Re: Tasks list for GNU Mach

[olafBuddenhagen]

Hi,

> please excuse my ignorance, but what's the Status of GNU Mach? I heard
> the L4 microkernel is favored for the Hurd. Do you guys try to catch
> up?

Well, the question is, favored by whom? It's actually quite a long
story, and also I weren't around to get all of it first hand; but I'll
try to sum up what I was able to gather.

A couple of years ago, Marcus (and some others), more and more exposed
to the problems in Mach, made the claim that Hurd/Mach has no future,
and further developement should be directed to a port to a modern
microkernel like L4. This wasn't really surprising, as today it's widely
known that Mach in it's existing form is severly flawed.

This, or maybe just unrelated loss of interest, made any work on GNUMach
practically stop for the last few years. So additionally to the design
flaws, it was also getting severly outdated.

After several years of design considerations (and waiting for L4
Pistachio), a mostly complete design was in place, and work on Hurd/L4
finally took off some 1 1/2 years ago. This spring it already looked
quite promising.

At this point, Marcus' rising doubts about the originally planned
capability system on top of L4 culiminated, and after learning about the
upcoming L4ng and L4.sec, he decided to rework the capability part for
these new L4 variants, while leaving the rest of the Hurd/L4 design
intact. However, further discussions with the L4 people and with
Jonathan Shapiro triggered more and more doubts, and a couple of months
ago, Marcus finally decided that the current half-implemented Hurd/L4
design is fundamentally flawed as well, and work on it should stop, in
favor of (another) complete redesign -- this time likely based on
Shapiro's Coyotos instead of L4, and most certainly based on some
EROS/Coyotos design ideas. (Whether this will still have *anything* to
do with the Hurd -- except for Marcus being the leader -- is arguable;
but that's another story...)

With all this confusion, and the future Hurd/whatever moving further and
further away -- or maybe just by a plain coincidence -- interest in
Hurd/Mach is on the rise again, and there is a remarkable revival in
GNUMach hacking.

I don't think anyone actually believes a hacked up GNUMach should try to
compete against L4. While it's certainly possible to turn Mach into
something L4-like by incremental improvment, some of the necessary
changes would be so fundamental, that it's likely easier to just move on
to L4 at some point. The idea is probably rather that there is still
*lots* of room for improvement with Hurd/Mach, and it's still a valuable
project to work on -- the one that is actually likely to give us a
useful and attractive Hurd in the forseeable future. Only once this room
has been fathomed, and the Hurd has been proven to be actually able to
deliver something useful not only visions, it's time to go for a further
improved architecture.

So the current situation seems to be that while the existing Hurd/L4
project is on ice probably for a long time, and Marcus' new
Hurd/whatever is far into the future (and might end up not really having
much to do with the Hurd), good old Hurd/Mach is on the rise again --
which is a good thing. At least that's how I see it.

-antrik-