Re: K42/HURDNG

[Andrew Baumann]

On Friday 11 August 2006 10:07, Guillaume FORTAINE wrote:
> Your knowledge in microkernel is unvaluable to us.

Hmm, the irony is great, but I'm going to assume you meant "invaluable".

> I would greatly appreciate that you help us  :
>
> http://lists.gnu.org/archive/html/l4-hurd/
>
> An interesting link :
>
> http://lists.gnu.org/archive/html/l4-hurd/2006-08/msg00017.html
>
> The choice for the language : Does Mercury seem to be a good candidate ?

I'm not sure why you're asking me about languages, since your subject seems to 
imply that you were interested in K42. My knowledge about language research 
is only that of a casual observer, but I can tell you some more about K42...

K42 as a project is working on building a Linux API/ABI-compatible OS. Some of 
the main aims of the group have been multiprocessor scalability and 
customisability. I'm not part of the K42 group at IBM, so this is just my 
view; they might disagree...

K42 is also built around a microkernel-style architecture -- it uses 
message-passing IPC and implements a lot of a lot of "traditional" kernel 
services like the file system in user-level servers. However, it's also not a 
microkernel in the L4 sense, because it uses Linux's device drivers and 
network stack, and runs them inside the kernel. It also puts more memory 
management policy in the kernel.

The underlying microkernel is pretty closely linked to the whole K42 source 
base. I don't know if HURD on top of K42 makes any sense, because I don't 
know enough about HURD. If it's sufficiently Unixish it might. The best 
approach in that case would probably be to take the whole K42 source base, 
and then start modifying the user-level parts and servers to match what you 
need. This would inevitably imply changes to the kernel -- K42 is not really 
an "off-the-shelf" microkernel that you can start writing alternative systems 
on.

Something else you said in that linked message demanded a response:
> As the L4 kernel has no more leader, and the community doesn't seem to be 
> very reactive, the only suitable alternative is the coyotos project. 

This sounds like you are implying that L4 is dead, which definitely isn't 
true. There are three research groups (here at UNSW/NICTA, UKa and TUD in 
Germany) actively doing stuff with L4, and I know that some people working on 
L4 systems are on this list.

Cheers,
Andrew