> Gerard Roudier says:
> >
> > I have had a look into UDI books. This looks to me as the result on N
> > years of brain masturbation from 'has been' people that enjoy turning
> > simple things into complex ones since this makes feel them intelligent.
> > Let me doubt that a driver blindly written using these specs will work
> > without change on any architecture and be optimal enough not to be
> > ridiculous. All the UDI kernel interface reinvention will make debugging
> > a pain. What about the overhead in code lines and CPU cycles due UDI
> > stuff ?
>
> The reason it's so complex is because all the *nix implementations are so
> different. It's quite an undertaking to attempt to create a standard interface
> among such varying implmentations.
A standard must allow differentiations and UDI is just trying to do the
opposite at driver level, as does I2O. So UDI is a _bad_ driver standard.
We know that:
- The unified form of life is the dynosaur.
- The unified car model is the trabant.
Do you want us to become dynosaurs driving trabants?
> UDI prototype environments (without kernel modifications, of course) already
> exist for Solaris and HP-UX. I don't know about performance numbers but as
> I've said before, our own UDI-ish driver interface which is designed along
> somewhat the same lines as UDI has shown no real noticable performance
> difference for our HP-UX implementation (I'm not sure of Solaris performance
> numbers). I also have a Linux implementation which has also shown very
> similar performance to a native Linux driver I also wrote for one of our FDDI
> devices.
Sun has developped a navigator under Java to promote their thing. This
does not mean that any Java applications will have the same
characteristics. I donnot doubt that high skilled and paied programmers
are able to write good UDI drivers. Pay me a lot and I will try to write
some UDI drivers. ;-)
On the other hand, UDI model may fit very well some O/S design but may
need bunches of bad glue code for some other O/S. A standard must be
_fair_ or really try to be so. My reading of UDI models let me think it is
not. BTW, the fact that SCO, Solaris and HP-UX already have UDI
environments tells me suspicious things about the fairness of UDI specs.
> > I am not against UDI. In fact I don't care of such proposal. I want an
> > O/S that fits my needs, and for now I have one that did'nt need UDI in
> > order to be so.
>
> UDI isn't geared toward users. It's geared toward driver developers. Once
> an OS environment is in place, you don't need to worry about it. You write
> a new driver to drive some piece of hardware with the knowledge that all of
> the OS dependencies have already been taken care of. It's a tool for more
> rapid, standardized driver development. Whether the drivers your OS uses
Look at applications developped under some rapid development tools. They
are generally not maintainable and so have to be rewritten each time you
want to add some feature.
If it was possible to develop rapidly software, M$ would'nt have hired
10,000 people for their R&D department.
> are written to UDI or not has no bearing on your use of that OS.
OS differences is a good thing. That's the resulted differentiation that
makes Linux better due to BSD projects and vice-versa.
Everybody doing exactly the _same_ things leads to everybody doing the
_wrong_ thing.
Regards,
Gerard.
PS: Let me know how many UDI drivers are currently shipped with commercial
O/Ses and at which URL(s) I can grab the source code.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/