Planning for IBM Software Hub on IBM Z and LinuxONE

When you install IBM Software Hub on IBM Z and LinuxONE, you can use certain capabilities of IBM Software Hub related to the platform, data preparation, model building, deployment, and monitoring. You must also consider the restrictions that apply when you install IBM Software Hub on IBM Z and LinuxONE.

Capabilities on IBM Z and LinuxONE

You can use the capabilities on Linux operating systems on IBM Z and LinuxONE platforms.

Component Capability
Platform
  • Use the platform services, including backup and restore.
  • Use the storage solutions for s390x architecture.
Data preparation

Use DataStage to move and transform data between operational, transactional, and analytical target systems.
Model building
  • Use Watson Studio to create projects and organize assets such as data, environments, notebooks.
  • Connecting to data sources and building predictive models with AutoAI.
  • Build machine learning models, Python functions, or Python scripts programmatically by using Jupyter Notebooks.
  • To run notebooks, Python functions, or Python scripts, you can create jobs.
  • Create custom environments, such as environments based on Python 3.11 in Watson Studio.
  • Create and run Spark applications on Analytics Engine powered by Apache Spark. The supported Spark versions include Spark 3.4 and Python 3.11.
Deployment
  • Deploy your AI assets by using Watson Machine Learning with Linux on IBM Z and LinuxONE.
  • Create deployment spaces to organize assets and deploy models by using Watson Machine Learning.
  • Create jobs to run notebooks, Python functions, or Python scripts.
Monitoring Monitor and evaluate your deployed model’s performance by using Watson OpenScale.
Governance Use AI Factsheets to track machine learning models from request to production in AI use cases. Use the detailed records in the factsheets to support your governance and compliance goals.

Restrictions on IBM Z and LinuxONE

Attention: AI model deployments in Watson Machine Learning in Cloud Pak for Data 5.1.1 release may not work on s390x. It is advisable to use a different version of Cloud Pak for Data when Watson Machine Learning deployments are required on s390x.

You cannot use the following capabilities with Linux operating system on IBM Z and IBM LinuxONE platforms.

Component Restrictions
Platform Offline volume backup is not supported with Linux on IBM Z and LinuxONE.
Data preparation
  • You cannot use the Sequential file connector to access Parquet files in Google Cloud Storage on DataStage on s390x hardware.
  • You cannot use the Teradata Python driver in DataStage on s390x hardware.
  • You cannot use the remote data plane and remote data engine in DataStage on s390x hardware.
Model building
  • You cannot build models and assets based on the following frameworks and runtime environments in Watson Studio with Linux on IBM Z and LinuxONE:
    • RStudio, including RShiny applications, or assets based on the R framework.
    • Spark, including Spark and R.
    • Python 3.7
    • ONNX 1.8.1
    • Deprecated or discontinued framework version, such as TensorFlow 2.4 or PyTorch 1.7.
    • Tensorflow model training is supported on z16 on-chip AI Accelerator on IBM Z and LinuxONE. However, if you encounter issues, you can disable the usage of AI Accelerator by setting the NNPA_DEVICES environment variable to 0: export NNPA_DEVICES=0.
  • You cannot use the parquet data format as input for an AutoAI time series experiment.
  • Adding data from an unsupported connection might result in kernel failure.
Deployment Deploying models with default runtimes might fail when you use Watson Machine Learning with Linux on IBM Z and LinuxONE. To prevent this, deploy the model by creating a custom runtime.
Monitoring You cannot use Optimized Row Columnar (ORC) input format to evaluate batch deployments with Watson OpenScale.
Governance Integration with OpenPages for synching governance activities with AI Factsheets is not supported. Governance of generative AI assets is not supported.