Tanenbaum-Torvalds Debate - Notes

This paper didn't come to any kind of conclusions so much as it was a good
discussion of the state and intentions of MINIX and Linux at the time. There
was a lot of misunderstanding about theory versus practice, especially about
the difference between Linux and MINIX. MINIX was intended as a simple code
base, easily understood by students. Linux on the other hand was intended both
to have more advanced features (purposefully excluded from MINIX) and to take
advantage of all of features that the Intel 386 architecture offered
(sacrificing platform independence). AST pointed to a lot of the existing
literature[1] proclaiming the technical and theoretical superiority of
microkernel design (after posters established that there were indeed successful
microkernel deployments in the wild[2]) but it was eventually realized that
Linux's monolithic design allowed it to be developed and delivered quicker.
This availability was considered a major win and why Linux steadily gained
popularity over the years. This was the main reason why many hackers of the
time were choosing Linux over MINIX, BSD, GNU, or other less free or less
available Unix-like systems. There were many references to the technical
deficiencies of MINIX without depth of discussion. A list of the features that
MINIX didn't have is: 386 support, virtual terminals, soft links, select()
syscall, PTY's, demand paging, swapping, shared text, shared libs, efficient
MM, flexible MM, X Windows, TCP/IP, or POSIX. Admittedly the version of MINIX
written at the time was started before POSIX was created and the rest of the
features would have needlessly complicated MINIX's codebase.
	Posters cited Linux's availability, quick evolution and performance as
much better than MINIX's. There was a lot of talk about MINIX's poor
multithreaded filesystem performance, making it perform poorly for mutlitasking
workloads such as compiling programs while reading other files, reading data
from a floppy drive and reading email at the same time, or using emacs. Most of
the (perceived) problems with MINIX lead to the development of Linux and the
growth of its community. Pragmatism won over theoretical idealism. The
pragmatism was both technical (the monolithic design, sacrificing platform
independence for more features) and social (accepting patches from others,
freely licensing the code) and can still be seen in the Linux Kernel community
today. This all shows how pragmatism is a strong factor in OS choice.

What's amusing is some of the incorrect predictions made by posters:
 - x86 will fade as the dominant architecture (x86 is still a large backwards-
   compatible hack that isn't even CISC internally, but it is the general
   performance king -- more pragmatism)
 - microkernels are the wave of the future (hard to say: WinNT, Mac OS, and
   QNX are arguably the only mainstream consumer OSes that are microkernels,
   but they aren't particularly loud about it, Linux is still very loudly
   monolithic)
 - GNU kernel will be released soon (we're still waiting for it, even today)

---
1 - A Comparison of Two Computer Systems: Amoeba and Sprite,
     1991, Douglis, Kaashoek, Osterhout, Tanenbaum
     ftp://ftp.cs.vu.nl/pub/papers/amoeba/cs91.ps.Z
    The Filesystem Belongs in the Kernel
     1991, Welch
     http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.51.3730
    Unix as an Application Program
     1990, Rashid, Golub, Dean, Forin
     http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.52.633
2 - AmigaOS, QNX, Chorus, Mach 3.0, Amoeba, GNU, Plan 9 (not really)