> > That's it so simple, that it's only a covered disk driver.
> No, even a file system supporting only a byte-stream model still provides
> much more service than a disk driver.
Depends - if you think about a disk driver as a RWTS thing,
you might be right, but still, byte stream is no more than
the most abstract minimum.
> > From
> > an OS I want OS services for common jobs and common solutions
> > to avoide the millionth invention of the weel. Take the simple
> > Apple DOS and their relave files - Within Unix not even this
> > simple kind of optimization is available - From a _real_ OS
> > I like to have services like SAM/ISAM fileslibrary management
> > etc. not just stupid, slow and clumbsy byte streams
> Why do you think it's better to put this functionality into an
> already-bloated monolithic kernel, rather than implementing it as
> a shared library in user space?
First, User space is user space, and all system services have to
be outside and protected. second Jep, I like to have it in system
space, as lodable services. and third, just look around among unix
systems - portable software means anything included or done from
the bottom. Reinventing the wheel. At this point, Unixes are more
worse than any stupid Windoze where every stupid applet of 1 K
coding installs 300 M of libraries...
> By keeping as much stuff as possible out of the kernel, you will
> generally end up with a more robust system.
You think so ? I you have to have one and the same coding
hundreds of times in the same system ? I call this blooded
overhead.
> > - If you
> > do it to simple, you miss the chance of geting high level
> > optimizations.
> I'm at somewhat of a loss to understand what optimizations
> the kernel could do better than a shared library with regard to
> fixed-length record access (as done by Apple DOS).
As seen above, we have to get a comon wording for kernel,
system or OS. Of course, there is _basicly_ no difference
between a shared library and a kernel service if you just
look at the coding. Its like with byte streams - basicly
its the same .... but just tell me how a kernel can do
buffer priorization as needed for a database system to
get maximum performance without knowing the meaning of
the buffers ? I inply here that you agree that priorizaton
of buffers is a function that has to be done at OS level
to improve over all performance. Just one example.
Or to get a more practical issue: Why can a programm
running in a classic /370 mainframe OS give a better
performance (in numbers of served requests and used
resources), than the same (source) programm running
on the same /370ish machine in an unix environment ?
Maybe bloaded OS vs. bloaded application is an advantage ?
Gruss
Hans
--
Ich denke, also bin ich, also gut
HRK
Received on Mon Dec 21 1998 - 13:28:22 GMT