AIX for a Linux admin: where Linux habits stop working

Tooling

• ▸ Package management: instead of zypper on SLES or yum on RHEL 7 there is installp and lslpp, working with filesets rather than packages. RPM exists on AIX, but only for open-source software alongside the system. The system level comes from oslevel, not from /etc/os-release.
• ▸ Configuration through SMIT: smit/smitty is a text menu that assembles commands for the administrator – and can show the command it will run (F6). YaST on SUSE is the closest counterpart; RHEL has none and the equivalent is editing files plus configuration management.
• ▸ ODM (Object Data Manager): AIX keeps device and system configuration in the binary ODM database, not in plain-text files under /etc. Editing a file by hand is therefore not enough – ODM is what knows the state.
• ▸ Errors and logs: instead of journalctl (systemd is standard on both SLES 12 and RHEL 7) there is errpt, and errpt -a for detail, over the system error log.
• ▸ Services: AIX has /etc/inittab and the System Resource Controller with startsrc/stopsrc/lssrc. Anyone used to a single systemctl looks for one tool where AIX has several layers.
• ▸ Boot: after changes to the boot chain you run bosboot, or the system may not boot.

LVM

AIX has its own LVM, older and different in many ways from LVM2 on Linux:

• ▸ the concepts are similar (PV, VG, LV), the commands are not (lspv, lsvg, lslv, mklv, extendvg);
• ▸ an LV typically carries a JFS2 file system directly – the role XFS plays as the RHEL 7 default, and Btrfs on the SLES 12 system volume;
• ▸ mounts are listed in /etc/filesystems, not in fstab;
• ▸ mirroring (mirrorvg) and the placement of copies across disks is part of LVM, not a separate layer;
• ▸ rootvg, the system VG, is managed with the same tools as data VGs.

LPARs: physical vs. virtual resources

Power machines are divided into logical partitions (LPARs) from the Hardware Management Console (HMC):

• ▸ a CPU can be dedicated or virtual (micro-partitioning from a shared pool) – capacity planning and performance measurement then work differently than on one physical CPU;
• ▸ memory and I/O are assigned to a partition from the HMC;
• ▸ I/O often does not pass through directly but through VIOS (Virtual I/O Server), which shares disks and networking between partitions – another layer Linux on a plain x86 machine does not have.

This layer is also the part that carries over: both SLES and RHEL run on Power in the ppc64le variant, in the same partitions and behind the same VIOS. Anyone who understands LPARs and the HMC assigns resources to a Linux partition exactly as to an AIX one – what changes is the operating system inside, not the virtualization model.

Enterprise storage

Disks usually do not arrive locally but from a SAN:

• ▸ attached through MPIO (multipath) – one disk is visible over several paths and the driver merges them into a single device, much like dm-multipath on SLES and RHEL;
• ▸ on AIX multipath is part of the base system, whereas on Linux multipathd and its multipath.conf are tuned for the particular array;
• ▸ losing one path must not take down access to the disk – that is a standard part of the design, not an exception.

Backups and deployment

• ▸ mksysb produces a bootable image of the whole system (rootvg) – recovery or cloning of a machine from a single artifact. On Linux the closest counterpart is ReaR (Relax-and-Recover), shipped by both RHEL and SLES; otherwise a data backup plus a fresh install from a template.
• ▸ NIM (Network Installation Management) installs and restores systems over the network – the role kickstart has on RHEL or AutoYaST on SLES, only with its own database and terminology.

High availability

Clustering and service takeover are handled by PowerHA (formerly HACMP). The concept is the same as Pacemaker with Corosync, which SUSE Linux Enterprise High Availability Extension and the Red Hat High Availability Add-On are built on. The configuration, terminology and tools differ – knowledge of one cluster carries into the other not as commands but as thinking about resource groups, dependencies and fencing.

What changed by 2024

The differences above still hold, but over those six years the Linux side moved more than AIX did:

• ▸ package management: RHEL 8 and 9 use dnf instead of yum, and SLES 15 split the system into modules with their own life cycles. installp and filesets stayed as they were, so the gap in tooling grew;
• ▸ containers: RHEL ships podman as a daemonless runtime, SLES offers it next to Docker. AIX has no container ecosystem and workloads are still split into LPARs – the unit of isolation is a partition, not a container;
• ▸ networking and firewalling: nftables on both distributions, the network managed by NetworkManager on RHEL and by wicked on SLES 15 (NetworkManager only from SLES 16) – against smitty and ODM on AIX;
• ▸ platform: POWER7 was followed by POWER8 to POWER10, AIX 7.1 by 7.2 (2015) and 7.3 (2021). HMC, LPARs and VIOS did not change in principle, so virtualization knowledge from the POWER7 days still holds;
• ▸ automation: Ansible covers both AIX and Linux, so part of the operations can be written once for both. The AIX modules are younger, though, and cover less than those for RHEL and SLES.

Summary

Moving from Linux to AIX is not about learning a new system from scratch but about unlearning details: where the configuration lives (ODM instead of /etc), how installation works (filesets and NIM), how LVM behaves and how partition performance is counted. Once those differences are clear, most operational habits carry over unchanged – and on Power one virtualization model serves AIX, SLES and RHEL alike.