OSDev.org

iansjack wrote:Brendan never said that Linux was designed with that purpose in view. I don't think that he was making any points about Linux per se but was commenting on the rather naive suggestion that one operating system can cover all uses. I agree with him that different applications can call for different operating systems.

He had quoted from a discussion about Linux. His point was that we cannot use the same system for different applications. My point is that having a universal system like Linux is not the same as "using the same operating system". Two Linux systems with different applications might well have qutie different kernels (as the kernels, remember, are optimised for their specific applications) but as they are only variants of the same kernel they are still compatible with each other and software written for one can (except for differences in hardware archietechture) run on the other.

Well, I'm sure that Brendan knew what he meant. And I still agree with my interpretation of that, that there is no "one size fits all" operating system.

There IS a one-size-fits-all operating system. It's Linux. But it's not "one-size-fits-all" in the sense of trying to make do with a single operating system in different applications; it's more a case of that it can be customised to fit whatever needs a particular application has. Why nobody can understand that I don't know. You honestly mean to tell me that smartphone manufacturers are so dumb to run a server operating system on their devices???

onlyonemac wrote:Why nobody can understand that I don't know.

Possibly it's because some people here have experience with a wider range of hardware than others. Have you worked with large database applications on IBM iSeries or zSeries, for example? What about real-time applications; are you familiar with them?

iansjack wrote:What about real-time applications; are you familiar with them?

No, but you can compile Linux for increased real-time support.

The fact that you can compile the Linux kernel for increased real-time support doesn't mean that you end up with an OS as well suited to the task as one designed from the outset as a RTOS. There are other factors to consider than the scheduling algorithms.

You, very sensibly, don't address the question of which operating system might be best suited for a particular application on a particular mainframe computer. They do tend to be a little different to personal computers.

Linux is a great all-round operating system, particularly on personal computers; that doesn't mean that it is best for all tasks on all types of computer. It would be very surprising if anyone could design an OS that was. And, let's face it, that would be a recipe for stagnation.

Increased real-time support still implies that Linux is still not hard real-time (and for that matter, it's very unlikely it will ever be), and thus will not be allowed for things like medical equipment or spacecraft controls.

Combuster wrote:spacecraft controls

Linux was used in LightSail and guess what?.. It hanged up in space because of the memory being filled by a log file. Hopefully, it was resolved by a random reboot (not by a watchdog, there was none at all!) some time after.

Rebooted by cosmic rays! Possibly not something that should be relied upon for a real-time OS controlling a nuclear power plant.

Roman wrote:

Combuster wrote:spacecraft controls

Linux was used in LightSail and guess what?.. It hanged up in space because of the memory being filled by a log file. Hopefully, it was resolved by a random reboot (not by a watchdog, there was none at all!) some time after.

That's not the fault of Linux itself. It's up to the user to configure the amount of information to log - and Linux (and much of the userspace) does have the functionality to make such a configuration.

I guess it could be argued that the OS most suitable for controlling a spacecraft (where the hardware is out of reach) wouldn't allow such misconfiguration. One of the reasons why it may be better to use an OS designed for the job rather than a general-purpose OS shoehorned into it.

Really, I find this whole discussion somewhat bizarre. What is someone who believes that there is only one OS needed, and that OS has already been written, doing on a website devoted to OS design? We should be welcoming diversity.

iansjack wrote:What is someone who believes that there is only one OS needed, and that OS has already been written, doing on a website devoted to OS design? We should be welcoming diversity.

So? It's fun to look at different designs and fiddle around with them and consider new concepts in operating systems, but it goes without saying that these "fun projects" are never going to beat Linux when it comes to use in the real world.

There are many platforms and applications for which the Linux kernel is ill-suited. Here's the first few that come to my mind:

• Any project whose deployment/maintenance scenarios will require a stable kernel ABI.
• 8 and 16-bit microcontrollers
• Projects without a need for process spaces or users.
• Projects incompatible with the GPL
• Processors with unusual architectures that would require significant porting efforts.
• Highly-reliable systems that require microkernel designs.
• Projects that require verified/provable code.
• Operating systems that strictly serve as execution environments for bytecode interpreters.
• Hard real-time applications with tight deadlines.

As for whether there's any "real" value in developing yet another hobby OS, it's worth remembering that's how Linux started.

Any project whose deployment/maintenance scenarios will require a stable kernel ABI.

That is the one and only reason ReactOS or windows is to be preferred. The rest has nothing to do with the subject.

In particular, these are nonsense claims when it comes to using or not using Linux:

Projects incompatible with the GPL
Processors with unusual architectures that would require significant porting efforts.
Operating systems that strictly serve as execution environments for bytecode interpreters.

Combuster wrote:That is the one and only reason ReactOS or windows is to be preferred. The rest has nothing to do with the subject.

ABI stability may be one of the most familiar drawbacks of Linux, but not the only one. Additionally, it's a stretch to say that "Linux is not well suited to 8 and 16-bit microcontrollers" has nothing to do with the subject of whether Linux is well suited to all applications.

In particular, these are nonsense claims when it comes to using or not using Linux:

Projects incompatible with the GPL
Processors with unusual architectures that would require significant porting efforts.
Operating systems that strictly serve as execution environments for bytecode interpreters.

I don't think it's nonsense..

• Projects incompatible with the GPL: I'm aware of several efforts where GPL code was not allowed to avoid having to meet GPL obligations.
• Processors with unusual architectures that would require significant porting efforts: custom soft processors in FPGAs, architectures without C compilers, move-triggered architectures, reversible computing architectures, etc.
• Operating systems that strictly serve as execution environments for bytecode interpreters: verified code platforms, Javacard implementations, SDR platforms, etc.