Hardware and software requirements

AI Framework for IBM® z/OS® requires the following hardware and software.

Tip: You can find additional practical guidance about these hardware and software requirements from IBM performance analysts in the white paper, IBM z/OS 3.1 AI-Powered WLM Batch Initiator Management Resource and Tuning Guidelines.

Hardware requirements

The z/OS AI Framework requires the following hardware:
  • Any of the following IBM Z servers:
    • IBM z17 — all models, including ME1 (machine type 9175)
    • IBM z16 — all models, including A01 (machine type 3931) and A02 (machine type 3932)
    • IBM z15 — all models, including T01 (machine type 8561) and T02 (machine type 8562)
  • Coupling facility (internal or external; required by EzNoSQL/VSAM record level sharing (RLS))
    • You will need a coupling facility cache structure of the appropriate size:
      • If you specify DIRONLY for the SMS DATACLAS RLS CF cache value (not caching data in the cache structure), a smaller structure size should suffice.
      • If you do not specify DIRONLY (caching data in the cache structure), a larger structure size will be needed. Use the Coupling Facility Structure Sizer (CFSizer) tool for guidance on sizing the structure.
      Guideline: The cache size (and local buffer pool size) should be a minimum of 20% of the data size.
    • If possible, do not share the cache structure. The guideline is to have separate caches for different RLS exploiters. For instance, if you are using HSM, it should have its own cache and the EzNoSQL databases should have their own. (It is possible to allow both HSM and EzNoSQL to share a cache, but you will need to increase the structure size accordingly.)
    • Start of changeThe configuration workflows for the various AI use cases (discussed later)End of change will guide you in configuring EzNoSQL. However, if you are interested in learning more detailed information, see Record-level Sharing CF Caching in z/OS DFSMS Using Data Sets and EzNoSQL for z/OS documentation.
  • IBM Z® Common Data Provider (ZCDP) runs on each z/OS LPAR where SMF records are to be collected for the z/OS AI Framework and requires the following minimum system capacity on each LPAR:
    • Memory: 1 GB
    • DASD: 8 GB
    • zIIP: ZCDP can offload work to IBM Z Integrated Information Processors (zIIPs) if one or more zIIPs are online before the ZCDP components are started.
  • Machine Learning for IBM z/OS Core Edition (MLz Core) runs on the z/OS LPAR where the z/OSMF server (AI Control Interface task) runs and requires the following minimum system capacity:
    • Processor: 1 shared zIIP or shared CP (2 suggested)
    • Memory: 31 GB
    • DASD: 10 GB (50 GB suggested)

    These capacities are in addition to those required for ZCDP running on the same LPAR.

Note: The preceding requirements are based on IBM internal testing and individual product documentation. Actual processor, memory, and DASD requirements will vary based on the number of participating z/OS LPARs, and the number of service classes per LPAR, the workload, and contention for shared resources. IBM strongly suggests that you monitor your system to ensure you have adequate memory to avoid paging, and adequate processor resource to avoid processor delays, both of which can adversely affect performance.

Software requirements

The z/OS AI Framework requires the following software:
  • IBM z/OS 3.1 or later (depending on use case)
    The z/OS AI Framework requires the following z/OS base elements:
    • Base Control Program (BCP), which includes the AI Base Component for z/OS (and workload management (WLM), which is a provider of an optional AI use case)
    • DFSMS, which includes EzNoSQL for z/OS  (Learn more about EzNoSQL for z/OS.)
      • EzNoSQL for z/OS uses VSAM record level sharing (RLS), which requires a coupling facility. Further, the VSAM server address space (SMSVSAM) must be initialized on each z/OS system in the sysplex that will be part of the z/OS AI Framework configuration.

        EzNoSQL requires the availability of one or more SMS storage classes (STORCLAS) which contain an RLS CACHESET with the name of a valid coupling facility cache structure. The STORCLAS names can be provided to the application architect or assigned by the SMS ACS routines.

      • Start of changeEach EzNoSQL database requires enough space to accommodate the minimum amount of training data needed by the specific AI use case (typically, 30–60 days of historical data). Refer to each use case's configuration workflow to determine the storage space requirements.End of change
      • Also see the coupling facility requirements related to the RLS cache structure earlier in Hardware requirements.
      • EzNoSQL must be available on each system where the z/OS AI Framework will be configured.
    • Start of changez/OS Communications Server, which provides TCP/IP networking and is a provider of an optional AI use caseEnd of change
    • z/OS Management Facility (z/OSMF), which includes the AI Control Interface for z/OS
      • The z/OSMF data directory must be mounted on a shared, global path, /global/zosmf, as described in Step 2: Run job IZUMKFS to create the z/OSMF user file system in IBM z/OS Management Facility Configuration Guide. This allows z/OS AI Framework components that are running on remote systems to access z/OSMF persistent data.
    • z/OS UNIX System Services
  • IBM Z Common Data Provider (ZCDP) 5.1
    • Start of changeFor AI use cases that depend on it for data collection, theEnd of change z/OS AI Framework requires its own running instance of ZCDP on each system where SMF records will be collected.
  • Machine Learning for IBM z/OS Core Edition (MLz Core) 3.2
    • MLz Core also requires IBM WebSphere Application Server for z/OS Liberty, version Start of change23.0.0.3End of change or later.
    • MLz Core includes the Spark runtime environment, which will be configured as part of the overall z/OS AI Framework configuration process (described later).
    • The home directory for MLz Core must have at least 10 GB of available space (50 GB suggested).
    • On z/OS 3.1, APAR OA67865 is required if you wish to use MLz Core 3.2.
  • A Java™ runtime, as follows:
    • MLz Core 3.2 and ZCDP 5.1 at PTF level UJ96789/UJ97146 or later require IBM Semeru Runtime® Certified Edition for z/OS, Version 17 (5655-UA1)

This is not an exhaustive list. The list covers the major software items that might require advance planning. Other software setup requirements, such as access and authorization to or for various components, are identified during the configuration phase using the Configure AI Framework for z/OS workflow. You can browse through the configuration workflow before you actually start performing the steps to learn what the configuration process entails. (See Configuring the z/OS AI Framework.)