Re: Rationale: No New OS.

Eugene Leitl (
Sun, 22 Nov 1998 02:07:00 +0100

Eliezer S. Yudkowsky writes:
> If you need an OS to do something, rewrite Linux.
> And hasn't Beowolf done distributed supercomputing already?

Err, you weren't listening. I said there is no need to rewrite anything for the original rationale, and singnificantly beyond that bickering about quality of OSses becomes pointless. (And if only because in advanced architectures the distinction between code and data grows meaningless, and the art of programming becomes first a lot like chip design, then is reduced to providing a set of constraints and then goes extinct altogether, along with the bipedal primate).

> There are only two real OSes. MacOS is the supreme OS for users, and Linux is
> the supreme OS for programmers. Everything else is the 90% of everything that
> is crap.

I won't discuss MacOS, but with a GUI a la KDE (Next version of TrollTech's Qt will be GPL, and then there is Gnome) and a streamlined installation even a typical housewife might find Linux attractive. Beyond the marketing smoke and mirrors, even Microsoft is very aware of that (see Halloween2).

As to Linux, it is far from being optimal. How are you supposed to build an affordable parallel system if your kernel alone takes up a MByte? You want fastest memory possible, you must use local on-die memory with kBit wide bus and no parasitic capacitance/inductance. Why having caches then? But die yield says your CPU must be way simpler (no pipeline/MMU/FPU/BPU etc.), and your core grains must be tiny. Linux/Unix is not scalable to a system of few 100 networked CPUs with a few MBit of RAM each. If your core is very limited, and sending an (asynchronous active) message is almost as fast as acessing on-die core we are obviously concerned more with hierarchical decomposition of objects so that their fragments fit into memory grains than with page faults, swaps, process context switches, threads, libraries, modules, and other vermin. I guess I'll never understand it why people think that building computers or programming them are rational enterprises, and not ad hoc bricolage following market conservativism and vogues. What's so rational in a Hollerith machine and punching cards or pushing abacus beads?