Hardware requirements for Jupyter Notebook Server

The hardware requirements of a Cognos Analytics for Jupyter Server installation depend on the number of concurrent Notebook users and the complexity of their work.

To determine the sizing requirements of your Cognos Analytics Jupyter Server installation, consider these two factors:
  • the number of active Notebook sessions that will run concurrently
  • the complexity of the operations being performed in Notebook
This section provides some rough estimates of sizing examples.
Note: In the following sizing examples, the first three are for a single Cognos Analytics Jupyter Server. However, you can also deploy a clustered architecture behind a reverse-proxy, as described in this high-failover example (https://community.ibm.com/community/user/businessanalytics/blogs/antonio-marziano/2019/06/18/setup-failover-jupyter-using-nginx-reverse-proxy).

Sizing example 1

  • Usage pattern: For 10-20 data scientist users with approximately five concurrent Notebook edit/run/schedule sessions.
  • Sizing requirements: Minimal.
    Note: Notebook consumers, such as widgets in reports/dashboards, or users who view notebooks in read-only (nbViewer) mode do not contribute significantly to the resource requirements. Only Notebook users who edit, run, and schedule notebooks concurrently require increased resources.

Sizing example 2

  • Usage pattern: For small-complexity usage patterns, such as data cleansing and visualizations.
  • Sizing requirements:
    • Number of CPU Cores: 4
    • RAM: 16 Gb

Sizing example 3

  • Usage pattern: For medium-complexity usage patterns, such as profile data quality, random forest classifier, KNN classification and decision trees.
  • Sizing requirements:
    • Number of CPU Cores: 16
    • RAM: 64 Gb

Sizing example 4

  • Usage pattern: For higher-complexity usage patterns, such as deep learning models or neural nets with pyTorch/Tensor Flow.
  • Sizing requirements:
    • high complexity usage patterns have a different resource profile
    • For example, here are the minimum requirements for a four-node IBM DSX-Local installation (https://www.ibm.com/support/knowledgecenter/SSAS34_1.2.1/local/requirements.html?view=kc#requirements__5node).