A comparison of some of the features of the three services:

Workbench

• Software that can be installed on a VM or cluster, typically run on NCSA Nebula, SDSC Cloud, Jetstream OpenStack NDS allocations.
• Always uses Kubernetes (either in single-node or cluster installation) for scalability
• Can run a wide variety of Docker-based applications, including multi-component services with dependencies
  • May also run Jupyter/RStudio and other similar interactive environments
• Distributed storage can be mounted into user environments
• Local authentication or Oauth supported
• Has built-in account registration/approval workflow
• Customizable resources constraints (per-container memory/CPU and per-user storage)

JupyterHub/BinderHub

• Software that can be installed on a VM or cluster
• Zero-to-JupterHub provides scalable solution using Kubernetes, typically run on AWS/GCE, but can be run on OpenStack
• Primarily for Jupyter environments, but can also support RStudio 
• Distributed storage can be mounted into user environments
• Customizable authentication
• Customizable resources constraints (per-container memory/CPU and per-user storage)
  
  

MyBinder

• Public BinderHub (JupterHub + repo2Docker)
• Given a Github repo with Juypter notebooks or R scripts, builds and runs a Docker image on behalf of the user
• Free, publicly available service hosted on AWS
• Resource limits maximum 2GB RAM, 10 minute inactivity timeout, 12 hour session
• Supports Jupyter, RStudio environments  
• Can be used for demos and workshops with no service guarantee