Content-Aware Storage (CAS)

This section answers questions that are related to Content-Aware Storage in IBM Fusion.

  1. What is Content-Aware Storage (CAS) in IBM Fusion Systems?
    CAS is a storage service that helps unlock the value of unstructured data through AI technologies. It enables the access to this value in new data-driven solutions, without the complexities of implementing and managing large AI solutions.
  2. What is the primary objective of the CAS architecture?
    The CAS architecture is designed to enable seamless interaction between large language models (LLMs) and vast volumes of unstructured data, enhancing insight generation and suggestion capabilities.
  3. How does CAS process unstructured data?
    CAS automatically processes files that are stored on a remote IBM Storage Scale cluster. It extracts parsed content and normalizes metadata, which is then stored along with vector embeddings in a managed vector database.
  4. What type of storage platforms does CAS support?
    CAS supports IBM Storage Scale software-based clusters, and Storage Scale Server 3500 and 6000 platforms. It uses remote mounting by using the IBM Fusion Global Data Platform service.
  5. How does CAS detect and ingest modified data?
    CAS uses IBM Storage Scale watch folders to detect changes (additions, deletions, or updates) in the file system. Only the modified content is ingested, optimizing performance and resource usage.
  6. Can CAS ingest data from external S3 sources?
    Yes, CAS supports ingest processing of files that are stored in external S3 sources by using Active File Management (AFM). It automatically scans AFM-S3 filesets to identify data for ingestion.
  7. What type of document processing services does CAS support?
    CAS integrates with IBM’s Docling Multimodal services and NVIDIA NeMo Retriever Extraction microservices for document parsing and content extraction.
  8. How does CAS perform search operations?
    CAS provides a search API that supports semantic, keyword, and hybrid search methods. It aggregates results and optionally uses NVIDIA’s reranking for enhanced accuracy.
  9. How do customer applications interact with CAS?
    Applications such as chatbots can send natural language queries to the CAS search API. CAS converts the query into a vector representation and performs a search against its vector database, returning optimized results.
  10. What is the role of NVIDIA reranking service in CAS search?
    In the NVIDIA multimodal pipeline, CAS uses an aggregator component to refine the top 100 search results. These results are further optimized by using keyword-based and optional reranking before they are returned to the API.
  11. Does CAS support Retrieval-Augmented Generation (RAG) workflows?
    Yes, CAS search results can be combined with user prompts and sent to an LLM to generate responses, enabling RAG-based client applications.
  12. What types of data does CAS support for semantic extraction?
    CAS focuses on unstructured text data. It uses natural language processing (NLP) and AI technologies to extract semantic meaning from documents.
  13. Which IBM Fusion version is required to install CAS?
    CAS requires the newest IBM Fusion version. Installation must be performed by using the default ibm-spectrum-fusion-ns namespace. Custom namespaces are not supported for CAS installation.
  14. What are the hardware requirements for CAS?
    To run CAS on three OpenShift® compute nodes, the following resources are required:
    • 3 × NVIDIA GPUs
    • One other GPU for optional reranker service
    • 166 vCPUs (83 physical cores)
    • 1024 GB memory
    • 1 TB IBM Storage Scale
  15. What is the recommended IBM Storage Scale configuration?
    The most common configuration includes:
    • 48 NVMe drives
    • 30 TB capacity (Enterprise Storage Server 6000)
    • 4 CX-7 network adapters
    • 1.5 TB memory
  16. What software versions are required for IBM Storage Scale?
    Ensure that IBM Storage Scale version 5.2.3.1 or later is installed. Remote file systems must be configured and available for CAS.
  17. How do you verify the availability of a storage class for CAS?
    In the OpenShift console, go to Storage > Storage Class and confirm that the storage class (for example, ibm-spectrum-fusion) created from the IBM Storage Scale remote file system is available and marked as default.
  18. What are the prerequisites for enabling GPU workloads in OpenShift?
    Install the following components:
    • Red Hat Node Feature Discovery Operator
    • NVIDIA GPU Operator (compatible version)
  19. What is NVIDIA NIM and why is it required?
    NVIDIA NIM is a set of optimized microservices for deploying generative AI models. CAS uses NVIDIA NeMo Retriever for document processing. Provide NVIDIA NGC license keys to pull components from the NVIDIA NGC registry.
  20. Is the NVIDIA text reranker mandatory?
    No, the NVIDIA text reranker is optional and requires an additional supported GPU. It enhances search result accuracy by using advanced scoring.
  21. What type of proxy URLs must be allowlisted for CAS installation?
    Add the following URLs to your proxy allowlist:

    • apigee.googleapis.com

    • apigeeconnect.googleapis.com

    • binaryauthorization.googleapis.com (optional)

    • gcr.io

    • raw.githubusercontent.com

  22. How do you configure CAS to use the NVIDIA reranker service?
    Add the following flags to the cas-config ConfigMap:

    • NVMM_NEMO_RANKER

    • NVMM_NEMO_RANKER_SERVICE

  23. What user permissions are required for remote file system setup?
    The Global Data Platform user must be part of the StorageAdmin group to enable watch creation. Membership in CsiAdmin or ContainerOperator alone is insufficient.
  24. How do you initiate the installation of CAS?
    You can start the installation from the IBM Fusion user interface by navigating to the Services page and selecting the Content-Aware Storage tile.
  25. What components are installed with CAS?
    The CAS installation includes Kafka, the CAS operator, and CNPG (Cloud Native PostgreSQL).
  26. What prerequisites must be met before installing CAS?
    Ensure the following items:

    • All prerequisites that are listed in the Planning and prerequisites section are fulfilled.

    • A default storage class (for example, ibm-spectrum-fusion) is available and marked as default.

    • All IBM Storage Scale projects and pods are running.

  27. What can you do if the installation pre-check raises a warning?
    Resolve the warning at the OpenShift level before proceeding. The pre-check ensures that the environment is ready for CAS deployment.
  28. How can you monitor the CAS installation progress?
    After initiating installation, a notification appears on the Services page. You can track progress under Services > Installed. Once completed, the CAS service status shows as Healthy.
  29. How do you verify that CAS is installed successfully?
    You can verify installation from:
    • IBM Fusion UI: Check service version and health status. Use the ellipsis menu to download logs and view documentation.
    • OpenShift Console: Go to Home > Projects and search for a namespace that contains ibm-cas. Then check Operators > Installed Operators for CAS availability and service version.
  30. How do you configure CAS after installation?
    Set up a Scale user to enable watch creation, configure domains and data sources, and map domains to data sources. For more information, see Configuring Content-Aware Storage (CAS).
  31. Can you download installation logs?
    Yes. From the ellipsis menu in the CAS service window, you can download logs and view related documentation. A success notification confirms log collection.
  32. What happens if the CAS service status is not healthy after installation?
    Check the logs for errors, verify that all prerequisites are met, and ensure that all required services (Kafka, CNPG, CAS operator) are running. Revalidate the environment and retry installation if needed.