IBM Maximo on AWS
Customer-managed IBM Maximo Application Suite on Red Hat OpenShift Service on AWS (ROSA) or self-managed OpenShift on EC2, with EBS, EFS, and IBM-supported database choices
Overview
IBM Maximo Application Suite (MAS) runs on Red Hat OpenShift, so running it on AWS means choosing how that OpenShift cluster is built and who operates it. IBM documents installing MAS on Red Hat OpenShift Service on AWS (ROSA) starting with MAS 8.11, and the MAS CLI can provision a self-managed OpenShift cluster on EC2 with the installer-provisioned infrastructure (IPI) method. On either path you pick the storage classes, typically EBS gp3 for ReadWriteOnce volumes and Amazon EFS for ReadWriteMany, and the dependencies: Db2 in the cluster or an external database for Maximo Manage, and MongoDB in the cluster or Amazon DocumentDB for MAS Core and Suite License Service. This is different from IBM's MAS SaaS offering, which AWS describes as running on AWS and operated by IBM Site Reliability Engineering. We plan, install, and upgrade customer-managed MAS in your AWS account and help you decide whether SaaS is a better fit.
What teams weigh before running Maximo on AWS
IBM MAS SaaS on AWS is not the same as MAS on your AWS account
There are two ways to run MAS on AWS: IBM's multi-tenant MAS SaaS, which IBM operates on AWS, and customer-managed MAS on ROSA or self-managed OpenShift in your own AWS account. IBM now documents its single-tenant MAS Dedicated service as offered exclusively on IBM Cloud, although a 2023 AWS blog post described a Dedicated option on AWS. With SaaS, IBM SRE handles installation, upgrades, database management, infrastructure, and network, and customization is UI and API driven. With customer-managed MAS, you own the AWS account, the OpenShift cluster choices, the upgrade schedule, and the operational work.
Reference architecture
ROSA is jointly supported by AWS and Red Hat, and Red Hat SREs handle cluster installation, maintenance, and upgrades. With ROSA hosted control planes (HCP), the control plane runs in a Red Hat-owned AWS account and worker nodes run in yours. With ROSA classic, both run in your account. IBM's ROSA install topic supports only the existing-cluster option, so you create the ROSA cluster first. A self-managed cluster built with OpenShift IPI on EC2 gives full control but leaves maintenance and upgrades to you.
IBM's MAS Performance Wiki recommends M5 or M6 instances, for example m5.4xlarge, as worker nodes and 10 Gb networking for production. Size machine pools from the IBM capacity planning guidance for the applications you deploy, since Manage, Monitor, and Visual Inspection have very different footprints.
The MAS CLI recognizes AWS storage and uses gp3-csi for ReadWriteOnce and EFS for ReadWriteMany, and you can override both. EBS volumes are ReadWriteOnce only, with io1 and io2 available where higher IOPS are needed. Red Hat documents OpenShift Data Foundation in internal mode on ROSA HCP using gp3-csi as the backing storage class.
The MAS CLI can install Db2 in the cluster with the Db2u operator, either shared or dedicated to Manage. MongoDB runs in the cluster through the MongoDB Community operator by default, or externally. IBM's operator catalog certifies Amazon DocumentDB engine 5.0.0 as an alternative. IBM documents Amazon RDS for Db2 for Maximo Real Estate and Facilities 9.2 on ROSA.
The MAS CLI has built-in support for Amazon Route 53 DNS, Amazon MSK as the Kafka provider for IoT, and an AWS provider for Manage attachments. These replace in-cluster components with managed AWS services where IBM supports it.
Visual Inspection needs NVIDIA GPUs to train models, with at least 16 GB of GPU memory, and IBM certifies devices including T4, P40, P100, V100, A10, and A100. The MAS Performance Wiki recommends P3 or P4 instances for production GPU nodes and g4dn for dev and test only. The NVIDIA GPU Operator and Node Feature Discovery must be installed in the cluster.
Trade-offs
MAS 8.11 first release with a documented ROSA install. IBM documents installing MAS on Red Hat OpenShift on AWS starting with MAS 8.11
2 vs 7-9 minimum EC2 instances, ROSA HCP vs classic. ROSA HCP needs two EC2 instances to create a cluster, classic needs seven (single-AZ) or nine (multi-AZ)
16 GB minimum GPU memory for Visual Inspection training. Maximo Visual Inspection requires GPUs to train all model types
ROSA moves cluster lifecycle work to Red Hat SREs, while self-managed OpenShift on EC2 gives more control and more operational work. Teams often pick one before knowing what MAS needs from the cluster.
IBM's SaaS on AWS removes infrastructure work but limits customization to UI and API driven changes and integrations can be more limited. Customer-managed MAS keeps full control and full responsibility.
Several MAS applications need ReadWriteMany volumes, which EBS cannot provide. EFS in bursting mode can slow down once burst credits run out, so throughput mode needs planning.
Not every managed AWS database is supported for every MAS component. DocumentDB has functional differences from MongoDB and cannot be used when MAS runs in FIPS mode, and RDS for Db2 is documented for specific applications.
The AWS classic load balancer has a 60 second idle timeout by default, which can cut off long-running requests, and its surge queue has a fixed limit that matters for IoT traffic.
MAS operator catalogs certify specific OpenShift releases, and ROSA follows Red Hat's release schedule. Upgrades need to keep MAS, OpenShift, and dependencies on a supported combination.
What we do
We compare IBM SaaS, ROSA, and self-managed OpenShift on EC2 against your integration, customization, and isolation needs, then plan the VPC, subnets, DNS, and IAM the cluster needs.
AI on AWS
If you choose ROSA, IBM provides a script to create the VPC and subnets and an Ansible role to create EFS storage before you install MAS from the AWS Marketplace BYOL listing or the MAS CLI.
We build ROSA or IPI clusters with machine pools sized for your applications, configure gp3-csi and EFS storage classes, and add OpenShift Data Foundation where you want in-cluster storage.
AI on AWS
For EFS we set throughput mode and CloudWatch monitoring of the burst credit balance so storage does not become the bottleneck after go-live.
We install Db2 with the Db2u operator or connect Manage to your existing Db2, Oracle, or SQL Server database, and set up MongoDB in the cluster or on Amazon DocumentDB.
AI on AWS
For DocumentDB we set RetryWrites to false in the Suite License Service and Suite configuration, as the MAS Performance Wiki requires, and confirm FIPS is not needed.
We add GPU machine pools with IBM-certified NVIDIA devices, install the NVIDIA GPU Operator and Node Feature Discovery, and confirm GPUs are allocatable before installing Visual Inspection.
AI on AWS
Separate GPU pools let you scale training capacity without changing the nodes that run Manage.
We plan OpenShift and MAS upgrades against IBM's operator catalog compatibility, tune load balancer timeouts, and monitor Amazon MSK where IoT applications use it.
AI on AWS
On ROSA HCP the control plane and each machine pool can be upgraded separately, which lets upgrades be staged.
Walk through your cluster, databases, integrations, and AI plans with an engineer who installs and runs MAS on OpenShift.
The stack
Requirements
IBM's ROSA procedure: create the VPC and subnets, install ROSA into them, create EFS with the ocp_efs role, then deploy MAS from the AWS Marketplace BYOL listing on the existing cluster.
Automated provisioning of self-managed OpenShift clusters on AWS with installer-provisioned infrastructure, in interactive or non-interactive mode.
Storage class detection for AWS (gp3-csi and EFS), Db2 options for Manage, MongoDB providers, Route 53 DNS, and Amazon MSK for Kafka.
MongoDB can run in or outside the cluster and must support TLS. DocumentDB cannot be used when MAS runs in FIPS mode.
Worker and GPU instance types, EBS and EFS behavior, DocumentDB settings, MSK monitoring, and load balancer timeouts.
Deploying ODF in internal mode on existing ROSA HCP clusters using the gp3-csi storage class.
Yes. MAS runs on Red Hat OpenShift, and on AWS that can be ROSA or a self-managed OpenShift cluster on EC2. IBM documents installing MAS on ROSA starting with MAS 8.11, and IBM also sells MAS SaaS, which runs on AWS and is operated by IBM.
With MAS SaaS, IBM SRE handles installation, upgrades, database management, infrastructure, and network, and you customize through the UI and APIs. With customer-managed MAS, the cluster runs in your AWS account and you control the OpenShift setup, integrations, and upgrade schedule. IBM also offers a Dedicated option that is single-organization and IBM-managed.
ROSA is a managed OpenShift service where Red Hat SREs install, maintain, and upgrade the cluster, with joint support from AWS and Red Hat. Self-managed OpenShift, which the MAS CLI can provision with IPI, gives full control of the cluster but you maintain and upgrade it yourself. The MAS application layer is your responsibility in both cases.
IBM's ROSA install topic does not distinguish HCP from classic, and Red Hat's community guide for MAS on ROSA builds on an HCP cluster. Because HCP places ingress and monitoring on worker nodes, test the MAS installer version you plan to use on an HCP cluster before production.
MAS needs a ReadWriteOnce storage class and a ReadWriteMany storage class. On AWS the MAS CLI uses EBS gp3-csi for ReadWriteOnce and Amazon EFS for ReadWriteMany. OpenShift Data Foundation in internal mode is another option, and Red Hat documents it on ROSA HCP.
Yes. IBM's operator catalog certifies DocumentDB engine 5.0.0 as an alternative to MongoDB for MAS Core and Suite License Service. You must set RetryWrites to false, and DocumentDB cannot be used if MAS runs in FIPS mode.
IBM documents Amazon RDS for Db2 as a managed database option for Maximo Real Estate and Facilities 9.2 on ROSA, with Db2 licenses obtained separately. For Maximo Manage, check the current IBM compatibility report before planning on it. Manage supports Db2 in the cluster through the Db2u operator and external Db2, Oracle, or SQL Server databases.
Visual Inspection needs NVIDIA GPUs with at least 16 GB of memory to train models, and IBM certifies T4, P40, P100, V100, A10, and A100 devices. The MAS Performance Wiki recommends P3 or P4 instances for production GPU nodes and g4dn for dev and test only.
Yes. IBM lists a BYOL MAS offering that deploys onto an existing ROSA cluster through CloudFormation, client-managed listings, and SaaS editions billed through your AWS account. Which one fits depends on whether you want to operate MAS yourself.
The MAS Performance Wiki highlights the 60 second default idle timeout on the AWS classic load balancer, EFS burst credits in bursting mode, and CloudWatch monitoring of Amazon MSK brokers. Each can affect performance once real users and IoT traffic arrive.
Tell us where Maximo runs today and where you want it to run. We will tell you honestly whether AWS is the right fit.
