Hacker News Reader

Comment by blacklion | Hacker News Reader

IoT and robotics should (dare I say "must"?) not use general-purpose OSes at all.

This «Linux have a finger in every pie» attitude is very harmful for industry, IMHO.

General purpose operating systems are fine and in some cases, preferable. However, they should be small, simple and designed with first class portability. Linux is none of those.

Why shouldn't they use the kernel, systemd, and a few core utilities? Why reinvent the wheel? There's nothing requiring them to pull in a typical desktop userspace.

Because different tasks requires different trade-offs and Linux has only one set of trade-offs. You cannot do good universal tool. It is like Leatherman, good enough to fix-up your bike on the side of the road, not so for normal workshop.

You say: reinvent the wheel.

I say: use pickup truck for every task, from farming to racing to commuting moving goods across continent. Is it possible? Of course. Is it good idea? I don't think so.

All cars are the same if you squint enough, wheels, engine, some frame, some controls, which are not very different between even F1 car and 18-wheel truck.

I agree but it's difficult to argue against it. There is just so much you get for free by starting with a Linux distro as your base. Developing against alternatives is very expensive and developing something new is even more expensive. The best we can hope for is that someone with deep pockets invests in good alternatives that everyone can benefit from.

How are you defining "general-purpose OS"? Are you saying IoT and robotics shouldn't use a Linux kernel at all? Or just not your general purpose distros? I would be interested to hear more of your logic here, since it seems like using the same FOSS operating system across various uses provides a lot of value to everyone.

I think, that I want at least hard-real-time OS in any computer which can move physical objects. Linux kernel cannot be it: hard RTOS cannot have virtual memory (mapping walks is unpredictable in case of TLB miss) and many other mechanisms which are desired in desktop/server OS are ill-suited for RTOS. Scheduler must be tuned differently, I/O must be done differently. It is not only «this process have RT priority, don't preempt it», it is design of whole kernel.

Better, this OS must be verified (as seL4). But I understand, that it is pipe dream. Heck, even RTOS is pipe dream.

About IoT: this word means nothing. Is connected TV IoT? I have no problems with Linux inside it. My lightbulb which can be turned on and off via ZigBee? Why do I need Linux here? My battery-powered weather station (because I cannot put 220v wiring in backyard)? Better no, I need as-low-power-as-possible solution.

To be honest, O think even using one kernel for different servers is technically wrong, because RDBMS, file server and computational node needs very different priories in kernel tuning too. I prefer network stack of FreeBSD, file server capabilities (native ZFS & Ko) of Solaris, transaction processing of Tandem/HPE NonStop OS and Wayland/GPU/Desktop support of Linux. But everything bar Linux is effectively dead. And Linux is only «good enough» in everything, mediocre.

I understand value of unification, but as engineer I'm sad.

Reply on news.ycombinator.com