Re: Hurd-related research help requested

[Tom Hart]

Alexander, I've been looking at GNU/Hurd as an undergrad, so I've got a 
few ideas that build on Niels' that might be of use to you. =)

Niels Möller wrote:

> Alexander Ward KULUNGOWSKI <akulungo@cs.ucsd.edu> writes:
>  
>
> Well, I don't really know what's expected of a "research paper", but
> some aspects one could focus on
>
> * Translator mechanism in general, and how it allows users to extend
>   the system.
>
> * Particular translators, like shadow-fs. As I understand it, it's
>   interactino with the rest of the system is not entirely trivial,
>   and it might be a good example for really understanding translators
>   and the dir_lookup mechanism.
>  
That sounds interesting. As I understand it, symlinks require a degree 
of support from the operating system, in that there's a filesystem flag 
that tells you if an inode is a symlink (if I remember what I've read 
correctly), and there's a hard-coded limit that says "If we go through 
more than 20 (or whatever the value is) symlinks, then we assume that 
we've hit a loop in the filesystem."

If you could design a protocol that removes the complications that 
translators introduce, that would be both an academic paper, and a 
benefit to the Hurd project.

> * Security, the auth protocol, and the mechanisms that allow
>   untrusted users to run their own file system servers.
>  
Security has the advantage of being formally-analyzable.

> > What about the efforts to port Hurd to L4?
> >    
>
> That's more fuzzy, or perhaps I should say closer to research. Finding
> the right questions to ask may well be harder than actually answering
> them ;-)
>  
The IBM/L4 guys are working on their own multi-server, called SawMill, 
which is based on decomposing Linux into a multi-server.

It seems to me that since multi-servers are known to be an advanced 
design (from an academic perspective), but little is known about how 
best to implement them, that you could write a decent paper comparing 
the approaches of different multi-server designs:
1. GNU/Hurd
2. SawMill (unfortunately, you can only get papers about this, no source 
or binaries)
3. QNX (you can get some good papers, and free binaries, but no source)

Good luck!

-- Tom Hart