynamic applications on cloud computing have been migrated to serverless computing platforms, such as Apache OpenWhisk, recently due to its elasticity and simplicity of compute resource provisioning. Mobile applications, image data processing and system log analysis are implemented as a stateless function to process given tasks on serverless computing environments. For example, less than 300 lines of a Python function with TensorFlow library identifies a thousand of training image data from ImageNet Large Visual Recognition Challenge in 2012 within a few seconds using serverless concurrent function invocations. Functions for heavy workloads with external libraries, however, would not be loaded on the serverless environments due to its size and its long execution time. Applications for big data, deep learning, computer vision and genomics typically require multiple and complicated libraries along with cpu and data intensive tasks.

We have three objectives to achieve during this experiment:

  • Sucessful deployment of Apache OpenWhisk on public clouds to comare its behavior with a series of functions for big data, deep learning and genomics applications
  • Benchmark regarding to concurrent function invocations on IaaS-powered serverless computing environments
  • Practical guide to building services on public clouds with the experience of deploying Apache OpenWhisk.

Target cloud providers are:

  • Oracle Cloud Infrastructure
  • Amazon EC2

Target serverless platform is:

  • Apache OpenWhisk
oracle cloud infrastructure Amazon EC2