System requirements

Decision Intelligence Client Managed Software containers are Red Hat® and IBM® certified. To use the Decision Intelligence Client Managed Software images, the administrator must make sure that the target cluster has the capacity.

For each stage in your operations, you must allocate a cluster of nodes before you install the Decision Intelligence Client Managed Software. Development, preproduction, and production are stages that are best run on different compute nodes. High-level resource objects are scoped within namespaces. Low-level resources, such as nodes and persistent volumes, are not in namespaces.

The Detailed system requirements External link opens a new window or tab page provides a cluster requirements guideline for Decision Intelligence Client Managed Software.

The minimum cluster configuration and physical resources that are needed to run the Decision Intelligence include the following elements:
  • Hardware architecture: Intel (amd64 or x86_64 the 64-bit edition for Linux® x86) on all platforms, Linux on IBM Z®, or Linux on Power®.
  • Node counts: Dual compute nodes for nonproduction and production clusters. A minimum of three nodes is needed for medium and large production environments and large test environments. Any cluster configuration needs to adapt to the size of the projects and the workload that is expected.
A cluster where you want to install Decision Intelligence needs as a minimum:
  • Master (3 nodes): 4 vCPU and 8 Gi memory on each node.
  • Worker (3 nodes): 4 vCPU and 10 Gi memory on each node.

Based on your cluster requirement, you can pick a deployment profile (deployment_profile_size) and enable it during installation. Decision Intelligence provides small, medium, large, and extra-large deployment profiles. You can set the profile during installation, in an update, and during an upgrade.

The default profile is small. Before you install Decision Intelligence, you can change the profile to a desired value. You can scale up or down a profile anytime after installation.

You can also adjust a profile size for Decision Designer or the decision runtime separately. For example, you can have a medium profile size for Decision Designer, and an extra-large profile size for the decision runtime to enable Horizontal Pod Autoscaler (HPA) only on the decision runtime.

Note:

The default profile size for the foundational services is small, the same as Decision Intelligence.

  • If you install Decision Intelligence with a medium profile, the foundational services profile size is initialized to medium.
  • If you install Decision Intelligence with a large profile, then the foundational services profile size is also large.

When you scale up or down the Decision Intelligence profile size afterward, the profile size for the foundational services remains unchanged. After installation, any change to the Decision Intelligence profile size is not reflected to the foundational services profile size. You must adapt it manually.

For more information about the foundational services, see the installing instructions in the GitHub repository: https://github.com/icp4a/automation-decision-services-kubernetes External link opens a new window or tab.

By default, the spec.zen.scale_config parameter takes the same value as the spec.deployment_profile_size parameter unless spec.zen.scale_config is explicitly set.

The following table describes each deployment profile.

Table 1. Deployment profiles and estimated workloads
Profile Description Scaling (per 8-hour day) Minimum number of worker nodes
Small (no HA) For environments that are used by a single department with a few users; useful for application development.
  • Processes 350,000 decisions
  • Supports failover
 3
Medium For environments that are used by a single department and by limited users.
  • Processes 700,000 decisions
  • Supports HA and failover
  • Provides at least two replicas of most services, if configuring failover
 3
Large For environments that are shared by multiple departments and users.
  • Processes 2,000,000 decisions
  • Supports HA and failover
  • Provides at least two replicas of most services, if configuring failover
 3

You can use custom resource templates to update the hardware requirements of the services that you want to install.

The following sections provide the default resources for Decision Intelligence.

Attention: The values in the hardware requirements tables were derived under specific operating and environment conditions. The information is accurate under the given conditions, but results that are obtained in your operating environments might vary significantly. Therefore, IBM cannot provide any representations, assurances, guarantees, or warranties regarding the performance of the profiles in your environment.

Operator default requirements

Table 2. Operator default requirements
Component CPU Request (m) CPU Limit (m) Memory Request (Mi) Memory Limit (Mi) Number of replicas Pods are licensed for production/nonproduction
ibm-ads-operator 10 500 64 512 1 No

Small profile hardware requirements

Table 3. Default requirements for a small profile
Component CPU Request (m) CPU Limit (m) Memory Request (Mi) Memory Limit (Mi) Number of replicas Pods are licensed for production/nonproduction Ephemeral storage Request Ephemeral storage Limit
ads-runtime 500 1000 2048 3072 1 Yes 600Mi 1700Mi
ads-credentials 250 1000 800 1536 1 No 100Mi 500Mi
ads-gitservice 500 1000 800 1536 1 No 400Mi 800Mi
ads-parsing 250 1000 800 1536 1 No 100Mi 800Mi
ads-restapi 500 1000 800 1536 1 No 100Mi 1300Mi
ads-run 500 1000 800 1536 1 No 100Mi 800Mi
Note:
Decision Intelligence also creates some jobs that request 200 m CPU and 256 Mi Memory. The following jobs are created at the beginning of the installation and do not last long:
  • ads-ltpa-creation
  • ads-ads-runtime-zen-translation-job
  • ads-designer-zen-translation-job
The following two cron jobs are running regularly for a short period of time (less than 2 minutes):
  • xx-ads-decision-runtime-cronjob
  • xx-ads-usage-metrics-reporter

Medium profile hardware requirements

Table 4. Default requirements for a medium profile
Component CPU Request (m) CPU Limit (m) Memory Request (Mi) Memory Limit (Mi) Number of replicas Pods are licensed for production/nonproduction Ephemeral storage Request Ephemeral storage Limit
ads-runtime 500 1000 2048 3072 2 Yes 1124Mi 1824Mi
ads-credentials 250 1000 800 1536 2 No 100Mi 500Mi
ads-gitservice 500 1000 800 1536 2 No 400Mi 800Mi
ads-parsing 250 1000 800 1536 2 No 100Mi 800Mi
ads-restapi 500 1000 800 1536 2 No 100Mi 1.3Gi
ads-run 500 1000 800 1536 2 No 100Mi 800Mi
Note:
Decision Intelligence also creates some jobs that request 200m CPU and 256Mi Memory. The following jobs are created at the beginning of the installation and do not last long:
  • ads-ltpa-creation
  • ads-ads-runtime-zen-translation-job
  • ads-designer-zen-translation-job
The following two cron jobs are running regularly for a short period of time (less than 2 minutes):
  • xx-ads-decision-runtime-cronjob
  • xx-ads-usage-metrics-reporter

Large profile hardware requirements

Table 5. Default requirements for a large profile
Component CPU Request (m) CPU Limit (m) Memory Request (Mi) Memory Limit (Mi) Number of replicas Pods are licensed for production/nonproduction Ephemeral storage Request Ephemeral storage Limit
ads-runtime 1000 2000 2048 3072 2 Yes 1274Mi 1974Mi
ads-credentials 250 2000 800 1536 2 No 250Mi 650Mi
ads-gitservice 500 2000 800 1536 2 No 550Mi 950Mi
ads-parsing 250 2000 800 1536 2 No 250Mi 950Mi
ads-restapi 500 2000 800 1536 2 No 250Mi 1450Mi
ads-run 500 2000 800 1536 2 No 250Mi 1250Mi
Note:
Decision Intelligence also creates some jobs that request 200 m CPU and 256 Mi Memory. The following jobs are created at the beginning of the installation and do not last long:
  • ads-ltpa-creation
  • ads-ads-runtime-zen-translation-job
  • ads-designer-zen-translation-job
The following two cron jobs are running regularly for a short period of time (less than 2 minutes):
  • xx-ads-decision-runtime-cronjob
  • xx-ads-usage-metrics-reporter

Extra large profile hardware requirements

The extra large profile can be set on Decision Designer and the decision runtime.

The extra large profile is not extending the CPU and memory limits of the large profile. Rather, it enables a HorizontalPodAutoscaler:
  • On the Decision Designer component, you can scale up to 5 pods of the parsing and run services.
  • On the decision runtime component, you can scale up to 5 pods of the runtime service.
You can change the maximum number of replica counts for each pod. For more information, see Configuration parameters.

You must then provision enough worker nodes or cluster resources to provide enough resources for the HorizontalPodAutoscaler to increase required pod counts, depending on the pods activity.

Also, check the resources of the Large profile hardware requirements.