root@sun:~# uname -a # SunOS 5.10 Generic_147440-01 sun4u sparc
Solaris, from 2.x to Oracle 11
What Linux Stole, and What’s Still Running
I have run Solaris since 2.x — SunOS 5.x on SPARC, when a workstation was a pizza box on your desk and a server was a Sun Fire that needed its own power circuit. Most of the people who write about Solaris in 2026 never ran it in production. I did, for decades, at banks and at a pension fund and anywhere else that needed a Unix that did not fall over. This is not a eulogy. Solaris is still running, still supported on 11.4, and — more usefully — most of what made it good is now inside the Linux you run every day, wearing different names. If you have ever wondered where ZFS, DTrace, zones, and the service model behind systemd came from, the answer is Sun, and the story is worth telling properly.
# the timeline, as lived — 2.x to 11.4
Solaris 2.x (1992–2000s). SunOS 5.x, System V Release 4 under the hood, on SPARC hardware that was genuinely better engineered than the x86 of the day. The work was VAX and Alpha shops moving to Sun, Novell NetWare giving way to Unix, and Solaris becoming the default serious Unix in finance and telecom. The machines were loud, hot, and reliable in a way that made you forgive both.
Solaris 8 and 9 (2000–2002). The mature, boring, dependable releases — the ones that ran the core banking and trading systems and were left alone for years because they worked. Trusted Extensions, RBAC, and the first real hardening model most of us ever used in anger.
Solaris 10 (January 2005). The release that matters most, because everything that survived the company was in it. ZFS, DTrace, Solaris Containers (zones), Service Management Facility, Fault Management Architecture, predictive self-healing, the lot — shipped in one release, on SPARC and x86, largely free. It was the most innovative operating system release of its decade, and Sun gave most of it away.
Oracle buys Sun (2010). The source stops being published in real time, OpenSolaris dies, and the community carries the open code forward as illumos — which is where OpenIndiana, OmniOS, and SmartOS come from. Solaris itself becomes a closed Oracle product again.
Solaris 11 (November 2011) and 11.4 (August 2018). Image Packaging System, network virtualization, kernel zones, and then continuous delivery — no 11.5, ever; features and fixes arrive as Support Repository Updates. 11.4 is still the current release in 2026, still receiving SRUs.
Solaris 8 and 9 (2000–2002). The mature, boring, dependable releases — the ones that ran the core banking and trading systems and were left alone for years because they worked. Trusted Extensions, RBAC, and the first real hardening model most of us ever used in anger.
Solaris 10 (January 2005). The release that matters most, because everything that survived the company was in it. ZFS, DTrace, Solaris Containers (zones), Service Management Facility, Fault Management Architecture, predictive self-healing, the lot — shipped in one release, on SPARC and x86, largely free. It was the most innovative operating system release of its decade, and Sun gave most of it away.
Oracle buys Sun (2010). The source stops being published in real time, OpenSolaris dies, and the community carries the open code forward as illumos — which is where OpenIndiana, OmniOS, and SmartOS come from. Solaris itself becomes a closed Oracle product again.
Solaris 11 (November 2011) and 11.4 (August 2018). Image Packaging System, network virtualization, kernel zones, and then continuous delivery — no 11.5, ever; features and fixes arrive as Support Repository Updates. 11.4 is still the current release in 2026, still receiving SRUs.
# what Linux stole — the ideas, wearing different names
None of this is a claim that Linux copied Solaris line for line. It is a claim about lineage: the problems Solaris solved first, and the shapes Linux’s solutions took afterwards. If you ran both, you can see the family resemblance, and knowing it makes you better at the Linux versions — because you understand what problem the feature was built to solve.
ZFS → OpenZFS, and the copy-on-write idea everywhere. ZFS (Solaris 10, 2005) was the first filesystem that treated storage as a pooled, checksummed, snapshotted whole instead of a stack of partitions and volumes. End-to-end checksums, snapshots, clones, compression, RAID-Z. Linux never got ZFS in-tree, for licensing reasons, but OpenZFS runs it on Linux today, and the ideas — pooled storage, snapshots as a first-class operation, checksums on everything — are now the default expectation. If you use snapshots on your Linux box, you are using a Solaris idea.
DTrace → eBPF and bpftrace. DTrace (Solaris 10) let you ask a running production system any question — which process is waiting on what, where the latency actually is — safely, with no reboot and no restart, in a language built for the purpose. Linux’s answer took a decade to arrive and came as eBPF, with bpftrace as its DTrace-shaped front end. The tools differ; the idea is the same, and the DTrace one-liner habit — ask the kernel, don’t guess — is still the right habit.
Zones → containers. Solaris Containers (Solaris 10) gave you isolated, lightweight, bootable environments sharing one kernel, with resource controls and their own network stack, years before Docker existed. Kernel zones in Solaris 11 took it further. Linux containers — LXC, then Docker, then Kubernetes — are the same idea at a different scale, with a different packaging story. Anyone who ran zones in 2006 understood containers in 2014 on day one.
SMF → the service model systemd inherited. The Service Management Facility replaced the init scripts with a service model: dependencies, restarts, fault management, and a single place to ask what a service is doing. systemd is not SMF — it is a different implementation, built for a different ecosystem — but the shape of the idea, services as managed objects with states and dependencies instead of scripts you hope ran in order, is SMF’s shape, and Solaris admins recognised it immediately.
And the rest, quietly: predictive self-healing and FMA (the box noticing its own faults before you do), RBAC and least-privilege as a designed-in model rather than a bolt-on, and the packaging discipline that IPS brought to Solaris 11. None of these are nostalgia. They are the features you now expect from a serious operating system, and they were Solaris features first.
ZFS → OpenZFS, and the copy-on-write idea everywhere. ZFS (Solaris 10, 2005) was the first filesystem that treated storage as a pooled, checksummed, snapshotted whole instead of a stack of partitions and volumes. End-to-end checksums, snapshots, clones, compression, RAID-Z. Linux never got ZFS in-tree, for licensing reasons, but OpenZFS runs it on Linux today, and the ideas — pooled storage, snapshots as a first-class operation, checksums on everything — are now the default expectation. If you use snapshots on your Linux box, you are using a Solaris idea.
DTrace → eBPF and bpftrace. DTrace (Solaris 10) let you ask a running production system any question — which process is waiting on what, where the latency actually is — safely, with no reboot and no restart, in a language built for the purpose. Linux’s answer took a decade to arrive and came as eBPF, with bpftrace as its DTrace-shaped front end. The tools differ; the idea is the same, and the DTrace one-liner habit — ask the kernel, don’t guess — is still the right habit.
Zones → containers. Solaris Containers (Solaris 10) gave you isolated, lightweight, bootable environments sharing one kernel, with resource controls and their own network stack, years before Docker existed. Kernel zones in Solaris 11 took it further. Linux containers — LXC, then Docker, then Kubernetes — are the same idea at a different scale, with a different packaging story. Anyone who ran zones in 2006 understood containers in 2014 on day one.
SMF → the service model systemd inherited. The Service Management Facility replaced the init scripts with a service model: dependencies, restarts, fault management, and a single place to ask what a service is doing. systemd is not SMF — it is a different implementation, built for a different ecosystem — but the shape of the idea, services as managed objects with states and dependencies instead of scripts you hope ran in order, is SMF’s shape, and Solaris admins recognised it immediately.
And the rest, quietly: predictive self-healing and FMA (the box noticing its own faults before you do), RBAC and least-privilege as a designed-in model rather than a bolt-on, and the packaging discipline that IPS brought to Solaris 11. None of these are nostalgia. They are the features you now expect from a serious operating system, and they were Solaris features first.
Knowing the lineage is not trivia. It is how you know what a Linux feature is for, what its limits are, and what its author was trying to fix — because you have seen the earlier version of the same answer.
# what’s still running — and the January 2027 date
Solaris is not dead, and the “dead” claim is the most common wrong thing said about it. Oracle Solaris 11.4 is the current release, still receiving Support Repository Updates, with Premier Support through at least November 2031 and Extended Support through at least November 2037 under Oracle’s Lifetime Support Policy. There is no 11.5 planned — continuous delivery is the model — but that is a release-model choice, not an abandonment.
The date that matters is January 2027. That is when Extended Support ends for Solaris 10 and for Solaris 11.3 — Oracle postponed it to January 2027 in October 2023, and its current policy documents still show it. After that, those releases move to Sustaining Support: indefinite, but no new fixes are written. No new security patches, ever. If you still have Solaris 10 or 11.3 in production — and plenty of banks, telcos, and government shops do — January 2027 is the deadline your migration plan is measured against, and it is a few months away.
Where it still runs, and why: core banking and trading systems, telecom billing, government registries, and anywhere a SPARC application was written against Solaris libraries and never ported. The reasons are unglamorous and real: the application works, the hardware is paid for, the risk of moving it is higher than the risk of keeping it, and the people who understand it are still in the building. That is not a failure. That is how infrastructure actually works.
The date that matters is January 2027. That is when Extended Support ends for Solaris 10 and for Solaris 11.3 — Oracle postponed it to January 2027 in October 2023, and its current policy documents still show it. After that, those releases move to Sustaining Support: indefinite, but no new fixes are written. No new security patches, ever. If you still have Solaris 10 or 11.3 in production — and plenty of banks, telcos, and government shops do — January 2027 is the deadline your migration plan is measured against, and it is a few months away.
Where it still runs, and why: core banking and trading systems, telecom billing, government registries, and anywhere a SPARC application was written against Solaris libraries and never ported. The reasons are unglamorous and real: the application works, the hardware is paid for, the risk of moving it is higher than the risk of keeping it, and the people who understand it are still in the building. That is not a failure. That is how infrastructure actually works.
# migrating off it — what moves cleanly, what doesn’t
What moves cleanly to AlmaLinux / RHEL: shell scripts and cron jobs (test them — Solaris’s default shell and tools have GNU differences, but most scripts move with small edits), NFS and standard network services, Java applications (the JVM is the JVM), and anything that was already written portably. The discipline that moves best is the Solaris discipline itself: zones map to containers, SMF manifests map to systemd units with a little thought, and the ZFS habit maps directly to OpenZFS.
What does not move cleanly: anything compiled for SPARC — SPARC binaries do not run on x86, full stop, and need a recompile or a replacement; Solaris-specific system calls and libraries; IPS packages, which have no Linux equivalent and need rebuilding as RPMs; and ZFS pools, which need a version check and a tested import into OpenZFS before you commit — never assume a pool imports, always test it on a copy first.
The migration order that works: inventory first (what is actually running, and who owns it), then the supported path — Solaris 10 or 11.3 to 11.4 if the application can stay on Solaris, or to Linux if it cannot — then a pilot on the hardest application, not the easiest, because the easy ones were never the risk. And the deadline is real: a migration that is still in planning in January 2027 is a migration that is running on an unsupported operating system.
The open-source alternative: illumos — OpenIndiana, OmniOS, Tribblix — keeps the Solaris lineage alive outside Oracle, with ZFS, zones, and DTrace intact. It is a real option for the workloads that need Solaris behaviour without an Oracle contract, and it is worth evaluating honestly against the Linux path rather than dismissing it as nostalgia.
What does not move cleanly: anything compiled for SPARC — SPARC binaries do not run on x86, full stop, and need a recompile or a replacement; Solaris-specific system calls and libraries; IPS packages, which have no Linux equivalent and need rebuilding as RPMs; and ZFS pools, which need a version check and a tested import into OpenZFS before you commit — never assume a pool imports, always test it on a copy first.
The migration order that works: inventory first (what is actually running, and who owns it), then the supported path — Solaris 10 or 11.3 to 11.4 if the application can stay on Solaris, or to Linux if it cannot — then a pilot on the hardest application, not the easiest, because the easy ones were never the risk. And the deadline is real: a migration that is still in planning in January 2027 is a migration that is running on an unsupported operating system.
The open-source alternative: illumos — OpenIndiana, OmniOS, Tribblix — keeps the Solaris lineage alive outside Oracle, with ZFS, zones, and DTrace intact. It is a real option for the workloads that need Solaris behaviour without an Oracle contract, and it is worth evaluating honestly against the Linux path rather than dismissing it as nostalgia.
# the hardware, briefly — because the hardware was half the story
Pizza-box SPARCstations on desks, Sun Fire servers in the machine room, zones and domains on M4000s — the hardware was engineered like the operating system was: built to run for years, with service processors, redundant everything, and a sense that a server was a piece of infrastructure, not a commodity. That era is over, and good riddance to the prices — but the engineering habit it taught, of expecting a system to be diagnosable, serviceable, and honest about its own faults, is the habit that makes a good Linux admin. The tools changed. The standard did not.
Support dates verified against Oracle’s Lifetime Support Policy and Oracle’s October 2023 announcement, November 7, 2026: Solaris 10 and 11.3 Extended Support ends January 2027; Solaris 11.4 Premier Support through at least November 2031, Extended through at least November 2037, with dates reviewed annually. Always re-check the current Oracle policy before committing to a plan. The history and the migration judgements are the author’s own, from running Solaris in production since 2.x.