The push to save the Itanium architecture in Linux
Linux kernel developers are removing support for Intel’s defunct Itanium (IA-64) architecture, prompting debate over how long open‑source projects should carry legacy hardware. Commenters note that most remaining Itanium systems run proprietary OSes like HP‑UX or VMS, hobby interest is tiny, and no companies are funding maintenance, so relying on old kernels or other Unix-like systems is more realistic than keeping IA‑64 in mainline Linux. The thread becomes a broader reflection on the costs of backward compatibility, the value of architectural diversity, and the practical need for maintainers and test hardware.
Who Still Uses Itanium and Why?
- Many assume remaining Itanium demand is tied to HP-UX and VMS, not Linux.
- Examples: Oracle databases on HP hardware, VMS clusters, custom COBOL on HP-UX, past HPC clusters and supercomputers running Linux.
- Several comments suggest active production Linux-on-Itanium is now tiny (<~1000 systems, and likely shrinking fast).
- Industrial and aerospace lifecycles are long (e.g., m68k in Eurofighter avionics), but Itanium doesn’t seem to fill that niche in practice.
Retro / Hobbyist Interest
- Hobbyist interest in Itanium is described as very small: hardware is rare, expensive, loud, power-hungry, and often tied to SANs and datacenter infrastructure.
- Some people would like Itanium boxes “for fun,” but contrast it with m68k, Amiga, SPARC, etc., which have stronger communities and good emulators.
- Lack of an Itanium emulator (like QEMU support for m68k) is seen as a factor limiting community maintenance.
Why Drop IA-64 from Linux?
- Central argument: no active maintainers or testers, very few users, and significant complexity cost for the rest of the kernel.
- Maintaining exotic architectures can prevent refactoring and performance improvements, and requires special cases for weakly used features.
- Some view attempts to keep IA-64 as sunk-cost fallacy. Others argue exotic architectures help keep the kernel portable and catch bugs.
- Consensus trend: if nobody steps up to maintain and test, code gets removed; mainline isn’t an entitlement.
“Just Use an Old Kernel” vs. Long-Term Support
- Many note users can stay on older kernels (e.g., LTS 6.1) indefinitely; this is already common in industry.
- Concerns: no systematic backporting of fixes for dead architectures; no support for new drivers or infrastructure changes.
- Some worry that aggressive removal of old drivers/architectures (e.g., QLogic NICs) accelerates hardware obsolescence and undermines long-term stability.
Alternatives and Broader Context
- NetBSD’s ia64 port exists but appears inactive.
- SPARC is cited as a still-viable “exotic” architecture (cheap used hardware, ongoing Fujitsu systems) providing diversity.
- RISC-V is mentioned as a future-dominant ISA, though others dismiss the idea of “ending” ISA wars.
- Several comments situate Itanium’s failure in its overreliance on compilers, the end of Dennard scaling, and competition from x86-64.